Laser repair apparatus for PCB and RDL circuit

Abstract

A PCB and RDL circuit laser repair device is disclosed, which can selectively and precisely remove only the metal in the area where the circuit pattern defect is formed while minimizing thermal damage to the substrate portion of the circuit board by removing circuit pattern defects using a non-thermal ablation method with a pulse laser light of a femtosecond (fs) or picosecond (ps) pulse width. A PCB and RDL circuit laser repair device is a PCB and RDL circuit laser repair device for removing circuit pattern defects of a circuit board by laser, comprising: a laser irradiation unit that generates processing laser light for removing circuit pattern defects of a circuit board and irradiates it toward a circuit board; an optical focusing unit that focuses the processing laser light emitted from the laser irradiation unit onto the surface of the circuit board and receives reflected light and emitted light from the circuit board; a vision unit that acquires an image of the circuit board via the optical focusing unit; an illumination unit that selectively irradiates fluorescent illumination and reflected illumination onto the circuit board and acquires fluorescent images and reflected images of the circuit board through the vision unit; a stage unit that supports the circuit board, moves the circuit board within a horizontal plane, and adjusts the position of the circuit board in the direction of the optical axis of the optical focusing unit; and a laser irradiation unit connected to the optical focusing unit, the vision unit, the illumination unit, and the stage unit, respectively, and identifies circuit pattern defects based on fluorescent images and reflected images acquired by the vision unit, and controls the laser irradiation unit and the stage unit to remove circuit pattern defects. It includes a control unit that performs a processing sequence.

Description

Technology Field

[0001] The present invention relates to a PCB and RDL circuit laser repair device, and more specifically, to a device that restores a circuit board to a normal circuit pattern by removing defects in a circuit pattern formed on a circuit board using a laser. In particular, the present invention relates to a PCB and RDL circuit laser repair device applicable to the fields of High Bandwidth Memory (HBM), AI semiconductors, and next-generation semiconductor packaging industries, which combines a non-thermal ablation method using a femtosecond (fs) or picosecond (ps) pulse laser with a high-speed precision scanning method based on Acousto-Optic Deflector (AOD), and combines time-division acquisition of fluorescence / reflection images using a single camera with AI-based defect detection and adaptive processing parameter control. Background Technology

[0002] With the recent rapid development of advanced industries such as artificial intelligence (AI), data centers, cloud computing, autonomous driving, and 5G / 6G communication, the miniaturization and high integration of semiconductor packaging technology for high-performance computing are accelerating. In particular, for High Bandwidth Memory (HBM), interposers for AI semiconductors, and high-bandwidth packages for next-generation data centers, the line-space (L / S) size of circuit patterns is being miniaturized to the 2㎛ level or less, and various types of circuit boards, such as Flip Chip Ball Grid Array (FCBGA) boards, glass interposers, and organic interposers, are being used as substrates on which circuit patterns are formed, in addition to conventional general PCB boards.

[0003] In the manufacturing process of such miniaturized and highly integrated circuit boards, there is a possibility that defects in circuit patterns may occur during multi-stage processes such as etching, plating, and exposure. Circuit pattern defects can be broadly classified into the following three types. First, short defects refer to defects where two adjacent metal patterns are unintentionally connected and short-circuited, and can occur in areas where metal is not sufficiently removed during the etching process. Second, protrusion defects refer to defects where a part of a metal pattern deviates from the intended circuit shape and protrudes outward. Third, residue defects refer to defects formed by metal or foreign substances that should have been removed during the etching process remaining between circuit patterns. Since these circuit pattern defects can cause critical failures in the final product, such as electrical short circuits, signal transmission failures, or reduced reliability, the detection and removal (repair) of circuit pattern defects during the mass production process is essential to ensure a good yield rate for circuit boards.

[0004] Conventional circuit pattern repair technology has utilized a pulse laser with a nanosecond (ns) pulse width to remove metal from defective areas via an ablation method. However, conventional nanosecond pulse lasers have a relatively long pulse width, which causes thermal diffusion from the processing area to adjacent areas while the pulse is applied. This results in thermal damage to the substrate of the circuit board (e.g., organic materials such as resin, ABF (Ajinomoto Build-up Film), and polyimide) or the formation of a Heat Affected Zone (HAZ). Such thermal damage causes a decrease in the mechanical strength or insulation reliability of the circuit board, and in particular, in circuit patterns where the line-space is fined to 2㎛ or less, the heat-affected zone may extend to adjacent normal patterns, potentially causing additional defects.

[0005] In addition, conventional circuit pattern repair devices adopt a method of using a galvano scanner or a Fast Steering Mirror (FSM) alone to deflect the processing laser light. While galvano scanners have the advantage of enabling wide-field scanning, their scanning speed typically remains at around 500 Hz, which presents a limitation in that it is difficult to satisfy the mass production throughput required in the advanced semiconductor packaging industry, such as HBM and AI semiconductors. Furthermore, there is a problem in that positional accuracy is degraded during high-speed scanning due to the inertia caused by the mirror rotation of the galvano scanner, making it difficult to perform precision processing on fine circuit pattern defects of 2 µm or less.

[0006] Furthermore, in conventional circuit pattern repair devices, a dual-camera system has been adopted as a vision system for detecting circuit pattern defects, equipped with separate cameras for acquiring fluorescent images and reflection images. While the dual-camera system has the advantage of being able to acquire two types of images simultaneously, separate calibration is required for image registration because the optical axes of the two cameras differ, and there are limitations to the registration precision due to optical deviations between the cameras. Additionally, the dual-camera system has the disadvantage of increasing the complexity of the optical system structure and raising the equipment volume and manufacturing costs due to the addition of two cameras and their corresponding optical systems.

[0007] Furthermore, in conventional circuit pattern repair devices, processing parameters (e.g., pulse width, pulse repetition frequency, pulse energy, scan pattern) applied during the processing of circuit pattern defects are determined by a pre-fixed recipe, making adaptive control based on the type and size of the defect or the material and thickness of the circuit board difficult. Consequently, there is a problem in that some defects are subjected to excessive processing, potentially causing damage to the substrate of the circuit board, while others are subjected to insufficient processing, resulting in defects that cannot be completely removed. Additionally, conventional circuit pattern repair devices have limitations in that they do not possess a closed-loop control function to automatically evaluate the presence of remaining defects after processing and adaptively reprocess them, or such function is very limited, making it difficult to guarantee the complete removal of defects.

[0008] In particular, the aforementioned limitations of conventional circuit pattern repair devices are further highlighted in glass interposers, organic interposers, or fine RDL circuit boards used for HBM, AI semiconductors, and semiconductor packaging for next-generation data centers. Glass interposers are more vulnerable to thermal damage caused by conventional nanosecond pulse lasers because the glass substrate portion is prone to cracking, while the ABF or polyimide substrates of organic interposers are sensitive to thermal deformation or carbonization. Furthermore, in AI semiconductor interposers, line spacing has been miniaturized to the 2㎛ level or less, resulting in insufficient processing precision with a conventional galvanometer scanner alone; additionally, with the increasing demand for HBM mass production, conventional galvanometer scan speeds are insufficient to meet the throughput requirements of mass production lines.

[0009] Therefore, there is a need to develop an advanced circuit pattern repair device that overcomes the aforementioned limitations of conventional circuit pattern repair devices, precisely removes fine circuit pattern defects without thermal damage to the substrate portion of the circuit board, secures significantly improved mass production throughput compared to conventional devices, adaptively determines processing parameters according to the type, size, and location of the defect, and automatically evaluates whether defects remain after processing to ensure complete removal. Prior art literature

[0010] KR 10-2001-0001728 The problem to be solved

[0011] The present invention provides a PCB and RDL circuit laser repair device capable of selectively and precisely removing only the metal in the area where the circuit pattern defect is formed while minimizing thermal damage to the substrate portion of the circuit board by removing circuit pattern defects using a non-thermal ablation method with a pulse laser light of a femtosecond (fs) or picosecond (ps) pulse width.

[0012] In addition, the present invention provides a PCB and RDL circuit laser repair device capable of simultaneously securing high-speed scanning speeds of approximately 10 to 20 times or more and fine processing precision compared to a conventional galvanometer scanner-only configuration by adopting a configuration in which an Acousto-Optic Deflector (AOD) and a galvanometer scanner cooperate in series to deflect processing laser light, thereby combining the wide field of view characteristics of the galvanometer scanner and the high-speed and high-precision characteristics of the AOD.

[0013] In addition, the present invention provides a PCB and RDL circuit laser repair device that can simplify the structure of the optical system compared to a conventional dual camera method, improve the alignment precision between two images, and reduce the volume and manufacturing cost of the equipment by adopting a configuration in which a camera unit equipped with a single image sensor alternately acquires fluorescent images and reflected images in a time-division manner synchronized with the time-division switching of an illumination unit. means of solving the problem

[0014] A PCB and RDL circuit laser repair device according to an embodiment of the present invention for solving the above problem is a PCB and RDL circuit laser repair device for removing circuit pattern defects of a circuit board by laser, comprising: a laser irradiation unit that generates processing laser light for removing circuit pattern defects of the circuit board and irradiates it toward the circuit board; an optical focusing unit that focuses the processing laser light emitted from the laser irradiation unit onto the surface of the circuit board and receives reflected light and light emission from the circuit board; a vision unit that acquires an image of the circuit board via the optical focusing unit; an illumination unit that selectively irradiates fluorescent illumination and reflected illumination onto the circuit board and acquires a fluorescent image and a reflected image of the circuit board through the vision unit; and a stage unit that supports the circuit board, moves the circuit board within a horizontal plane, and adjusts the position of the circuit board in the direction of the optical axis of the optical focusing unit. and a control unit connected to the laser irradiation unit, the optical focusing unit, the vision unit, the illumination unit, and the stage unit, respectively, and identifying the circuit pattern defect based on the fluorescence image and the reflection image acquired from the vision unit, and controlling the laser irradiation unit and the stage unit to perform a processing sequence to remove the circuit pattern defect;

[0015] The laser irradiation unit comprises: a processing laser unit that generates pulsed laser light with a femtosecond (fs) or picosecond (ps) pulse width as the processing laser light; a beam expander unit that expands the beam diameter of the processing laser light generated by the processing laser unit; an AOD unit that receives the processing laser light emitted from the beam expander unit and changes the deflection direction of the processing laser light by an acousto-optical effect; and a DM unit having wavelength selectivity that reflects the processing laser light emitted from the AOD unit and transmits light in the visible light region; wherein the AOD unit comprises an X-axis AOD member that deflects the processing laser light in the X-axis direction; and a Y-axis AOD member that deflects the processing laser light in the Y-axis direction; wherein the processing laser light is configured to be deflected in a two-dimensional direction, and the optical focusing unit comprises: an objective lens unit disposed on the optical path between the DM unit and the stage unit to focus the processing laser light onto the surface of the circuit board and receive the reflected light and the light emission from the circuit board; a tube lens unit that forms an image of the reflected light and the light emission received from the objective lens unit onto an image sensor of the camera unit of the vision unit; and a BS unit disposed on the optical path between the tube lens unit and the objective lens unit to branch or combine the optical path of the processing laser light and the optical path of the light incident on the vision unit; and the vision unit comprises: a camera unit equipped with a single image sensor to acquire both the fluorescent image and the reflected image; and a fluorescent blocking filter unit coupled to the incident end of the camera unit, which blocks the output wavelength of the fluorescent illumination of the illumination unit but passes the wavelength of the fluorescent emission from the circuit board and the wavelength of the reflected illumination. and an image processing unit that identifies the location and shape of the circuit pattern defect by combining or comparing the fluorescent image and the reflection image acquired from the camera unit;The lighting unit comprises: a fluorescent lighting unit that induces fluorescent emission on a substrate portion of the circuit board by irradiating light of an ultraviolet or near-ultraviolet wavelength onto the circuit board; a coaxial lighting unit that irradiates light in the visible light region onto the circuit board in a direction coaxial with the optical axis of the optical focusing unit; and a ring lighting unit that irradiates light in the visible light region onto the circuit board from a lateral direction. The lighting controller is electrically connected to each of the fluorescent lighting unit, the coaxial lighting unit, and the ring lighting unit, performs time-division switching control so that at least one of the fluorescent lighting unit, the coaxial lighting unit, and the ring lighting unit is mutually exclusively turned ON / OFF, and outputs a trigger signal synchronized with the exposure timing of the camera unit; wherein the lighting controller is configured such that the camera unit acquires the fluorescent image during the period when the fluorescent lighting unit is ON, and the camera unit acquires the reflection image during the period when the coaxial lighting unit or the ring lighting unit is ON, so that the single image sensor alternately acquires the fluorescent image and the reflection image in a time-division manner; and the control unit causes the vision unit to acquire the fluorescent image and the reflection image, and then performs processing by the laser irradiation unit based on the location and shape of the circuit pattern defect identified by the image processing unit, and after the processing is performed, acquires the fluorescent image and the reflection image again to determine the residual of the circuit pattern defect It includes: a processing sequence control unit that determines whether there is a defect in the circuit pattern and performs a closed-loop processing sequence of shooting-processing-shooting-reprocessing to repeat the processing if it is determined that the defect remains; and a synchronization control unit that aligns the operation triggers of the laser irradiation unit, the vision unit, and the lighting unit, respectively, on the time axis.

[0016] The processing laser section of the laser irradiation unit comprises: a laser light source member that emits pulsed laser light with a femtosecond (fs) or picosecond (ps) pulse width; and a pulse control member connected to the laser light source member and variablely controlling the pulse width, pulse repetition frequency, and pulse energy of the pulsed laser light; the beam expander section comprises: a first lens member that expands the diameter of the processing laser light incident from the processing laser section; and a second lens member that shapes the processing laser light emitted from the first lens member into parallel light; and the AOD section comprises an RF driving member that controls the acoustic frequency of the X-axis AOD member and the Y-axis AOD member by applying a variable frequency RF signal to the X-axis AOD member and the Y-axis AOD member, respectively. and further comprising a first-order diffraction selection member that blocks zero-order diffracted light among the light emitted from the X-axis AOD member and the Y-axis AOD member and passes only first-order diffracted light; and the laser irradiation unit further comprises a galvano-auxiliary member disposed on the optical path between the AOD part and the DM part to further change the deflection direction of the processing laser light emitted from the AOD part; and the galvano-auxiliary member comprises an X-galvano-mirror member that deflects the processing laser light in the X-axis direction; and a Y-galvano-mirror member that deflects the processing laser light in the Y-axis direction; and a galvanometer driving member that controls the rotation angle of the X galvanometer mirror member and the Y galvanometer mirror member; wherein the galvanometer auxiliary member is responsible for scanning a wide field of view area on the circuit board, and the AOD member is responsible for fine precision scanning within the wide field of view area, and the galvanometer auxiliary member and the AOD member are configured to cooperate in series to deflect the processing laser light, and the DM member comprises: a dichroic mirror member having wavelength selectivity; and a mirror holder member that finely adjusts the angle with respect to the optical axis of the dichroic mirror member.The optical focusing unit comprises an objective lens portion including an objective lens member that focuses the processing laser light onto the circuit board, the tube lens portion including a tube lens member that forms the reflected light and the light emission onto the image sensor of the camera portion of the vision unit, the BS portion including a beam splitter member that branches or combines the optical path of the processing laser light and the optical path of the light incident on the vision unit, and the camera portion of the vision unit includes a CCD or CMOS type image sensor member; a global shutter control member that performs simultaneous exposure and simultaneous reading for all pixels of the image sensor member; and an exposure time control member that variably adjusts the exposure time of the image sensor member; and the fluorescence blocking filter portion includes a notch filter member that blocks the output wavelength band of the fluorescence illumination portion. and a filter holder member for mounting the notch filter member at the incident end of the camera unit; wherein the image processing unit comprises: an image matching member that aligns pixels of the fluorescent image and the reflection image; a difference image generation member that calculates the difference between the fluorescent image and the reflection image aligned by the image matching member to generate a difference image; a rule-based defect region extraction member that applies a rule-based image processing algorithm to the difference image generated by the difference image generation member to extract a region where the circuit pattern defect exists; and a rule-based defect classification member that classifies the circuit pattern defect extracted by the rule-based defect region extraction member into one of a short defect, a protrusion defect, and a residue defect; and wherein the fluorescent lighting unit of the lighting unit comprises an NUV light source member that generates light of an ultraviolet or near-ultraviolet wavelength; and a polarizer member (p) that shapes light emitted from the NUV light source member into p-polarization; wherein the coaxial illumination member comprises: an RB light source member that generates red and blue visible light; and a polarizer member (s) that shapes light emitted from the RB light source member into s-polarization;and a beam alignment member that directs light emitted from the RB light source member to the surface of the circuit board in a coaxial direction through the BS member; wherein the ring illumination unit comprises: a ring LED array member arranged annularly on the outer side of the objective lens member of the optical focusing unit to irradiate visible light onto the circuit board from a lateral direction; and an angle adjustment member that variably adjusts the irradiation angle of the ring LED array member; and the illumination controller comprises: a switching timing control member that performs time-division switching control so that only one of the fluorescent illumination unit, the coaxial illumination unit, and the ring illumination unit is turned ON at a single point in time; and a synchronization pulse generation member that generates a trigger pulse and outputs it to the camera unit so that the switching operation of the switching timing control member and the exposure operation of the camera unit are synchronized. and a light intensity control member for variably adjusting the light intensity of each of the fluorescent lighting unit, the coaxial lighting unit, and the ring lighting unit; wherein the processing sequence control unit of the control unit includes a closed-loop trigger member that triggers a reprocessing sequence depending on whether the circuit pattern defect remains; a residual defect determination member that determines whether the circuit pattern defect remains based on the fluorescent image and the reflection image acquired after processing; a repetition count upper limit setting member that sets an upper limit value for the number of repetitions of the closed-loop processing sequence; and a sequence termination condition determination member that determines the termination of the closed-loop processing sequence when the circuit pattern defect does not remain or when the number of repetitions reaches the upper limit value; and the synchronization control unit includes a master clock member that generates a reference clock signal; and a trigger distribution member that distributes the reference clock signal generated from the master clock member as a trigger signal to each of the laser irradiation unit, the vision unit, and the lighting unit.

[0017] The control unit further includes an AI vision control unit that takes the fluorescent image and the reflection image acquired from the vision unit as inputs, identifies the location, shape, and type of the circuit pattern defect, and determines whether to perform processing on the identified circuit pattern defect and the result after processing; wherein the AI ​​vision control unit comprises: an image preprocessing member that performs noise removal, contrast correction, and pixel alignment on the fluorescent image and the reflection image to align the fluorescent image and the reflection image to the same coordinate system; a multimodal fusion member that combines the fluorescent image and the reflection image aligned by the image preprocessing member using a channel-wise fusion method or a post-fusion method to generate a single multimodal input data; and a defect pattern recognition member that extracts the location and shape of the area where the circuit pattern defect exists from the multimodal input data generated by the multimodal fusion member using a pre-trained neural network. A neural network defect classification member that classifies the circuit pattern defect extracted from the defect pattern recognition member into one of short defects, protrusion defects, and residue defects by a pre-trained neural network; a segmentation member that divides the boundaries of the circuit pattern defect extracted from the defect pattern recognition member into pixel units and sets them as a processing area (ROI: Region Of Interest); a defect importance evaluation member that calculates a risk score by quantifying the degree of influence of the circuit pattern defect on the circuit function of the circuit board based on the type of the circuit pattern defect classified by the neural network defect classification member and the size and location of the processing area set by the segmentation member; and a processing priority setting member that determines the processing order of the plurality of circuit pattern defects based on the risk score calculated by the defect importance evaluation member when a plurality of circuit pattern defects are identified on the circuit board.A processing difficulty prediction member that predicts the processing difficulty in advance based on the type and shape of the circuit pattern defect and the material information of the circuit board; a processing completion determination member that determines whether the circuit pattern defect remains in the processing area by comparing and analyzing the fluorescent image and the reflection image before processing with the fluorescent image and the reflection image after processing; a reprocessing sequence member that, if the processing completion determination member determines that the circuit pattern defect remains, automatically triggers a reprocessing sequence for the remaining circuit pattern defect and transmits a reprocessing command to the closed loop rigger member of the processing sequence control unit; and a processing trace analysis member that detects damage or residues occurring in the substrate portion of the circuit board by analyzing the surface condition around the processing area from the fluorescent image and the reflection image after processing. A model selection member that receives input information regarding the type of the circuit board, the material of the circuit board, and the thickness of the metal pattern formed on the circuit board, selects a model corresponding to the input information from among a plurality of pre-trained models stored in advance, and applies it to the defect pattern recognition member and the neural network defect classification member; a transfer learning member that fine-tunes the parameters of the pre-trained model selected by the model selection member to match the process conditions of the circuit board currently being worked on; an online learning member that accumulates the fluorescent image, the reflection image, and the judgment result of the processing completion judgment member collected during the processing sequence as training data, and updates the pre-trained model in real time; and a comprehensive diagnostic member that, after the completion of the processing sequence, synthesizes the distribution of the plurality of circuit pattern defects identified on the circuit board, the judgment result of the processing completion judgment member, and the analysis result of the processing trace analysis member to generate a diagnostic report for the circuit board.The defect pattern recognition member comprises: a feature extractor that extracts a feature map from the multimodal input data by a pre-trained convolutional neural network including a plurality of convolutional layers; a candidate region detector that detects candidate regions of the circuit pattern defect from the feature map extracted by the feature extractor; and a region determination member that determines the existence of the circuit pattern defect for each of the candidate regions detected by the candidate region detector and outputs the location coordinates of the circuit pattern defect; and the segmentation member comprises a pixel labeling member that labels whether each pixel of the multimodal input data corresponds to the circuit pattern defect using a pre-trained semantic segmentation neural network of an encoder-decoder structure. and a boundary calculation unit that calculates the boundary coordinates of the processing area based on the results labeled by the pixel labeling unit; wherein the processing completion determination unit comprises: a difference image calculation unit that generates a difference image by calculating the difference between the fluorescent image and the reflection image before processing and the fluorescent image and the reflection image after processing in pixel units; a threshold determination unit that determines whether there exists an area where the intensity of the difference image generated by the difference image calculation unit exceeds a preset threshold value; and a determination result output unit that outputs a final determination result regarding whether the processing is completed based on the determination result of the threshold determination unit.

[0018] The control unit further includes an AI parameter control unit that adaptively determines processing parameters of the laser irradiation unit and the AOD unit based on the location, shape, and type of the circuit pattern defect identified by the AI ​​vision control unit, and dynamically readjusts the processing parameters according to the result after processing; wherein the AI ​​parameter control unit includes a parameter prediction member that takes as input information on the type, shape, and location of the circuit pattern defect, the material of the circuit board, and the thickness information of the metal pattern formed on the circuit board, and outputs processing parameters of the laser irradiation unit and scan parameters of the AOD unit by a pre-trained deep neural network model; a laser parameter control member that variably controls the pulse width, pulse repetition frequency, pulse energy, and number of pulses of the processing laser unit of the laser irradiation unit based on the result output from the parameter prediction member; and an AOD scan pattern control member that variably controls the scan trajectory, scan speed, scan area, and scan pattern of the AOD unit based on the result output from the parameter prediction member. A beam shape control member that controls the beam diameter and beam profile of the processing laser light by variably controlling the beam magnification ratio of the beam expander unit based on the result output from the parameter prediction member; a pulse timing control member that controls the processing laser light to be accurately irradiated at an intended location by synchronizing the pulse emission timing of the processing laser unit with the acoustic frequency change timing of the AOD unit; a processing path optimization member that receives the location coordinates of a plurality of circuit pattern defects identified on the circuit board and calculates a processing sequence and a processing path such that the total travel distance of the processing path between the plurality of circuit pattern defects is minimized; and a reinforcement learning member that learns the correlation between the processing parameters applied during the processing of the circuit pattern defects and the results after processing based on reinforcement learning, and updates the policy of the processing parameters to be applied in the next processing sequence.A Bayesian optimization member that searches for the optimal value of the processing parameter with a small number of trials using a Bayesian optimization algorithm when there is insufficient collected data for a new type of circuit board; a model predictive control member that determines the processing parameter to be applied in the next iteration step of the closed-loop processing sequence based on a model predictive control algorithm; a hybrid control member that secures control stability of the processing parameter by generating a final control value through weighted combination of the output of the parameter prediction member and the output of the PID (Proportional-Integral-Derivative) control algorithm; an adaptive control member that receives residual defect information as feedback from the processing completion judgment member of the AI ​​vision control unit as a result of the previous processing sequence and adaptively corrects the processing parameter to be applied in the next processing sequence based on the feedback information; and a safety control member that dynamically limits the upper limit of the processing parameter so that the processing parameter of the processing laser unit does not exceed a threshold value that causes damage to the substrate portion of the circuit board. A multi-objective optimization member that searches for a Pareto optimum for a multi-objective function in which the speed of the processing sequence, processing quality, and the degree of damage to the substrate portion of the circuit board are used as evaluation indicators; an energy distribution optimization member that learns the energy distribution formed on the circuit board for each scan pattern of the AOD unit and provides a correction value to the AOD scan pattern control member so that a uniform energy distribution is formed within the processing area; an environment correction member that receives the temperature and humidity of the working environment, calculates an environment correction value for the processing parameters, and applies it to the laser parameter control member and the AOD scan pattern control member; and an output correction member that measures the change in the output light quantity of the processing laser unit over time, calculates an output correction value to compensate for the measured change in light quantity, and applies it to the laser parameter control member.A digital twin component that generates a virtual model of the circuit board and the circuit pattern defect, and performs a processing simulation on the virtual model to pre-verify the processing parameters before actual processing; a parameter database component that stores optimal parameter information corresponding to a combination of the type of circuit board, defect type, and processing parameters in advance, and outputs the optimal parameter information upon a request from the parameter prediction component; a batch learning component that analyzes processing result data collected in batch units consisting of a plurality of circuit boards to derive an improved profile of the processing parameters to be applied to the next batch; and a quality-based learning component that updates the policy of the reinforcement learning component and the weights of the parameter prediction component using the processing quality evaluation result measured after the completion of the processing sequence as a compensation function. A federation learning member that operates in a distributed learning manner, wherein the raw data of each device is preserved within each device and only update information of model parameters is integrated, by integrating training data collected from each of the plurality of PCB and RDL circuit laser repair devices to update the pre-trained deep neural network model; wherein the parameter prediction member comprises: an input converter that converts information on the circuit pattern defect and information on the circuit board into an input vector; a prediction output unit that inputs the input vector into the pre-trained deep neural network model and outputs predicted values ​​of the processing parameters and the scan parameters; and a constraint verification unit that determines the final parameter values ​​by verifying whether the predicted values ​​output from the prediction output unit are within a physically feasible range; and wherein the reinforcement learning member comprises: a state observer that generates a state vector by observing the current state of the processing sequence; an action selector that selects the processing parameters to be applied as an action based on the state vector; and a policy updater that updates a policy using the degree of processing quality improvement measured after the execution of the action and the degree of damage to the substrate portion of the circuit board as a compensation function.The adaptive control member comprises: a feedback receiver that receives residual defect information from the processing completion judgment member; an adjustment amount calculation member that calculates an adjustment amount of the processing parameter by comparing and analyzing the residual defect information and the processing parameter applied to the previous processing sequence; and a parameter update member that applies the adjustment amount calculated by the adjustment amount calculation member to the laser parameter control member and the AOD scan pattern control member to update the processing parameter of the next processing sequence; and the laser irradiation unit comprises: a power monitor member that branches off a portion of the processing laser light generated by the processing laser member to measure the output light amount in real time and outputs the measured light amount information to the control unit; and a laser cooling member coupled to the processing laser member that maintains the temperature of the processing laser member within a certain range by absorbing heat generated during the oscillation of the processing laser member and releasing it to the outside. AOD calibration member that periodically measures the correspondence between the acoustic frequency of the AOD unit and the deflection angle of the processing laser light to calibrate the frequency of the RF signal applied to the RF driving member; and a shutter unit that blocks the optical path of the processing laser light while the processing sequence is not performed to prevent unnecessary light irradiation to the circuit board; wherein the shutter unit includes a beam blocking member that is movably arranged on the optical path of the processing laser light to selectively block the processing laser light; and a shutter driving member that drives the beam blocking member to block or open the optical path; and wherein the objective lens unit of the optical focusing unit has a plurality of objective lenses having different magnifications and automatically switches to selectively place one of the objective lenses on the optical path of the optical focusing unit based on the risk score calculated by the defect importance evaluation member or a preset precision requirement level according to the defect type.The system further comprises: an alignment mark recognition unit that captures and recognizes an alignment mark pre-formed on the circuit board and aligns the design coordinate system of the circuit board with the actual coordinate system of the stage unit based on the position information of the recognized alignment mark; an autofocus unit that measures the distance between the surface of the circuit board and the objective lens unit of the optical focusing unit in real time and automatically drives the stage unit based on the measured distance information to align the focus of the processing laser light with the surface of the circuit board; and a substrate thickness measurement unit that measures the thickness of the circuit board and outputs it to the control unit, and causes the control unit to correct the processing parameters of the laser irradiation unit based on the measured thickness information; wherein the alignment mark recognition unit includes a mark capturing member that captures the alignment mark on the circuit board; and a coordinate matching member that extracts the coordinates of the alignment mark from an image captured by the mark capturing member and aligns the extracted coordinates with the coordinates on the design data of the circuit board to calculate a coordinate transformation matrix; wherein the autofocus unit includes a distance measuring member that measures the distance to the surface of the circuit board; and an optical axis direction correction member that converts the distance information measured by the distance measuring member into an optical axis direction driving value of the stage unit and outputs it to the stage unit; and the control unit includes an environment monitoring unit that monitors at least one of the temperature, humidity, and cleanliness of the environment in which the PCB and RDL circuit laser repair device is installed in real time and outputs the monitored environment information to the environment correction member; and a process history storage unit that stores the execution history of the processing sequence, the processing parameters, the distribution of the circuit pattern defects, and the judgment result of the processing completion judgment member in a database for each circuit board, and provides the stored information as learning data according to the request of the online learning member and the batch learning member.The apparatus further includes: an interlock safety unit that detects the door opening / closing status, emergency stop signal, and worker intrusion of the PCB and RDL circuit laser repair device, and stops the operation of the laser irradiation unit when an abnormal state is detected; a board identification unit that recognizes an identification code attached to the circuit board to obtain identification information for the circuit board and calls a processing recipe corresponding to the circuit board based on the identification information; and a beam alignment correction unit that periodically inspects the alignment status on the optical path of the processing laser light and, if a misalignment of the optical path is detected, corrects the position or angle of an optical component on the optical path to maintain the processing laser light in a position where it is accurately irradiated to an intended location on the circuit board. Effects of the invention

[0019] According to the present invention, by removing circuit pattern defects using a non-thermal ablation method with a pulsed laser light of a femtosecond (fs) or picosecond (ps) pulse width, thermal damage to the substrate portion of the circuit board is minimized compared to the conventional nanosecond pulsed laser method, and the formation of a heat-affected zone (HAZ) is suppressed, so that even fine circuit pattern defects with a line-space of 2 μm or less can be precisely removed without damaging adjacent normal patterns.

[0020] In addition, through the serial cooperative configuration of the AOD and the galvanometer scanner, the galvanometer scanner is responsible for scanning a wide field of view area on the circuit board, and the AOD is responsible for fine precision scanning within the wide field of view area. This allows for the simultaneous achievement of high-speed scanning speeds of approximately 10 to 20 times or more compared to a conventional galvanometer scanner-only configuration, along with fine processing precision. Consequently, it is possible to satisfy the mass production throughput required in the semiconductor packaging industries for HBM, AI semiconductors, and next-generation data centers.

[0021] Furthermore, since the camera unit equipped with a single image sensor is synchronized with time-division switching to alternately acquire fluorescent and reflection images, the registration calibration between the two cameras required in conventional dual-camera systems becomes unnecessary. Additionally, because both images are acquired through the same optical axis and the same image sensor, pixel-level registration precision is improved, and optical deviations or calibration errors between cameras are fundamentally eliminated. Moreover, the structure of the optical system is simplified, resulting in reduced equipment volume and manufacturing costs.

[0022] Furthermore, by providing a fluorescence blocking filter at the incident end of the camera unit that selectively blocks the output wavelength of the fluorescent illumination, the NUV reflected light of the fluorescent illumination is blocked from entering the camera unit, allowing only the fluorescent emission signal generated from the substrate portion of the circuit board to reach the camera unit, thereby maximizing the contrast in the fluorescent image. In addition, by controlling the ON / OFF timing of the fluorescent and reflected illumination using a time-division switching method, mutual interference between the two light sources is fundamentally prevented, thereby simultaneously improving the detection accuracy of both the fluorescent image and the reflected image.

[0023] Furthermore, through adaptive closed-loop control by an AI vision control unit and an AI parameter control unit based on a pre-trained deep neural network model, processing parameters are adaptively determined for each defect according to its type, size, and location; the presence of residual defects is automatically evaluated after processing; and if residual defects are present, reprocessing is performed using adaptively adjusted processing parameters that reflect the defect information. This automatically executes a closed-loop processing sequence of shooting-processing-shooting-reprocessing. As a result, damage to the substrate portion of the circuit board can be minimized while ensuring the complete removal of defects, thereby significantly improving the processing yield rate compared to conventional fixed-recipe methods.

[0024] In addition, through the operation of the model selection member, transfer learning member, and online learning member of the AI ​​vision control unit, a model optimized for each type of substrate is automatically selected and applied to various types of circuit boards, such as general PCB substrates, FCBGA substrates, glass interposers, and organic interposers, thereby ensuring uniform processing quality regardless of the type of circuit board. Accordingly, the present invention enables mass production application to various circuit boards in the field of the advanced semiconductor packaging industry, such as HBM, interposers for AI semiconductors, and high-bandwidth packages for next-generation data centers.

[0025] In addition, through the federation learning component of the AI ​​parameter control unit, processing experience collected from each of the multiple PCB and RDL circuit laser repair devices is shared among all devices in a distributed learning manner that integrates only the update information of model parameters. This has the effect of continuously improving the processing quality of the entire mass production line without leaking sensitive raw data, such as circuit board design information, to the outside. Brief explanation of the drawing

[0026] FIG. 1 is a block diagram illustrating the overall structure of a PCB and RDL circuit laser repair device according to one embodiment of the present invention. FIG. 2 is a diagram illustrating the configuration of an optical system of a PCB and RDL circuit laser repair device according to one embodiment of the present invention. FIG. 3 is a diagram illustrating the optical path in the optical system of a PCB and RDL circuit laser repair device according to one embodiment of the present invention. FIG. 4 is a drawing illustrating the mechanical configuration of a stage unit according to one embodiment of the present invention. FIG. 5 is a schematic diagram illustrating the structure of a laser irradiation unit according to one embodiment of the present invention. FIG. 6 is a schematic diagram illustrating the structure of an optical focusing unit according to one embodiment of the present invention. FIG. 7 is a schematic diagram illustrating the structure of a vision unit according to one embodiment of the present invention. FIG. 8 is a schematic diagram illustrating the structure of a lighting unit according to one embodiment of the present invention. FIG. 9 is a schematic diagram illustrating the structure of a stage unit according to one embodiment of the present invention. FIG. 10 is a schematic diagram illustrating the structure of a control unit according to one embodiment of the present invention. FIG. 11 is a schematic diagram illustrating the structure of an AI vision control unit according to one embodiment of the present invention. FIG. 12 is a schematic diagram illustrating the structure of an AI parameter control unit according to one embodiment of the present invention. Specific details for implementing the invention

[0027] Hereinafter, embodiments are described in detail with reference to the attached drawings. However, various modifications may be made to the embodiments, and thus the scope of the patent application is not limited or restricted by these embodiments. It should be understood that all modifications, equivalents, and substitutions to the embodiments are included within the scope of the rights.

[0028] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Accordingly, the embodiments are not limited to the specific disclosed forms, and the scope of this specification includes modifications, equivalents, or substitutions that fall within the technical concept.

[0029] Terms such as "first" or "second" may be used to describe various components, but these terms should be interpreted solely for the purpose of distinguishing one component from another. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0030] When it is stated that a component is "connected" to another component, it should be understood that it may be directly connected to or coupled with that other component, or that there may be other components in between.

[0031] The terms used in the embodiments are for illustrative purposes only and should not be interpreted as intended to be limiting. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as "comprising" or "having" are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0032] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the embodiments pertain. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0033] In addition, when describing with reference to the attached drawings, identical components are assigned the same reference numeral regardless of drawing symbols, and redundant descriptions thereof are omitted. In describing the embodiments, if it is determined that a detailed description of related prior art could unnecessarily obscure the essence of the embodiments, such detailed description is omitted.

[0034] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims.

[0035] In the embodiments of the present invention, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in the embodiments of the present invention.

[0036] The shapes, sizes, ratios, angles, numbers, etc. disclosed in the drawings for explaining embodiments of the present invention are exemplary, and therefore the present invention is not limited to the depicted details. Furthermore, in describing the present invention, if it is determined that a detailed description of related known technology may unnecessarily obscure the essence of the present invention, such detailed description is omitted. Where terms such as "includes," "has," or "is made up" are used in this specification, other parts may be added unless "only" is used. Where a component is expressed in the singular, it includes cases where it includes the plural unless specifically stated otherwise.

[0037] In interpreting the components, they are interpreted to include a margin of error even in the absence of a separate explicit statement.

[0038] In the case of describing a positional relationship, for example, when the positional relationship between two parts is described using expressions such as 'on,' 'upper,' 'lower,' or 'next to,' one or more other parts may be located between the two parts unless 'immediately' or 'directly' is used.

[0039] When elements or layers are referred to as "on" another element or layer, this includes cases where another layer or element is placed directly on top of or in between. Throughout the specification, the same reference numerals refer to the same components.

[0040] The size and thickness of each component shown in the drawings are illustrated for convenience of explanation, and the present invention is not necessarily limited to the size and thickness of the illustrated components.

[0041] The features of each of the various embodiments of the present invention may be combined or combined with one another, either partially or wholly, and as will be fully understood by those skilled in the art, various technical interlocking and operation are possible, and each embodiment may be implemented independently of one another or together in an interlocking relationship.

[0043] FIG. 1 is a block diagram illustrating the overall structure of a PCB and RDL circuit laser repair device according to an embodiment of the present invention. FIG. 2 is a diagram illustrating the configuration of the optical system of a PCB and RDL circuit laser repair device according to an embodiment of the present invention. FIG. 3 is a diagram illustrating the optical path in the optical system of a PCB and RDL circuit laser repair device according to an embodiment of the present invention. FIG. 4 is a diagram illustrating the mechanical configuration of a stage unit according to an embodiment of the present invention. FIG. 5 is a diagram schematically illustrating the structure of a laser irradiation unit according to an embodiment of the present invention. FIG. 6 is a diagram schematically illustrating the structure of an optical focusing unit according to an embodiment of the present invention. FIG. 7 is a diagram schematically illustrating the structure of a vision unit according to an embodiment of the present invention. FIG. 8 is a diagram schematically illustrating the structure of an illumination unit according to an embodiment of the present invention. FIG. 9 is a diagram schematically illustrating the structure of a stage unit according to an embodiment of the present invention. FIG. 10 is a diagram schematically illustrating the structure of a control unit according to an embodiment of the present invention. FIG. 11 is a schematic diagram illustrating the structure of an AI vision control unit according to one embodiment of the present invention. FIG. 12 is a schematic diagram illustrating the structure of an AI parameter control unit according to one embodiment of the present invention.

[0044] Hereinafter, with reference to FIGS. 1 to 12, the structure of a PCB and RDL circuit laser repair device (1000) according to one embodiment of the present invention will be described.

[0045] A PCB and RDL circuit laser repair device (1000) according to one embodiment of the present invention is a device that restores a circuit board (O) to a normal circuit pattern by removing defects in a circuit pattern formed on a circuit board (O) with a laser.

[0046] In the present invention, "circuit pattern defect" refers to any type of defect in which a metal pattern formed on a circuit board (O) deviates from the designed circuit shape, and specifically includes the following three types. First, a short defect refers to a defect in which two adjacent metal patterns are unintentionally connected and short-circuited, and may occur in an area where the metal is not sufficiently removed during the etching process. Second, a protrusion defect refers to a defect in which a part of the metal pattern deviates from the intended circuit shape and protrudes outward. Third, a residue defect refers to a defect formed by metal or foreign substances that should be removed during the etching process remaining between the circuit patterns. The PCB and RDL circuit laser repair device (1000) of the present invention performs the function of restoring the circuit board (O) to a normal circuit pattern by selectively removing the above three types of defects using processing laser light.

[0047] In addition, the term "circuit board (O)" in the present invention is used as a concept including a PCB (Printed Circuit Board) board and an RDL (Re-Distribution Layer) circuit board for semiconductor packaging, and the types thereof may include a general PCB board, an FCBGA (Flip Chip Ball Grid Array) board, a glass interposer, an organic interposer, etc. The circuit board (O) typically has a circuit pattern formed of a metal material such as copper (Cu) on a substrate portion of an organic or inorganic material such as resin, ABF (Ajinomoto Build-up Film), polyimide, or glass.

[0048] A PCB and RDL circuit laser repair device (1000) includes a laser irradiation unit (1100), an optical focusing unit (1200), a vision unit (1300), an illumination unit (1400), a stage unit (1500), and a control unit (1600).

[0049] A laser irradiation unit (1100) is positioned facing the surface of a circuit board (O) and generates a processing laser beam to remove circuit pattern defects on the circuit board (O) and irradiates it toward the circuit board (O). The processing laser beam generated by the laser irradiation unit (1100) is focused onto the surface of the circuit board (O) via an optical focusing unit (1200), thereby selectively removing the metal in the area where circuit pattern defects are formed on the circuit board (O).

[0050] The optical focusing unit (1200) focuses the processing laser light emitted from the laser irradiation unit (1100) onto the surface of the circuit board (O) and receives reflected light and emitted light from the circuit board (O). The optical focusing unit (1200) integrates the optical path of the processing laser light and the optical path of the light incident on the vision unit (1300) coaxially, thereby enabling processing and image acquisition to be performed on the same optical axis.

[0051] The vision unit (1300) acquires an image of the circuit board (O) via the optical focusing unit (1200). The vision unit (1300) is configured to acquire both a fluorescent image from fluorescent illumination and a reflected image from reflected illumination of the illumination unit (1400) described later, through a camera unit (1310) equipped with a single image sensor. That is, unlike the conventional method of separately providing a camera for acquiring fluorescent images and a camera for acquiring reflected images, the present invention simplifies the structure of the optical system and improves the precision of registration between the two images by having a single camera unit (1310) alternately acquire fluorescent images and reflected images in time division.

[0052] The lighting unit (1400) selectively irradiates fluorescent lighting and reflective lighting onto the circuit board (O) to acquire fluorescent images and reflective images of the circuit board (O) through the vision unit (1300). The lighting unit (1400) includes a fluorescent lighting unit (1410) that induces fluorescent emission on the substrate portion of the circuit board (O) by irradiating light of ultraviolet or near-ultraviolet wavelength, a coaxial lighting unit (1420) that irradiates light in the visible light region in a coaxial direction, a ring lighting unit (1430) that irradiates light in the visible light region from a lateral direction, and a lighting controller (1440) that controls the ON / OFF timing of each of these lighting units in a time-division switching manner.

[0053] The stage unit (1500) supports the circuit board (O), moves the circuit board (O) in the X-axis and Y-axis directions within a horizontal plane, and adjusts the position of the circuit board (O) in the optical axis direction of the optical focusing unit (1200). The stage unit (1500) stably fixes the circuit board (O) while precisely controlling its position so that sequential processing can be performed on multiple measurement points of the circuit board (O).

[0054] The control unit (1600) is connected to the laser irradiation unit (1100), optical focusing unit (1200), vision unit (1300), lighting unit (1400), and stage unit (1500), respectively, identifies circuit pattern defects based on fluorescent images and reflection images acquired from the vision unit (1300), and performs a processing sequence to remove circuit pattern defects by controlling the laser irradiation unit (1100) and the stage unit (1500). The control unit (1600) includes a processing sequence control unit (1610), a synchronization control unit (1620), an AI vision control unit (1640), an AI parameter control unit (1650), an environment monitoring unit (1670), a process history storage unit (1680), an interlock safety unit (1690), a substrate identification unit (1695), and a beam alignment correction unit (1696). By doing so, the PCB and RDL circuit laser repair device (1000) of the present invention can integrally perform automatic detection of defects, adaptive determination of processing parameters, automatic evaluation of results after processing, and closed-loop reprocessing.

[0055] <Flow of light and signal conversion in the present invention>

[0056] In the PCB and RDL circuit laser repair device (1000) of the present invention, the light energy of the processing laser light generated by the laser irradiation unit (1100) is converted into ablation energy for removing metal pattern defects on the circuit board (O), and the reflected light and light emitted from the circuit board (O) are detected again as light signals by the vision unit (1300), and the detected light signals are converted into electrical signals and processed as image data. Through this conversion process, circuit pattern defects on the circuit board (O) are detected and removed.

[0057] Specifically, when pulsed laser light with a femtosecond (fs) or picosecond (ps) pulse width generated from a laser light source member (1110a) is focused on the surface of a circuit board (O), multi-photon absorption and photo-ionization occur in the area where metal pattern defects are formed by the peak power of the pulse, and thereby the metal is removed by non-thermal ablation (cold ablation) (light → ablation conversion). Non-thermal ablation is a method in which processing is completed before thermal diffusion occurs, and has the advantage of being able to selectively remove only metal pattern defects while minimizing thermal damage to the substrate portion of the circuit board (O).

[0058] Additionally, the fluorescent light irradiated from the lighting unit (1400) is absorbed by the substrate portion of the circuit board (O) to induce fluorescent emission in the visible light region (light → fluorescent emission conversion), and the reflected light irradiated from the coaxial lighting unit (1420) or the ring lighting unit (1430) is reflected from the surface of the circuit board (O). The fluorescent emission and reflected light from the circuit board (O) are received through the optical focusing unit (1200), converted into electrical signals by the image sensor member (1310a) of the camera unit (1310) of the vision unit (1300), and output as fluorescent images and reflected images (light → electrical conversion). Thus, the present invention has an integrated structure of optical-electrical conversion in which ablation by processing laser light and image acquisition by fluorescent / reflective lighting are performed through the same optical system.

[0059] The main types of light used in the present invention are as follows. Processing laser light is generated from a laser light source member (1110a) of a processing laser unit (1110), has a femtosecond (fs) or picosecond (ps) pulse width, and is irradiated onto the surface of a circuit board (O) to remove circuit pattern defects in an ablation manner. Fluorescent illumination light is generated from an NUV light source member (1410a) of a fluorescent illumination unit (1410), has an ultraviolet or near-ultraviolet wavelength band (e.g., 350 nm to 410 nm), and is used as an excitation light to induce fluorescent emission on a substrate portion of the circuit board (O). Reflected illumination light is generated from the RB light source member (1420a) of the coaxial illumination unit (1420) or the ring LED array member (1430a) of the ring illumination unit (1430), has a visible light wavelength band such as red (R) and blue (B), and is reflected from the surface of the circuit board (O) to contribute to the formation of a reflected image. Fluorescent emission is light in the visible light region emitted by the substrate part of the circuit board (O) excited by the fluorescent illumination light, and contributes to the formation of a fluorescent image. Reflected light is light reflected from the metal pattern and substrate part of the circuit board (O) by the reflected illumination light, and contributes to the formation of a reflected image.

[0060] Confluence, branching, and turning points of the scenic route

[0061] In the PCB and RDL circuit laser repair device (1000) of the present invention, a plurality of junction points and branching points exist on the optical path, an optical-to-ablation conversion point exists on the surface of the circuit board (O), and an optical-to-electrical conversion point exists in the camera unit (1310).

[0062] As for the junction point of the optical paths, the optical path of the processing laser light from the laser irradiation unit (1100) and the optical path on the vision unit (1300) side are combined at the dichroic mirror member (1140a) of the first DM section (1140), so that the two optical paths are aligned at the same processing point on the circuit board (O). Second, the fluorescent illumination light from the fluorescent illumination section (1410) and the reflected illumination light from the coaxial illumination section (1420) are combined at the beam splitter member (1230a) and polarizing beam splitter member (1230b) of the BS section (1230), so that the two illumination lights are irradiated onto the circuit board (O) through the optical focusing unit (1200). At this time, the polarizing beam splitter member (1230b) is configured to transmit light of the p-polarized component and reflect light of the s-polarized component, and the p-polarized fluorescent illumination light passing through the polarizer member (p) (1410b) in the fluorescent illumination unit (1410) and the s-polarized reflected illumination light passing through the polarizer member (s) (1420b) in the coaxial illumination unit (1420) are combined into the same optical path according to the polarization characteristics.

[0063] As for the branching point of the optical path, the reflected light and fluorescent light emitted from the circuit board (O) are branched from the beam splitter member (1230a) of the BS section (1230) into an optical path directed toward the camera section (1310).

[0064] At the light-to-ablation conversion point, the pulse energy of the processing laser light is absorbed by the metal pattern defects on the surface (upper surface) of the circuit board (O), and the metal is removed by ablation through multiphoton absorption and photoionization.

[0065] At the light-to-electricity conversion point, reflected light and fluorescent light from the circuit board (O) are converted into pixel-unit electrical signals at the image sensor member (1310a) of the camera part (1310) of the vision unit (1300) to form a fluorescent image and a reflected image.

[0066] Additionally, the PCB and RDL circuit laser repair device (1000) of the present invention performs a time-division synchronization operation to acquire both fluorescent images and reflection images with a single camera unit (1310) by separating fluorescent lighting and reflection lighting in the time domain and using them alternately in a time-division manner. Specifically, the lighting controller (1440) controls the fluorescent lighting unit (1410) to be ON and the coaxial lighting unit (1420) and ring lighting unit (1430) to be OFF at any time point t1, and applies an exposure trigger to the camera unit (1310) so that the camera unit (1310) acquires a fluorescent image of the circuit board (O). Next, at time point t2, the lighting controller (1440) switches the fluorescent lighting unit (1410) to the OFF state and switches the coaxial lighting unit (1420) to the ON state, and applies an exposure trigger to the camera unit (1310) again so that the camera unit (1310) acquires a reflected image of the circuit board (O). The interval between time point t1 and time point t2 can be set to a level of several milliseconds to tens of milliseconds corresponding to the frame rate of the camera unit (1310), and through this time-division switching, the fluorescent image and the reflected image can be acquired substantially simultaneously with a single camera unit (1310).

[0067] The reflected image is an image formed by the reflection of visible light from the surface of the circuit board (O). Since both the metal pattern and the substrate portion form reflected light, the contrast between the two regions may be relatively low. In particular, as the circuit pattern becomes finer, it becomes difficult to visually distinguish between the defect portion of a fine defect, such as a short circuit, and the normal pattern, making it difficult to detect fine defects using only the reflected image.

[0068] In contrast, the fluorescent image is formed in such a way that light of ultraviolet or near-ultraviolet wavelength irradiated from the fluorescent illumination unit (1410) is absorbed by the substrate portion of the circuit board (O) to induce fluorescent emission in the visible light region. Since the metal pattern does not cause fluorescent emission, it is expressed darkly in the fluorescent image, whereas the substrate portion is expressed brightly due to the fluorescent emission, thereby forming a stark contrast between the metal pattern and the substrate portion. As a result, even fine defects that are difficult to detect in the reflection image can be clearly detected in the fluorescent image. Accordingly, the present invention enables the comprehensive detection of defects that are difficult to detect with only one image by alternately acquiring the fluorescent image and the reflection image in time division through a single camera unit (1310) and combining or comparing and analyzing the two images in the image processing unit (1330) or the AI ​​vision control unit (1640).

[0069] Additionally, the PCB and RDL circuit laser repair device (1000) of the present invention performs a closed-loop processing sequence by the processing sequence control unit (1610). Specifically, the processing sequence control unit (1610) sequentially performs the following steps on the circuit board (O). First, after loading the circuit board (O), coordinate alignment by the alignment mark recognition unit (1550), focus adjustment by the autofocus unit (1560), and thickness measurement by the substrate thickness measurement unit (1570) are performed. Subsequently, a fluorescent image and a reflected image of the circuit board (O) are acquired by time-division switching control of the lighting controller (1440), and the AI ​​vision control unit (1640) analyzes the two acquired images to identify the location, shape, and type of the circuit pattern defect. When a defect is identified, the AI ​​parameter control unit (1650) determines processing parameters for the laser irradiation unit (1100) and the AOD unit (1130) based on information regarding the identified defect, and the laser irradiation unit (1100) irradiates processing laser light according to the determined processing parameters to perform processing on the defect area. After processing is performed, a fluorescent image and a reflection image are acquired again, and a residual defect determination unit (1610b) determines whether the defect remains. If a residual defect exists and the number of repetitions does not reach the upper limit value set by the repetition count upper limit setting unit (1610c), reprocessing is performed after adaptive readjustment of the processing parameters. If no residual defect exists or the number of repetitions reaches the upper limit value, the comprehensive diagnosis unit (1640o) generates a diagnosis report for the circuit board (O), stores the processing history in the process history storage unit (1680), and then the circuit board (O) is unloaded. By means of this closed-loop processing sequence of shooting-processing-shooting-reprocessing, the present invention can minimize damage to the substrate portion of the circuit board (O) while ensuring the complete removal of defects.

[0070] Below, the detailed configuration of each unit will be described in detail sequentially with reference to the drawings.

[0071] Laser Irradiation Unit

[0072] A laser irradiation unit (1100) is positioned facing the surface of a circuit board (O) and generates a processing laser light to remove circuit pattern defects of the circuit board (O) and irradiates it toward the circuit board (O).

[0073] The laser irradiation unit (1100) includes a processing laser section (1110), a beam expander section (1120), an AOD section (1130), a DM section (1140), a galvano-auxiliary section (1150), and a shutter section (1160).

[0074] The processing laser light generated in the laser irradiation unit (1100) is generated in the processing laser section (1110), then the beam diameter is expanded in the beam expander section (1120), the deflection direction is changed in the AOD section (1130), additional deflection is performed in the galvano-aux section (1150), and it reaches the optical focusing unit (1200) via the DM section (1140). That is, the processing laser light propagates along the optical path in the order of processing laser section (1110) → beam expander section (1120) → AOD section (1130) → galvano-aux section (1150) → DM section (1140) → optical focusing unit (1200) → circuit board (O), and this optical path is called the processing beam incident path.

[0075] Unlike conventional circuit pattern repair devices that use nanosecond (ns) pulse lasers, the laser irradiation unit (1100) uses pulse laser light with a femtosecond (fs) or picosecond (ps) pulse width to perform processing by a non-thermal ablation (cold ablation) method. By doing so, it is possible to selectively and precisely remove only the metal in the area where the circuit pattern defect is formed while minimizing thermal damage to the substrate portion of the circuit board (O).

[0076] In addition, the laser irradiation unit (1100) adopts a configuration in which the AOD unit (1130) and the galvano auxiliary unit (1150) cooperate in series to deflect the processing laser light, unlike the method using a conventional galvano scanner or a standalone FSM (Fast Steering Mirror) scanner. Specifically, the two deflection means operate by sharing roles so that the galvano auxiliary unit (1150) is responsible for scanning a wide field area on the circuit board (O), and the AOD unit (1130) is responsible for fine precision scanning within the wide field area. As a result, a high-speed scanning speed of about 10 to 20 times or more compared to a conventional galvano scanner is secured, while simultaneously achieving fine processing precision.

[0077] The laser irradiation unit (1100) may further include a collimator (not shown) in the optical path between the processing laser unit (1110) and the beam expander unit (1120), or in the optical path between the beam expander unit (1120) and the AOD unit (1130). The collimator converts the divergence of processing laser light emitted from the processing laser unit (1110) into parallel light, thereby ensuring that optical operations in the beam expander unit (1120) and the AOD unit (1130) are performed normally. Additionally, the laser irradiation unit (1100) may further include an optical isolator (not shown) immediately after the processing laser unit (1110). The optical isolator prevents back-reflected light from the optical system from returning to the light source of the processing laser unit (1110), thereby ensuring the stability of the light source. Additionally, the laser irradiation unit (1100) may further include a half-wave plate (not shown) immediately after the processing laser section (1110), and the half-wave plate improves the processing quality of the circuit board (O) by adjusting the polarization direction of the processing laser light. Additionally, the laser irradiation unit (1100) may further include one or more folding mirrors (not shown) on the optical path between the AOD section (1130) and the DM section (1140), and the folding mirrors fold the optical path to make the overall size of the laser irradiation unit (1100) compact.

[0078] Processing Laser Department

[0079] The processing laser unit (1110) serves as a light source for the laser irradiation unit (1100) and generates processing laser light. The processing laser unit (1110) is positioned toward the beam expander unit (1120), and the generated processing laser light passes through the beam expander unit (1120) and the AOD unit (1130) to reach the DM unit (1140), after which it is irradiated onto the surface of the circuit board (O) through the optical focusing unit (1200).

[0080] The processing laser unit (1110) includes a laser light source member (1110a), a pulse control member (1110b), a power monitor member (1110c), and a laser cooling member (1110d).

[0081] The laser light source member (1110a) emits pulsed laser light with a pulse width of femtosecond (fs) or picosecond (ps). When the pulsed laser light generated from the laser light source member (1110a) is irradiated onto the surface of the circuit board (O), multi-photon absorption and photo-ionization occur in the region where the metal pattern defect is formed due to the peak power of the pulse, thereby removing the metal by a non-thermal ablation method. Since the pulse width of the femtosecond or picosecond pulsed laser light is very short, almost no thermal diffusion occurs while the pulse is applied, and accordingly, thermal damage to the substrate portion (organic material such as resin, ABF, etc.) of the circuit board (O) is minimized.

[0082] In one embodiment, the laser light source member (1110a) can emit pulsed laser light having a pulse width selected in the range of 100 femtoseconds (fs) or more and 100 picoseconds (ps) or less. More specifically, the laser light source member (1110a) can have a pulse width in the range of 100 femtoseconds (fs) to 10 picoseconds (ps), and in this pulse width range, processing of less than the thermal diffusion time is possible, so that thermal influence on the substrate portion of the circuit board (O) hardly occurs.

[0083] In one embodiment, the laser light source member (1110a) may be implemented as a fiber laser, a solid-state laser, or a semiconductor laser. Fiber lasers have the advantage of excellent beam quality and easy maintenance, while solid-state lasers (e.g., Yb:YAG, Nd:YAG) have the advantage of being able to secure high pulse energy.

[0084] In one embodiment, the center wavelength of the pulsed laser light oscillating from the laser light source member (1110a) may be set to any one of the near-infrared region in the range of 1030 nm to 1064 nm, the visible light region in the range of 515 nm to 532 nm, or the ultraviolet region in the range of 343 nm to 355 nm. When the center wavelength is in the visible light or ultraviolet region, the light absorption rate of the metal pattern (particularly copper (Cu)) of the circuit board (O) is improved, thereby increasing processing efficiency, and at the same time, the light penetration depth into the substrate portion of the circuit board (O) is reduced, thereby further suppressing damage to the substrate portion.

[0085] The pulse control member (1110b) is connected to the laser light source member (1110a) and variably controls the pulse width, pulse repetition rate, and pulse energy of the pulsed laser light. The pulse control member (1110b) sets the pulse width, pulse repetition rate, and pulse energy of the pulsed laser light to a predetermined value by controlling the driving current, pumping energy, or Q-switching timing of the laser light source member (1110a).

[0086] The pulse repetition frequency directly affects the processing speed, and a high pulse repetition frequency enables processing more measurement points per unit time. In one embodiment, the pulse control member (1110b) can set a pulse repetition frequency in the range of 100 kHz to 10 MHz. The pulse energy affects the processing depth and the size of the processing area, and the optimal pulse energy is set according to the material of the circuit board (O) and the thickness of the metal pattern. In addition, the pulse control member (1110b) controls the pulse energy so that it is below the damage threshold of the substrate part to prevent damage to the substrate part of the circuit board (O). In one embodiment, the pulse control member (1110b) receives a control signal output from the laser parameter control member (1650b) of the AI ​​parameter control unit (1650) described later, and variably controls the pulse parameters in real time according to the type of circuit pattern defect and the material of the circuit board (O).

[0087] The power monitor member (1110c) branches off a portion of the processing laser light generated by the processing laser unit (1110) to measure the output light amount in real time and outputs the measured light amount information to the control unit (1600). The power monitor member (1110c) may be configured to include a beam sampler (not shown) that is positioned on the optical path of the processing laser light and branches off a portion of the processing laser light (typically 1% to 5%) into a measurement path, and a photodiode (not shown) that receives the light branched into the measurement path and converts it into an electrical signal. The light amount information measured by the power monitor member (1110c) is transmitted to the output correction member (1650q) of the AI ​​parameter control unit (1650) of the control unit (1600) and is used to calculate a correction value to compensate for changes in the light amount of the processing laser unit (1110) over time.

[0088] The laser cooling member (1110d) is coupled to the processing laser unit (1110) and absorbs the heat generated during the oscillation of the processing laser unit (1110) and releases it to the outside, thereby maintaining the temperature of the processing laser unit (1110) within a certain range. In one embodiment, the laser cooling member (1110d) may be implemented as a water-cooled chiller or an air-cooled cooling system. Since the output stability of a femtosecond or picosecond pulse laser may be degraded by the heat generated during oscillation, thermal management by the laser cooling member (1110d) is essential for ensuring the long-term stability of the processing laser unit (1110). In one embodiment, the laser cooling member (1110d) performs precise temperature control of 25±0.1°C to maintain the output stability of the processing laser unit (1110) within ±1%.

[0089] Beam Expander Unit

[0090] The beam expander unit (1120) is positioned on the optical path between the processing laser unit (1110) and the AOD unit (1130) and expands the beam diameter of the processing laser light generated from the processing laser unit (1110). More specifically, the beam expander unit (1120) is located at the rear end of the processing laser unit (1110) and the front end of the AOD unit (1130) in the processing beam incidence path, receives the processing laser light emitted from the processing laser unit (1110), expands the beam diameter, and then emits it toward the AOD unit (1130).

[0091] As the beam diameter of the processing laser light is expanded by the beam expander (1120), the diffraction limit at the objective lens (1210) of the optical focusing unit (1200) is reduced. The diffraction limit is inversely proportional to the numerical aperture (NA) of the focusing optical system, and the expansion of the incident beam diameter consequently has the effect of reducing the diameter of the focusing spot. Accordingly, the beam expander (1120) minimizes the size of the processing spot formed on the circuit board (O), thereby enabling precision processing of fine circuit patterns (e.g., LS 2㎛ class).

[0092] The beam expander part (1120) includes a first lens member (1120a) and a second lens member (1120b).

[0093] The first lens member (1120a) expands the diameter of the processing laser light incident from the processing laser unit (1110). In one embodiment, the first lens member (1120a) may be implemented as a diverging lens having a negative focal length, causing the diameter of the processing laser light to gradually expand as it passes through. In another embodiment, the first lens member (1120a) may be implemented as a converging lens having a positive focal length, thereby expanding the beam by focusing the processing laser light to a single point and then diverging it again (Keplerian beam expander).

[0094] The second lens member (1120b) shapes the processing laser light emitted from the first lens member (1120a) into parallel light. In one embodiment, the second lens member (1120b) is implemented as a converging lens having a positive focal length, and converts the processing laser light magnified by the first lens member (1120a) into parallel light and directs it to the AOD unit (1130). The magnification ratio of the beam expander unit (1120) is determined according to the focal length ratio of the first lens member (1120a) and the second lens member (1120b), and in one embodiment, the magnification ratio can be set in the range of 2 to 10 times. In one embodiment, the second lens member (1120b) is provided to be movable in the direction of the optical axis, and its position is adjusted according to a control signal output from the beam shape control member (1650d) of the AI ​​parameter control unit (1650) described later, thereby allowing the beam diameter and beam profile of the processing laser light to be variably adjusted.

[0095] <AOD부>

[0096] The AOD unit (1130) is positioned in the optical path between the beam expander unit (1120) and the DM unit (1140), receives the processing laser light emitted from the beam expander unit (1120), and changes the deflection direction of the processing laser light by an acousto-optic effect. That is, the AOD unit (1130) is located at the rear end of the beam expander unit (1120) and the front end of the DM unit (1140) in the processing beam incidence path, and changes the direction of travel of the processing laser light shaped by the beam expander unit (1120) at high speed so that the processing laser light is irradiated to a desired processing point on the surface of the circuit board (O).

[0097] The AOD unit (1130) provides a high-speed scanning speed that is approximately 10 to 20 times faster than that of a conventional galvanometer scanner alone. While the scanning speed of a galvanometer scanner is typically around 500 Hz, the AOD can achieve a scanning speed of several tens of kHz or more. However, since the AOD has the characteristic of having a narrower scanning angle range compared to a galvanometer scanner, the present invention works in series with the galvanometer auxiliary unit (1150) described later, so that the AOD unit (1130) is responsible for fine precision scanning within a wide field of view and the galvanometer auxiliary unit (1150) is responsible for wide field of view scanning. By doing so, the high-speed scanning characteristics of the AOD and the wide field of view characteristics of the galvanometer scanner are combined, thereby simultaneously achieving three characteristics of wide field of view + high speed + high precision that could not be achieved with a conventional standalone scanner.

[0098] The AOD unit (1130) includes an X-axis AOD member (1130a), a Y-axis AOD member (1130b), an RF driving member (1130c), a first-order diffraction selection member (1130d), and an AOD correction member (1130e).

[0099] The X-axis AOD member (1130a) deflects the processing laser light in the X-axis direction. The X-axis AOD member (1130a) is a member comprising an acousto-optic crystal, wherein when an RF signal is applied to the acousto-optic crystal, an acoustic wave is generated inside the crystal, and the refractive index of the crystal is periodically modulated by this acoustic wave, so that the incident processing laser light is deflected by Bragg diffraction. In one embodiment, the acousto-optic crystal of the X-axis AOD member (1130a) may be implemented as any one of tellurium oxide (TeO2), lead molybdate (PbMoO4), or crystalline germanium.

[0100] The Y-axis AOD member (1130b) deflects the processing laser light in the Y-axis direction. The Y-axis AOD member (1130b) is arranged in series with the X-axis AOD member (1130a), and its crystal axis is rotated 90° with respect to the crystal axis of the X-axis AOD member (1130a), so that the two members each perform deflection in the orthogonal direction. By the serial combination of the X-axis AOD member (1130a) and the Y-axis AOD member (1130b), a two-dimensional (2D) AOD configuration is achieved in which the processing laser light can be deflected simultaneously in the X-axis and Y-axis directions.

[0101] The RF driving member (1130c) applies a variable frequency RF signal to the X-axis AOD member (1130a) and the Y-axis AOD member (1130b), respectively, to control the acoustic frequency of the X-axis AOD member (1130a) and the Y-axis AOD member (1130b). When the frequency of the RF signal is changed, the wavelength of the acoustic wave within the acousto-optic crystal is changed, and accordingly, the diffraction angle is changed, thereby controlling the deflection direction of the processing laser light. In one embodiment, the RF driving member (1130c) generates an RF signal in the range of 30 MHz to 200 MHz to form a deflection angle in the range of several mrad. The RF driving member (1130c) variably controls the RF frequency according to a control signal output from the AOD scan pattern control member (1650c) of the AI ​​parameter control unit (1650) described later, so that processing laser light is irradiated to a desired processing point on the circuit board (O).

[0102] The first-order diffraction selection member (1130d) blocks zero-order diffraction light emitted from the X-axis AOD member (1130a) and the Y-axis AOD member (1130b), and allows only first-order diffraction light to pass through. Light incident on the AOD is split into zero-order diffraction light (unbiased residual light) and first-order diffraction light (biased light) and emitted, but in the present invention, only the first-order diffraction light is used for processing. In one embodiment, the first-order diffraction selection member (1130d) is configured in the form of a slit, and blocks zero-order diffraction light by forming an opening only at the location where the first-order diffraction light passes through. The zero-order diffracted light blocked by the first-order diffracted selection member (1130d) travels in a direction different from the first-order diffracted light (the direction indicated by the "fs laser beam scattering" region in FIG. 3) as shown in FIG. 3, and is safely absorbed by a beam dump (not shown) outside the optical system, thereby preventing scattering within the optical system and the resulting noise generation.

[0103] The AOD calibration unit (1130e) periodically measures the correspondence between the acoustic frequency of the AOD unit (1130) and the deflection angle of the processing laser light to calibrate the frequency of the RF signal applied to the RF driving unit (1130c). Since the acoustic frequency-deflection angle relationship of the AOD may fluctuate over time due to the temperature, aging, etc. of the acousto-optical crystal, the AOD calibration unit (1130e) performs calibration at a fixed period (e.g., per processing substrate or per 8 hours) to maintain the deflection precision of the AOD unit (1130) for a long period. In one embodiment, the AOD calibration unit (1130e) operates by irradiating the processing laser light onto a circuit board (O) or a separate calibration target while changing the RF frequency stepwise, and by measuring the actual irradiation position through the optical focusing unit (1200) and the vision unit (1300) to update the mapping data between the RF frequency and the actual deflection position.

[0104] <DM부>

[0105] The DM section (1140) is positioned in the optical path between the AOD section (1130) and the optical focusing unit (1200) and has wavelength selectivity that reflects the processing laser light emitted from the AOD section (1130) and transmits the light in the visible light region. That is, the DM section (1140) is located at the rear end of the AOD section (1130) (or galvano-aux section (1150)) and the front end of the optical focusing unit (1200) in the processing beam incident path, and reflects the processing laser light toward the optical focusing unit (1200), while transmitting the reflected light and fluorescent emission (visible light region) from the circuit board (O) as is to be transmitted toward the vision unit (1300).

[0106] The DM section (1140) includes a dichroic mirror member (1140a) and a mirror holder member (1140b).

[0107] The dichroic mirror member (1140a) is an optical member having wavelength selectivity, which reflects the wavelength of the processing laser light (e.g., 1030 nm, 515 nm, or 343 nm) and transmits the wavelength of the visible light region (e.g., 400 nm to 700 nm). In one embodiment, the dichroic mirror member (1140a) may be implemented as a structure in which a multi-layer thin film coating is formed on a glass substrate. By controlling the refractive index and thickness of the multi-layer thin film coating, the reflectance and transmittance for a desired wavelength band can be precisely designed. In one embodiment, the dichroic mirror member (1140a) may be designed to have a reflectance of 99% or more for the processing laser light wavelength and a transmittance of 95% or more for light in the visible light region.

[0108] The mirror holder member (1140b) finely adjusts the angle with respect to the optical axis of the dichroic mirror member (1140a). In one embodiment, the mirror holder member (1140b) is equipped with a fine adjustment mechanism of three or four axes, so that the tilt angle and rotation angle of the dichroic mirror member (1140a) can be precisely adjusted. The angle adjustment of the dichroic mirror member (1140a) by the mirror holder member (1140b) is used to align the incident direction of the processing laser light to the optical focusing unit (1200), and can automatically perform correction when an optical path misalignment is detected by linking with the beam alignment correction unit (1696) of the control unit (1600) described later.

[0109] Galvanobo Jobu

[0110] The galvano-auxiliary section (1150) is positioned in the optical path between the AOD section (1130) and the DM section (1140) to further change the deflection direction of the processing laser light emitted from the AOD section (1130). That is, the galvano-auxiliary section (1150) is located at the rear end of the AOD section (1130) and the front end of the DM section (1140) in the processing beam incidence path, and performs the deflection of the processing laser light in cooperation with the AOD section (1130) in series.

[0111] By the division of roles between the galvano-auxiliary unit (1150) and the AOD unit (1130) of the present invention, the galvano-auxiliary unit (1150) is responsible for scanning a wide field of view area (e.g., a tens of mm × tens of mm area) on a circuit board (O), and the AOD unit (1130) is responsible for fine precision scanning within the wide field of view area determined by the galvano-auxiliary unit (1150) (e.g., precision bias within a several μm × several μm area). Through this division of roles, the wide field of view characteristics of the galvano scanner and the high-speed and high-precision characteristics of the AOD are combined, so that three characteristics of wide field of view + high speed + high precision, which could not be achieved with a conventional single scanner, are simultaneously achieved.

[0112] The galvano auxiliary member (1150) includes an X-galvano mirror member (1150a), a Y-galvano mirror member (1150b), and a galvano driving member (1150c).

[0113] The X-galvanometer mirror member (1150a) deflects the processing laser light in the X-axis direction. The X-galvanometer mirror member (1150a) is implemented as one of a pair of mirrors coupled to a galvanometer motor, and changes the reflection direction of the processing laser light by changing the angle of the mirror in proportion to the rotation angle of the galvanometer motor.

[0114] The Y-galvanometer mirror member (1150b) deflects the processing laser light in the Y-axis direction. The Y-galvanometer mirror member (1150b) is arranged in series with the X-galvanometer mirror member (1150a), and deflects the processing laser light deflected in the X-axis direction by the X-galvanometer mirror member (1150a) back into the Y-axis direction so that the processing laser light reaches any point on a two-dimensional plane.

[0115] The galvanometer driving member (1150c) controls the rotation angle of the X-galvanometer mirror member (1150a) and the Y-galvanometer mirror member (1150b). The galvanometer driving member (1150c) applies a control voltage or a control current to the galvanometer motors respectively coupled to the X-galvanometer mirror member (1150a) and the Y-galvanometer mirror member (1150b) to precisely control the rotation angle and rotation speed of the mirrors. The galvanometer driving member (1150c) coordinately controls the X-galvanometer mirror member (1150a) and the Y-galvanometer mirror member (1150b) according to the processing path information output from the processing path optimization member (1650f) of the AI ​​parameter control unit (1650) described later, and irradiates a processing laser light at a desired location within a wide field of view on the circuit board (O).

[0116] Shutter section

[0117] The shutter unit (1160) blocks the optical path of the processing laser light while the processing sequence is not being performed, thereby preventing unnecessary light irradiation to the circuit board (O). In one embodiment, the shutter unit (1160) may be positioned on the optical path between the processing laser unit (1110) and the beam expander unit (1120), or on the optical path between the beam expander unit (1120) and the AOD unit (1130).

[0118] The shutter unit (1160) includes a beam blocking member (1160a) and a shutter driving member (1160b).

[0119] The beam blocking member (1160a) is provided to move in and out along the optical path of the processing laser light to selectively block the processing laser light. In one embodiment, the beam blocking member (1160a) may be implemented in the form of a black anodized metal plate or a mirror with a reflective coating. When processing is not performed, the beam blocking member (1160a) enters the optical path to block the processing laser light, and when processing is performed, the beam blocking member (1160a) moves out of the optical path to allow the processing laser light to proceed normally.

[0120] The shutter driving member (1160b) drives the beam blocking member (1160a) to block or open the optical path. In one embodiment, the shutter driving member (1160b) may be implemented as a solenoid, an electromagnet, or a motor, and switches the position of the beam blocking member (1160a) according to a control signal output from the processing sequence control unit (1610) of the control unit (1600). In one embodiment, the shutter driving member (1160b) has a fast response time of a few milliseconds (ms) or less, so it can respond quickly to the start and end of the processing sequence.

[0121] The laser irradiation unit (1100) is optically connected to the optical focusing unit (1200) so that the processing laser light is focused onto the circuit board (O) through the optical focusing unit (1200), and the processing operation is controlled by the processing sequence control unit (1610) and the synchronization control unit (1620) of the control unit (1600). In addition, the processing parameters of the laser irradiation unit (1100) are varied in real time by control signals output from the laser parameter control member (1650b), AOD scan pattern control member (1650c), beam shape control member (1650d), and pulse timing control member (1650e) of the AI ​​parameter control unit (1650).

[0122] Optical Focusing Unit

[0123] The optical focusing unit (1200) focuses the processing laser light emitted from the laser irradiation unit (1100) onto the surface of the circuit board (O) and receives reflected light and emitted light from the circuit board (O). The optical focusing unit (1200) is positioned between the DM section (1140) of the laser irradiation unit (1100) and the stage unit (1500), and integrates the optical path of the processing laser light and the optical path of the light incident on the vision unit (1300) coaxially so that processing and image acquisition can be performed on the same optical axis.

[0124] The optical focusing unit (1200) includes an objective lens part (1210), a tube lens part (1220), and a BS part (1230).

[0125] The light propagation path in the optical focusing unit (1200) is as follows. First, the processing laser light incident from the laser irradiation unit (1100) is reflected from the DM section (1140) and then focused onto the surface of the circuit board (O) through the objective lens section (1210). That is, the processing laser light propagates along the path of DM section (1140) → objective lens section (1210) → circuit board (O). Second, the fluorescent illumination light and reflected illumination light from the illumination unit (1400) are combined in the BS section (1230), pass through the DM section (1140), and then irradiate onto the circuit board (O) through the objective lens section (1210). That is, the illumination light propagates along the path of illumination unit (1400) → BS section (1230) → DM section (1140) → objective lens section (1210) → circuit board (O). Third, the reflected light and fluorescent light generated from the circuit board (O) are received by the objective lens unit (1210), pass through the DM unit (1140), and branch into an optical path on the vision unit (1300) side at the BS unit (1230), pass through the tube lens unit (1220), and are formed on the camera unit (1310) of the vision unit (1300). That is, the vision light propagates along the path of circuit board (O) → objective lens unit (1210) → DM unit (1140) → BS unit (1230) → tube lens unit (1220) → camera unit (1310).

[0126] The optical focusing unit (1200) adopts a configuration in which the processing optical path and the vision optical path are integrated coaxially, unlike the method in which the processing optical path and the vision optical path are configured separately and equipped with separate cameras and optical systems in a conventional circuit pattern repair device. As a result, the structure of the optical system is simplified, and the processing area and the image acquisition area are precisely aligned, thereby improving the precision of determining the processing position.

[0127] Objective lens section

[0128] The objective lens unit (1210) is positioned in the optical path between the DM unit (1140) and the stage unit (1500) to focus the processing laser light reflected from the DM unit (1140) onto the surface of the circuit board (O). At the same time, the objective lens unit (1210) also performs the role of receiving reflected light and fluorescent light from the circuit board (O) and transmitting them to the DM unit (1140). That is, the objective lens unit (1210) is a bidirectional optical member that simultaneously performs the focusing of the processing laser light and the receiving of the vision light with a single optical member.

[0129] The objective lens portion (1210) includes an objective lens member (1210a) and a lens exchange member (1210b).

[0130] The objective lens member (1210a) focuses processing laser light onto the circuit board (O). In one embodiment, the objective lens member (1210a) can be implemented as an infinity-corrected objective lens composed of a combination of multiple optical lenses. The infinity-corrected objective lens has a structure in which incident light is incident as parallel light, passes through the objective lens, and is focused at a point of light collection on the circuit board (O). This has the advantage that image quality is maintained even if the optical path length between the objective lens member (1210a) and the tube lens member (1220a) is variable.

[0131] In one embodiment, the objective lens member (1210a) may have a numerical aperture (NA) of 0.3 or more and 0.9 or less. As the numerical aperture increases, the size of the light-gathering spot becomes smaller, making it possible to process finer circuit pattern defects, but there is a trade-off relationship in that the working distance becomes shorter. In one embodiment, the objective lens member (1210a) of the present invention is designed to have a numerical aperture in the range of 0.5 to 0.7 and a working distance in the range of 5 mm to 20 mm, thereby securing both fine processing precision and sufficient working space.

[0132] In one embodiment, the objective lens member (1210a) may adopt an apochromatic or superchromatic optical design in which chromatic aberration is corrected for both the wavelength of the processing laser light (e.g., 1030 nm, 515 nm, or 343 nm) and the wavelength of the vision light (e.g., 400 nm to 700 nm). Due to the chromatic aberration correction, the focusing point of the processing laser light and the imaging point of the vision light are formed at the same location on the same optical axis, thereby minimizing the alignment error between the processing position and the image acquisition position.

[0133] The lens exchange member (1210b) has a plurality of objective lenses having different magnifications and automatically switches so that one of the objective lenses is selectively placed on the optical path of the optical focusing unit (1200) according to the size of the circuit pattern defect and the required level of processing precision. In one embodiment, the lens exchange member (1210b) may be implemented as an automatic turret structure equipped with four objective lenses with magnifications of 5x, 10x, 20x, and 50x, and rotates so that an objective lens of an appropriate magnification is automatically placed on the optical path based on a preset precision requirement level according to the risk score or defect type calculated by the defect importance evaluation member (1640f) of the AI ​​vision control unit (1640) described later. For example, when the size of the defect is relatively large (e.g., 10 μm or more), a low magnification (5x or 10x) objective lens is selected so that processing over a wide field of view is performed quickly, and when the size of the defect is fine (e.g., 2 μm or less), a high magnification (20x or 50x) objective lens is selected so that fine processing precision is ensured.

[0134] Tube lens section

[0135] The tube lens unit (1220) is positioned in the optical path between the BS unit (1230) within the optical focusing unit (1200) and the camera unit (1310) of the vision unit (1300), and forms an image of reflected light and fluorescent light received from the objective lens unit (1210) on the image sensor member (1310a) of the camera unit (1310) of the vision unit (1300). The tube lens unit (1220) forms an imaging optical system together with the objective lens member (1210a), and converges the vision light emitted as parallel light from the objective lens member (1210a) back to a single point to form an image of the circuit board (O) on the light receiving surface of the image sensor member (1310a).

[0136] The tube lens portion (1220) includes a tube lens member (1220a).

[0137] The tube lens member (1220a) forms reflected light and fluorescent light emission onto the image sensor member (1310a) of the camera unit (1310) of the vision unit (1300). In one embodiment, the tube lens member (1220a) may be implemented as an imaging lens having a focal length in the range of 100 mm to 300 mm. The overall imaging magnification of the optical system is determined according to the ratio of the focal lengths of the tube lens member (1220a) and the objective lens member (1210a), and in one embodiment, the imaging magnification may be variably set in the range of 5 to 50 times. In addition, in one embodiment, the tube lens member (1220a) adopts an apochromatic optical design with chromatic aberration corrected for the visible light region (400 nm to 700 nm) to ensure that the imaging quality of the fluorescent image and the reflected image is at an equivalent level.

[0138] <BS부>

[0139] The BS section (1230) is positioned on the optical path between the tube lens section (1220) and the objective lens section (1210) within the optical focusing unit (1200), and branches or combines the optical path of the processing laser light and the optical path of the light incident on the vision unit (1300). That is, the BS section (1230) is a branching and combining optical member that integrates and processes the illumination optical path (path from the BS section to the circuit board) irradiated on the circuit board (O) and the vision optical path (path from the BS section to the camera) from the circuit board (O) to the camera section (1310) into the same optical member.

[0140] The BS section (1230) includes a beam splitter member (1230a) and a polarizing beam splitter member (1230b).

[0141] The beam splitter member (1230a) splits or combines the optical path of the processing laser light and the optical path of the light incident on the vision unit (1300). In one embodiment, the beam splitter member (1230a) may be implemented as a cube beam splitter or a plate beam splitter. The cube beam splitter has a structure in which two right-angle prisms are bonded together, and a semi-transparent coating is formed on the bonding surface to reflect a certain proportion of the incident light and transmit the remainder. The plate beam splitter has a structure in which a semi-transparent coating is formed on a single flat glass substrate, and has the advantage of having less influence on the optical path length compared to the cube type.

[0142] In one embodiment, the beam splitter member (1230a) may be designed to have a reflectance and transmittance of 50:50 for the incident light, and in another embodiment, it may be designed with an asymmetric ratio such as 30:70 or 70:30. An asymmetric ratio design is advantageous for maximizing the amount of light toward a specific light path (e.g., illumination light path) while appropriately controlling the amount of light toward an opposite light path (e.g., vision light path).

[0143] The polarizing beam splitter member (1230b) is positioned at the location where the optical paths of the fluorescent illumination unit (1410) and the coaxial illumination unit (1420) intersect, and synthesizes the fluorescent illumination light and the reflected illumination light into the same optical path according to polarization characteristics. In one embodiment, the polarizing beam splitter member (1230b) can be implemented as a cube-shaped polarizing beam splitter and has the characteristic of transmitting light of the p-polarization component and reflecting light of the s-polarization component.

[0144] The operation of the polarizing beam splitter member (1230b) is as follows. Light emitted from the fluorescent illumination unit (1410) passes through the polarizer member (p) (1410b) to be shaped into p-polarized light and then enters the polarizing beam splitter member (1230b), and is transmitted as is due to the polarization characteristics of the polarizing beam splitter member (1230b) and reaches the beam splitter member (1230a). Meanwhile, light emitted from the coaxial illumination unit (1420) passes through the polarizer member (s) (1420b) to be shaped into s-polarized light and then enters the polarizing beam splitter member (1230b), and is reflected due to the polarization characteristics of the polarizing beam splitter member (1230b) and reaches the beam splitter member (1230a) in the same optical path as the transmitted p-polarized light. By synthesizing the optical path based on these polarization characteristics, the fluorescent illumination light and the reflected illumination light are irradiated onto the circuit board (O) along the same optical path, while the polarization states of the two illumination lights are separated from each other, thereby minimizing optical interference.

[0145] In one embodiment, the BS section (1230) may further include a beam synthesizer (not shown) in addition to the beam splitter member (1230a) and the polarizing beam splitter member (1230b). The beam synthesizer is an optical member that assists in synthesizing the optical paths from the fluorescent illumination section (1410), the coaxial illumination section (1420), and the ring illumination section (1430) into the same coaxial optical path, and may be implemented in the form of an optical flat plate or cube with a multilayer thin film coating formed thereon.

[0146] Additionally, in one embodiment, the BS section (1230) may be implemented as a variable beam splitter with an adjustable reflectance coating applied to the optical path of the beam splitter member (1230a), in which case the ratio of reflectance to transmittance is dynamically adjusted according to a control signal from the control unit (1600) described later, so that the distribution of light can be optimized according to processing conditions or image acquisition conditions.

[0147] The optical focusing unit (1200) is optically connected to the DM section (1140) of the laser irradiation unit (1100) to receive processing laser light, is optically connected to the camera section (1310) of the vision unit (1300) to transmit vision light, and is optically connected to the fluorescent lighting section (1410), coaxial lighting section (1420), and ring lighting section (1430) of the lighting unit (1400) to receive lighting light. Additionally, since the focusing state of the processing laser light is determined by the optical axis distance between the stage unit (1500) and the optical focusing unit (1200), the optical focusing unit (1200) cooperates with the autofocus section (1560) described later to ensure that the focus of the processing laser light is accurately aligned with the surface of the circuit board (O).

[0148] Vision Unit

[0149] The vision unit (1300) acquires an image of the circuit board (O) via the optical focusing unit (1200). Unlike the conventional method of separately providing a camera for acquiring fluorescent images and a camera for acquiring reflective images, the vision unit (1300) adopts a configuration that alternately acquires fluorescent images and reflective images in time division through a single image sensor. As a result, the structure of the optical system is simplified, and since the two images are acquired through the same optical axis and the same image sensor, the precision of registration at the pixel level is improved, and optical deviations or calibration errors between cameras are fundamentally eliminated.

[0150] The vision unit (1300) includes a camera unit (1310), a fluorescence blocking filter unit (1320), and an image processing unit (1330).

[0151] The conversion flow of light and signals in the vision unit (1300) is as follows. Fluorescent light and reflected light from the circuit board (O) pass sequentially through the objective lens section (1210), DM section (1140), BS section (1230), and tube lens section (1220) of the optical focusing unit (1200) and are incident on the vision unit (1300). The incident vision light passes through the fluorescence blocking filter section (1320) and is then formed on the image sensor member (1310a) of the camera section (1310) and converted into an electrical signal, and the converted electrical signal is processed into digital image data in the image processing section (1330) and output as a fluorescent image or a reflected image.

[0152] Camera Department

[0153] The camera unit (1310) is equipped with a single image sensor to acquire both fluorescent images and reflection images. That is, unlike the conventional dual camera method, the camera unit (1310) performs a time-division operation in which a single image sensor is synchronized with the time-division switching of the lighting unit (1400) to acquire a fluorescent image in a fluorescent lighting state and a reflection image in a reflection lighting state.

[0154] The camera unit (1310) includes an image sensor member (1310a), a global shutter control member (1310b), and an exposure time control member (1310c).

[0155] The image sensor member (1310a) is a photoelectric conversion member that receives reflected light and fluorescent light emission from a circuit board (O) and converts them into electrical signals, and in one embodiment, it can be implemented in a CCD (Charge-Coupled Device) or CMOS (Complementary Metal-Oxide-Semiconductor) manner. The CCD method has the advantage of excellent noise characteristics and high photoelectric conversion efficiency, while the CMOS method has the advantage of enabling high-speed operation and low-power operation.

[0156] In one embodiment, the image sensor member (1310a) has a resolution of 2 million pixels (2 megapixels) or more and 20 million pixels (20 megapixels) or less, and may have a frame rate of 30 frames (30fps) or more per second and 200 frames (200fps). In addition, the image sensor member (1310a) is designed to have excellent light sensitivity for the visible light region (e.g., 400nm to 700nm), and in one embodiment, a back-side illuminated (BSI) structure may be adopted to improve quantum efficiency.

[0157] The global shutter control member (1310b) performs simultaneous exposure and simultaneous reading for all pixels of the image sensor member (1310a). The global shutter method is a method in which all pixels start exposure simultaneously and end exposure simultaneously at the same point in time, and has the advantage of not causing motion artifacts in dynamic scenes compared to the rolling shutter method, in which exposure proceeds sequentially line by line. In the present invention, since the fluorescent lighting and the reflective lighting are alternately turned ON / OFF at intervals of several milliseconds to tens of milliseconds by the time-division switching of the lighting controller (1440), it is essential to secure time alignment precision using the global shutter method. The global shutter control member (1310b) receives a trigger pulse output from the synchronization pulse generating member (1440b) of the lighting controller (1440) and performs simultaneous exposure that is accurately synchronized in the fluorescent lighting ON section and the reflective lighting ON section, respectively.

[0158] The exposure time control member (1310c) variably controls the exposure time of the image sensor member (1310a). Since the intensity of fluorescence emission in a fluorescent image is generally weaker than the intensity of reflected light in a reflected image, a relatively long exposure time is required when acquiring a fluorescent image, and a relatively short exposure time may be applied when acquiring a reflected image. In one embodiment, the exposure time control member (1310c) can variably set the exposure time in the range of 10 microseconds (μs) to 100 milliseconds (ms), and in cooperation with the light intensity control member (1440c) of the lighting controller (1440), it independently optimizes the exposure conditions of the fluorescent image and the reflected image, respectively.

[0159] Fluorescent blocking filter section

[0160] The fluorescence blocking filter unit (1320) is coupled to the incident end of the camera unit (1310) to block the output wavelength of the fluorescence illumination of the illumination unit (1400), while allowing the wavelength of the fluorescence emission from the circuit board (O) and the wavelength of the reflected illumination to pass through. The operation of the fluorescence blocking filter unit (1320) plays a key role in the acquisition of the fluorescence image of the present invention.

[0161] Specifically, ultraviolet or near-ultraviolet (NUV) light irradiated onto the circuit board (O) from the fluorescent illumination unit (1410) may be partially reflected from the surface of the circuit board (O) and incident on the camera unit (1310) with high intensity. If this reflected NUV light reaches the image sensor member (1310a) as is, a problem arises where the fluorescent emission signal (relatively weak visible light) generated in the substrate portion of the circuit board (O) is obscured by the strong NUV reflected light and cannot be detected. The fluorescent blocking filter unit (1320) selectively blocks this NUV reflected light just before it is incident on the camera unit (1310), thereby allowing only the fluorescent emission signal to reach the camera unit (1310). As a result, the contrast in the fluorescent image is maximized, allowing circuit pattern defects to be clearly detected.

[0162] The fluorescent blocking filter section (1320) includes a notch filter member (1320a) and a filter holder member (1320c).

[0163] The notch filter member (1320a) blocks the output wavelength band of the fluorescent illumination unit (1410). The notch filter is an optical filter that selectively blocks (reflects or absorbs) only a specific narrow wavelength band and allows all other wavelengths to pass through. In one embodiment, it has a blocking band with a width of ±10 nm to ±30 nm centered on the output wavelength (e.g., 405 nm) of the fluorescent illumination unit (1410) and is designed to block the corresponding wavelength with an optical density of 99% or more. In one embodiment, the notch filter member (1320a) may be implemented as a multilayer thin film coating or a volume phase hologram structure on a glass substrate.

[0164] In one embodiment, the notch filter member (1320a) may be implemented as a single-notch filter that blocks only a single wavelength band, and in another embodiment, it may be implemented as a multi-notch filter that blocks multiple wavelength bands simultaneously. When using a multi-notch filter, it is possible to additionally prevent scattered light of the processing laser light from being incident on the camera unit (1310) in addition to the output wavelength of the fluorescent lighting unit (1410), so there is an advantage that a vision image can be stably acquired even during processing.

[0165] Additionally, in one embodiment, the fluorescence blocking filter section (1320) may further include a longpass filter (not shown) in addition to the notch filter member (1320a). The longpass filter is a filter that allows only light of a specific wavelength or higher to pass through, blocking light of the NUV output wavelength (e.g., 405 nm) or lower of the fluorescence illumination section (1410) and allowing only visible light of a higher wavelength (e.g., 450 nm or higher) to pass through. By combining the notch filter member (1320a) and the longpass filter, the blocking performance for NUV wavelengths is further improved, while allowing sufficient light of the fluorescence emission region to pass through at the same time.

[0166] The filter holder member (1320c) mounts the notch filter member (1320a) to the entrance end of the camera unit (1310). In one embodiment, the filter holder member (1320c) may be implemented as a screw coupling structure or a bayonet coupling structure compatible with the lens mount of the camera unit (1310), and maintains the notch filter member (1320a) in a state of accurate optical axis alignment. In one embodiment, the filter holder member (1320c) may be implemented as a rotating filter wheel structure so as to be rotatably arranged so that a plurality of notch filter members (1320a) are selectively placed on the optical path. Due to the rotating filter wheel structure, the notch filter can be quickly replaced with an appropriate notch filter in response to changes in the material of the circuit board (O) or the wavelength of the light source of the fluorescent lighting unit (1410).

[0167] Image Processing Unit

[0168] The image processing unit (1330) combines or compares fluorescent images and reflection images acquired from the camera unit (1310) to identify the location and shape of circuit pattern defects. The image processing unit (1330) performs the role of detecting defects by processing digital image data output from the camera unit (1310) using a rule-based image processing algorithm, and this can operate in parallel or complementarily with the neural network-based defect detection of the AI ​​vision control unit (1640) described later.

[0169] The image processing unit (1330) includes an image matching member (1330a), a difference image generation member (1330b), a rule-based defect region extraction member (1330c), and a rule-based defect classification member (1330d).

[0170] The image matching member (1330a) aligns the pixels of the fluorescent image and the reflection image. In the present invention, since a single image sensor member (1310a) acquires the fluorescent image and the reflection image in time division, the coordinate systems of the two images are, in principle, identical. However, since minute pixel misalignment between the two images may occur due to minute vibrations of the stage unit (1500) or thermal deformation of the circuit board (O) during the time division time interval (several milliseconds to tens of milliseconds), the image matching member (1330a) corrects these residual errors to accurately align the two images at the pixel level. In one embodiment, the image matching member (1330a) calculates a displacement vector between the two images using any one of a phase correlation algorithm, a feature point matching algorithm, or an optical flow algorithm, and aligns the two images by correcting one of the images by translation or rotation according to the calculated displacement vector.

[0171] The differential image generating member (1330b) generates a differential image by calculating the difference between the fluorescent image and the reflection image aligned in the image matching member (1330a). The differential image is an image formed by calculating the difference in pixel values ​​between the two images on a pixel-by-pixel basis. Normal pattern areas that appear in common in both images are represented darkly with small differential values, while defect areas that appear differently in the two images are represented brightly with large differential values. In one embodiment, the differential image generating member (1330b) may use various differential operation methods, such as weighted subtraction, normalized subtraction, or ratio image, in addition to simple pixel subtraction.

[0172] Through difference operations, common information appearing identically in both the fluorescence and reflection images (e.g., the location of normal circuit patterns, the general appearance of the substrate surface) is removed, and only information differing between the two images (e.g., micro-short defects detected only in the fluorescence image, specific surface defects detected only in the reflection image) is highlighted. Due to this effect of common-mode rejection, defects that are difficult to detect in a single image become clearly visible in the difference image.

[0173] The rule-based defect region extraction member (1330c) applies a rule-based image processing algorithm to the difference image generated by the difference image generation member (1330b) to extract a region where circuit pattern defects exist. In one embodiment, the rule-based defect region extraction member (1330c) sequentially applies classical image processing algorithms such as thresholding, morphological operation, connected component analysis, or blob analysis. Specifically, defect candidate pixels are extracted by applying a preset threshold to the difference image, noise is removed by morphological operation (erosion, dilation, opening, closing, etc.), and adjacent defect pixels are grouped into defect object units by connected component analysis to extract a defect region. The rule-based defect region extraction member (1330c) can be operated in parallel with the defect pattern recognition member (1640c) of the AI ​​vision control unit (1640) described later in one embodiment, and can also be used to mutually verify the rule-based detection result and the neural network-based detection result.

[0174] The rule-based defect classification member (1330d) classifies the circuit pattern defect extracted from the rule-based defect area extraction member (1330c) into one of a short defect, a protruding defect, and a residue defect. In one embodiment, the rule-based defect classification member (1330d) analyzes the shape features (e.g., area, perimeter, aspect ratio, circularity) of the extracted defect area and determines the type of defect according to a pre-set classification rule. For example, if a defect is detected as a long, narrow area connecting two metal patterns, it is classified as a short defect; a defect protruding outward from the edge of the metal pattern is classified as a protruding defect; and a defect formed at a location separated from the metal pattern is classified as a residue defect.

[0175] The vision unit (1300) is optically connected to the optical focusing unit (1200) to receive an image of the circuit board (O), is electrically connected to the lighting controller (1440) of the lighting unit (1400) to perform an exposure operation synchronized with time-division switching, and is electrically connected to the processing sequence control unit (1610), AI vision control unit (1640), and synchronization control unit (1620) of the control unit (1600) to transmit image data and receive control signals.

[0176] Lighting Unit

[0177] The lighting unit (1400) selectively irradiates fluorescent lighting and reflective lighting onto the circuit board (O) to acquire fluorescent images and reflective images of the circuit board (O) through the vision unit (1300). The lighting unit (1400) is composed of a fluorescent lighting unit (1410) that induces fluorescent light emission on the substrate portion of the circuit board (O), a coaxial lighting unit (1420) and a ring lighting unit (1430) that form reflected light on the surface of the circuit board (O), and a lighting controller (1440) that controls the ON / OFF timing of these lighting units in a time-division switching manner.

[0178] The lighting unit (1400) includes a fluorescent lighting section (1410), a coaxial lighting section (1420), a ring lighting section (1430), and a lighting controller (1440).

[0179] The core operating principle of the lighting unit (1400) is time-division switching. That is, the fluorescent lighting unit (1410) and the coaxial lighting unit (1420) (or ring lighting unit (1430)) are not simultaneously turned ON at any given time, but are controlled to be mutually exclusive so that only one of them is turned ON at any given time. Through this time-division switching, a single camera unit (1310) can alternately acquire fluorescent images and reflection images depending on the time.

[0180] Fluorescent lighting section

[0181] The fluorescent illumination unit (1410) irradiates light of ultraviolet or near-ultraviolet wavelength onto the circuit board (O) to induce fluorescent emission on the substrate portion of the circuit board (O). The fluorescent illumination unit (1410) serves as a light source for acquiring a fluorescent image according to the present invention, and the light emitted from the fluorescent illumination unit (1410) is irradiated onto the circuit board (O) via the polarizing beam splitter member (1230b) and beam splitter member (1230a) of the BS unit (1230), the dichroic mirror member (1140a) of the DM unit (1140), and the objective lens unit (1210) of the optical focusing unit (1200).

[0182] The fluorescent lighting unit (1410) includes an NUV light source member (1410a) and a polarizer member (p) (1410b).

[0183] The NUV light source member (1410a) generates light of ultraviolet or near-ultraviolet wavelength. In one embodiment, the NUV light source member (1410a) may be implemented as an ultraviolet LED or ultraviolet laser diode having a center wavelength in the range of 350 nm to 410 nm. More specifically, the NUV light source member (1410a) may be implemented as an LED with a center wavelength of 405 nm, so that fluorescent emission in the visible light region (mainly 450 nm to 550 nm) can be efficiently induced from organic materials such as resin, ABF, and polyimide used in the substrate portion of the circuit board (O).

[0184] When light in the NUV wavelength range is irradiated onto the circuit board (O), organic molecules in the substrate part absorb light energy and transition to an excited state, and then return to the ground state while emitting photons in the visible light range (fluorescence emission). Meanwhile, the metal pattern (e.g., copper) on the circuit board (O) does not cause fluorescence emission even when it absorbs NUV light, but rather converts the light energy into heat or reflects it; consequently, a stark contrast is formed in the fluorescent image, where the substrate part is displayed brightly and the metal pattern is displayed darkly.

[0185] In one embodiment, the NUV light source member (1410a) may additionally include an illumination focusing lens (not shown) and an illumination notch filter (not shown) in the light path. The illumination focusing lens shapes the light emitted from the NUV light source member (1410a) into parallel light or appropriately focused light, so that it is efficiently irradiated onto the circuit board (O) through the BS section (1230) and the optical focusing unit (1200). The illumination notch filter blocks visible light or infrared light components that may be incidentally included in the output light of the NUV light source member (1410a), thereby further enhancing the contrast of the fluorescent image.

[0186] The polarizer member (p) (1410b) shapes the light emitted from the NUV light source member (1410a) into p-polarization. A polarizer is an optical member that allows only the component of a specific polarization direction of the incident light to pass through, and a p-polarization polarizer allows only the electric field component parallel to the plane of incidence to pass through. The p-polarized light shaped by the polarizer member (p) (1410b) has a transmission characteristic in the polarization beam splitter member (1230b) of the subsequent BS unit (1230), and is irradiated onto the circuit board (O) while separated from the s-polarized light of the coaxial illumination unit (1420). In one embodiment, the polarizer member (p) (1410b) may be implemented as any one of an absorptive polarizer film, a wire grid polarizer, or a photonic crystal polarizer.

[0187] Coaxial Lighting Unit

[0188] The coaxial illumination unit (1420) irradiates light in the visible light region onto the circuit board (O) in a direction coaxial with the optical axis of the optical focusing unit (1200). Coaxial illumination is a method in which the illumination light is incident on the circuit board (O) along the optical axis of the optical focusing unit (1200). Since the optical axis of the optical member and the illumination light path coincide, specular reflection components are well formed, allowing for the acquisition of a reflection image that richly reflects information regarding the surface shape of the circuit board (O).

[0189] The coaxial illumination unit (1420) includes an RB light source member (1420a), a polarizer member(s) (1420b), and a beam alignment member (1420c).

[0190] The RB light source member (1420a) generates red (R) and blue (B) visible light. In one embodiment, the RB light source member (1420a) may be implemented as a multi-wavelength LED light source or a white LED, which is a combination of a red LED and a blue LED. Since red (e.g., near 625 nm) and blue (e.g., near 460 nm) visible light exhibit different reflection characteristics on the copper metal pattern on the circuit board (O), using a combination of the two wavelengths allows for the acquisition of richer reflection information than when acquiring a single wavelength. Additionally, in one embodiment, the RB light source member (1420a) may additionally include a green (G) LED to provide light across the entire RGB range.

[0191] In one embodiment, the RB light source member (1420a) may additionally include an illumination focusing lens (not shown) in the light path. The illumination focusing lens shapes the light emitted from the RB light source member (1420a) into parallel light or appropriately focused light, so that it is efficiently irradiated onto the circuit board (O) through the BS part (1230) and the optical focusing unit (1200).

[0192] The polarizer member(s) (1420b) shapes the light emitted from the RB light source member (1420a) into s-polarization. The s-polarization polarizer allows only the electric field component perpendicular to the plane of incidence to pass through, and the s-polarization light shaped by the polarizer member(s) (1420b) has a reflection characteristic at the polarization beam splitter member (1230b) of the subsequent BS section (1230), so that it is separated from the p-polarization light of the fluorescent illumination section (1410) while being combined into the same optical path and irradiated onto the circuit board (O).

[0193] The beam alignment member (1420c) directs light emitted from the RB light source member (1420a) to the surface of the circuit board (O) in a coaxial direction through the BS section (1230). In one embodiment, the beam alignment member (1420c) is implemented as a mirror, prism, or optical mount placed on the optical path, and finely adjusts the optical path so that light from the coaxial illumination section (1420) travels in a direction that exactly matches the optical axis of the optical focusing unit (1200). Since the coaxial alignment precision of the beam alignment member (1420c) directly affects the uniformity of the reflected image, in one embodiment, the beam alignment member (1420c) is provided with a precision adjustment mechanism of three or four axes to enable fine adjustment of the angle and position of the optical path.

[0194] Ring lighting

[0195] The ring illumination unit (1430) irradiates light in the visible light region onto the circuit board (O) from a side direction. Side illumination (or oblique illumination) is a method in which light is incident on the surface of the circuit board (O) at an oblique angle, and since shadows are emphasized by the surface irregularities of the circuit board (O), it is advantageous for detecting fine surface defects (e.g., protrusions, steps) that are difficult to detect with coaxial illumination alone.

[0196] The ring lighting unit (1430) includes a ring LED array member (1430a) and an angle adjustment member (1430b).

[0197] A ring LED array member (1430a) is arranged in a ring shape on the outer side of the objective lens portion (1210) of the optical focusing unit (1200) to irradiate visible light onto a circuit board (O) from a side direction. In one embodiment, the ring LED array member (1430a) is provided in a circular or polygonal array structure surrounding the objective lens member (1210a), and a plurality of LEDs are evenly arranged along the circumferential direction to provide uniform side illumination to the circuit board (O). In one embodiment, the ring LED array member (1430a) may be configured to include red (R), green (G), blue (B) LEDs or white LEDs, and light of an appropriate color may be selectively used depending on the material or defect type of the circuit board (O).

[0198] The angle adjustment member (1430b) variably adjusts the irradiation angle of the ring LED array member (1430a). Since the optimal value of the incident angle of the side illumination varies depending on the type and depth of the defect to be detected, allowing the incident angle to be variably adjusted by the angle adjustment member (1430b) enables flexible response to various defect detection conditions. In one embodiment, the angle adjustment member (1430b) may be implemented as a mechanical mechanism capable of rotating or tilting the ring LED array member (1430a) within a range of 0° to 60° with respect to the optical axis, or as a multi-ring structure in which a plurality of ring arrays are pre-arranged at different angles and selected from among them.

[0199] Lighting Controller

[0200] The lighting controller (1440) is electrically connected to each of the fluorescent lighting unit (1410), the coaxial lighting unit (1420), and the ring lighting unit (1430), performs time-division switching control so that at least one of the fluorescent lighting unit (1410), the coaxial lighting unit (1420), and the ring lighting unit (1430) is mutually exclusively turned ON / OFF, and outputs a trigger signal synchronized with the exposure timing of the camera unit (1310).

[0201] The lighting controller (1440) includes a switching timing control member (1440a), a synchronization pulse generating member (1440b), and a light intensity control member (1440c).

[0202] The switching timing control member (1440a) performs time-division switching control such that only one of the fluorescent lighting unit (1410), the coaxial lighting unit (1420), and the ring lighting unit (1430) is ON at any given time. That is, the switching timing control member (1440a) controls the fluorescent lighting unit (1410) to the ON state and the coaxial lighting unit (1420) and the ring lighting unit (1430) to the OFF state at any time point t1, switches the fluorescent lighting unit (1410) to the OFF state and the coaxial lighting unit (1420) to the ON state at time point t2, and switches the coaxial lighting unit (1420) to the OFF state and the ring lighting unit (1430) to the ON state at time point t3, thereby performing sequential time-division switching. In one embodiment, the switching period is determined in synchronization with the frame rate of the camera unit (1310) and is typically set to a level of several milliseconds to tens of milliseconds.

[0203] In one embodiment, the switching timing control member (1440a) can be implemented as a Field-Programmable Gate Array (FPGA), a microcontroller (MCU), or a digital signal processor (DSP), and precisely controls the ON / OFF timing of each lighting unit according to a preset switching sequence. In one embodiment, the response time of the switching operation is set to 1 microsecond (μs) or less, thereby minimizing the temporal discrepancy between the exposure of the camera unit (1310) and the lighting ON period.

[0204] The synchronization pulse generating member (1440b) generates a trigger pulse and outputs it to the camera unit (1310) so that the switching operation of the switching timing control member (1440a) and the exposure operation of the camera unit (1310) are synchronized. The trigger pulse generated by the synchronization pulse generating member (1440b) is input to the global shutter control member (1310b) of the camera unit (1310) in correspondence with the ON time of each lighting unit, causing the camera unit (1310) to start exposure during each lighting ON period and end exposure just before the end of the ON period. In one embodiment, the synchronization pulse generating member (1440b) is synchronized with a reference clock signal from the master clock member (1620a) of the synchronization control unit (1620) of the control unit (1600) described later, thereby securing the time synchronization precision of the entire system.

[0205] The light intensity control member (1440c) variably controls the light intensity of each of the fluorescent lighting unit (1410), the coaxial lighting unit (1420), and the ring lighting unit (1430). In one embodiment, the light intensity control member (1440c) controls the output light intensity of the lighting light within a range of 0% to 100% by varying the driving current applied to the LED of each lighting unit using a Pulse Width Modulation (PWM) method or an analog current control method. By controlling the light intensity, the light intensity of each lighting unit is set to an optimal value according to the difference in surface reflectance or fluorescent luminescence efficiency of the circuit board (O), so that the exposure of both the fluorescent image and the reflected image is performed within an appropriate range.

[0206] The lighting unit (1400) is optically connected to the BS section (1230) of the optical focusing unit (1200) to irradiate lighting light onto the circuit board (O), is electrically connected to the camera section (1310) of the vision unit (1300) to transmit an exposure trigger, and is electrically connected to the synchronization control section (1620) of the control unit (1600) to participate in the time synchronization of the entire system.

[0207] Stage Unit

[0208] The stage unit (1500) supports the circuit board (O), moves the circuit board (O) in the X-axis and Y-axis directions within a horizontal plane, and adjusts the position of the circuit board (O) in the optical axis direction (typically the Z-axis direction) of the optical focusing unit (1200). The stage unit (1500) stably fixes the circuit board (O) while precisely controlling its position so that sequential processing of multiple processing points on the circuit board (O) is possible.

[0209] The stage unit (1500) includes an X-axis stage section (1510), a Y-axis stage section (1520), an optical axis position adjustment section (1530), a substrate fixing section (1540), an alignment mark recognition section (1550), an autofocus section (1560), and a substrate thickness measuring section (1570).

[0210] The mechanical structure of the stage unit (1500) is as follows. An X-axis stage section (1510) is placed on the upper part of a base plate, a Y-axis stage section (1520) is stacked on the upper part of the X-axis stage section (1510), an optical axis position adjustment section (1530) is stacked on the upper part of the Y-axis stage section (1520), and a substrate fixing section (1540) is placed on the upper part of the optical axis position adjustment section (1530) to support a circuit board (O). That is, movement in the 3-axis orthogonal coordinate system of the XYZ three axes is implemented in a stacked structure, so that the circuit board (O) can be moved to any position in three-dimensional space.

[0211] <X축스테이지부>

[0212] The X-axis stage unit (1510) moves the circuit board (O) in the X-axis direction within a horizontal plane. In one embodiment, the X-axis stage unit (1510) is implemented with a structure combining a precision linear guide rail and a linear motor, and accurately moves the circuit board (O) to a preset X-axis coordinate.

[0213] The X-axis stage section (1510) includes an X-axis linear motor member (1510a) and an X-axis position sensor member (1510b).

[0214] The X-axis linear motor member (1510a) performs precise linear driving in the X-axis direction. In one embodiment, the X-axis linear motor member (1510a) can be implemented as a linear motor in which a magnet and a coil are arranged in a straight line, and controls the X-axis position of the circuit board (O) with a resolution of sub-micrometers (sub-μm). Linear motor driving has the advantage of having no backlash and excellent response speed compared to conventional ball screw driving, making it advantageous for precise alignment of the location of fine defects on the circuit board (O).

[0215] The X-axis position sensor member (1510b) measures the position of the X-axis stage member (1510) in real time and provides a feedback signal necessary for closed-loop control. In one embodiment, the X-axis position sensor member (1510b) may be implemented as a linear encoder or an optical scale and has a position resolution in the range of 10 nm to 100 nm. Position information measured by the X-axis position sensor member (1510b) is transmitted to the control unit (1600), an error with respect to the target position is calculated, and a correction signal is applied to the X-axis linear motor member (1510a), thereby performing closed-loop position control.

[0216] <Y축스테이지부>

[0217] The Y-axis stage section (1520) moves the circuit board (O) in the Y-axis direction within a horizontal plane. The Y-axis stage section (1520) is stacked orthogonally on top of the X-axis stage section (1510) to perform movement in the Y-axis direction independently of X-axis movement.

[0218] The Y-axis stage section (1520) includes a Y-axis linear motor member (1520a) and a Y-axis position sensor member (1520b).

[0219] The Y-axis linear motor member (1520a) performs precise linear driving in the Y-axis direction, and the Y-axis position sensor member (1520b) measures the position of the Y-axis stage member (1520) in real time. Since the specific configuration and operating principles of the Y-axis linear motor member (1520a) and the Y-axis position sensor member (1520b) are substantially the same as those of the aforementioned X-axis linear motor member (1510a) and X-axis position sensor member (1510b), redundant descriptions are omitted.

[0220] The X-axis stage section (1510) and the Y-axis stage section (1520) are typically combined in an H-drive or gantry structure so that the circuit board (O) can be moved to any XY coordinate within a horizontal plane.

[0221] Optical axis position adjustment unit

[0222] The optical axis direction position adjustment unit (1530) adjusts the distance in the optical axis direction (typically the Z-axis direction) between the circuit board (O) and the optical focusing unit (1200). By the optical axis direction position adjustment unit (1530), the focal point of the processing laser light is controlled to precisely match the surface of the circuit board (O), thereby ensuring processing precision.

[0223] The optical axis direction position adjustment unit (1530) includes an optical axis direction driving member (1530a) and a distance sensor member (1530b).

[0224] The optical axis driving member (1530a) moves the circuit board (O) in the optical axis direction. In one embodiment, the optical axis driving member (1530a) may be implemented as a precision linear actuator, a piezoelectric actuator, or a voice coil motor. Piezoelectric actuators have the advantage of having ultra-precision resolution at the nanometer (nm) level, and voice coil motors have excellent response speed, making them suitable for auto-focus operation. In one embodiment, the optical axis driving member (1530a) is arranged as a pair on both the left and right sides of the optical axis position adjustment unit (1530) to perform balanced linear movement.

[0225] The distance sensor member (1530b) measures the optical axis position of the optical axis position control unit (1530) in real time. In one embodiment, the distance sensor member (1530b) may be implemented as an optical distance sensor, a capacitive distance sensor, or a laser displacement meter. Distance information measured by the distance sensor member (1530b) is transmitted to the control unit (1600) and used for closed-loop control of the optical axis driving member (1530a).

[0226] <Board Fixing Unit>

[0227] The substrate fixing part (1540) is positioned above the optical axis position adjustment part (1530) to support and fix the circuit board (O). The substrate fixing part (1540) firmly fixes the circuit board (O) so that it maintains a stable position even during processing, and is provided so that the circuit board (O) can be easily separated after processing is completed.

[0228] The substrate fixing part (1540) includes a vacuum chuck member (1540a).

[0229] The vacuum chuck member (1540a) fixes the circuit board (O) to the upper surface of the substrate fixing part (1540) by a vacuum adsorption method. In one embodiment, the vacuum chuck member (1540a) is configured to include a chuck plate made of a porous ceramic or porous metal material and a vacuum channel formed inside or below the chuck plate. The vacuum channel is connected to a vacuum pump to form negative pressure between the chuck plate and the circuit board (O), thereby uniformly adsorbing and fixing the circuit board (O) to the surface of the chuck plate. In one embodiment, the vacuum adsorption pressure can be set in the range of -20kPa to -80kPa and is appropriately adjusted according to the size and thickness of the circuit board (O).

[0230] In one embodiment, the vacuum chuck member (1540a) may additionally include a clamp (not shown) structure. The clamp is a method of mechanically fixing the edge portion of the circuit board (O) and can be used when vacuum suction is insufficient or the surface of the circuit board (O) is not smooth.

[0231] Alignment Mark Recognition Unit

[0232] The alignment mark recognition unit (1550) captures and recognizes an alignment mark (fiducial mark) that has been pre-formed on the circuit board (O), and registers the design coordinate system of the circuit board (O) and the actual coordinate system of the stage unit (1500) based on the position information of the recognized alignment mark. Since slight positional misalignment or rotational misalignment may occur when the circuit board (O) is loaded onto the stage unit (1500), the alignment mark recognition unit (1550) compensates for these errors to ensure an accurate correspondence between the design coordinates (e.g., defect location in CAD data) and the actual processing coordinates.

[0233] The alignment mark recognition unit (1550) includes a mark imaging member (1550a) and a coordinate matching member (1550b).

[0234] The mark imaging member (1550a) captures an alignment mark on the circuit board (O). In one embodiment, the mark imaging member (1550a) may be implemented as a separate auxiliary camera or by sharing the main camera unit (1310) of the vision unit (1300). The separate camera method has the advantage of being able to perform alignment in parallel with the main processing and inspection sequence, while the shared method has the advantage of simplifying the optical system and reducing costs.

[0235] The coordinate matching member (1550b) extracts the coordinates of the alignment mark from the image captured by the mark capturing member (1550a) and aligns the extracted coordinates with the coordinates on the design data of the circuit board (O) to calculate a coordinate transformation matrix. In one embodiment, the coordinate matching member (1550b) calculates the center coordinates of the mark by applying pattern matching or moment analysis to the image of the alignment mark, and calculates a homography transformation matrix that expresses the difference between the calculated actual coordinates and the design coordinates as rotation, translation, and scaling transformations. The calculated coordinate transformation matrix is ​​transmitted to the control unit (1600), and in the subsequent processing sequence, all processing coordinates are corrected by this matrix and then converted into driving signals for the X-axis stage unit (1510) and the Y-axis stage unit (1520).

[0236] Autofocus

[0237] The autofocus unit (1560) measures the distance between the surface of the circuit board (O) and the objective lens unit (1210) of the optical focusing unit (1200) in real time, and automatically drives the stage unit (1500) based on the measured distance information to align the focus of the processing laser light with the surface of the circuit board (O). The autofocus unit (1560) compensates for focus deviation caused by flatness error or surface step difference of the circuit board (O), thereby ensuring uniform processing quality across the entire area of ​​the circuit board (O).

[0238] The autofocus unit (1560) includes a distance measuring member (1560a) and an optical axis direction correction member (1560b).

[0239] The distance measuring member (1560a) measures the distance to the surface of the circuit board (O). In one embodiment, the distance measuring member (1560a) may be implemented as a confocal displacement sensor, a laser triangulation sensor, a white-light interferometer, or a capacitive sensor. A confocal displacement sensor can achieve a resolution of tens of nanometers (nm) in an optically non-contact manner, making it particularly suitable for micro-precision machining applications such as the present invention.

[0240] The optical axis direction correction member (1560b) converts the distance information measured by the distance measuring member (1560a) into an optical axis direction driving value of the stage unit (1500) and outputs it to the stage unit (1500). That is, the optical axis direction correction member (1560b) calculates the distance error between the surface of the circuit board (O) and the objective lens unit (1210) from the distance measurement result, generates a driving signal of the optical axis direction position adjustment unit (1530) to compensate for this, and transmits it to the optical axis direction driving member (1530a). In one embodiment, the optical axis direction correction member (1560b) forms a real-time feedback control loop to immediately compensate for surface height changes that occur when the circuit board (O) is moved in the XY direction.

[0241] <Substrate Thickness Measurement Unit>

[0242] The substrate thickness measuring unit (1570) measures the thickness of the circuit board (O) and outputs it to the control unit (1600). The control unit (1600) corrects the processing parameters of the laser irradiation unit (1100) based on the measured thickness information. The thickness may vary depending on the type and manufacturing process of the circuit board (O), and even for the same type of substrate, there may be variations in thickness due to manufacturing deviations. Since the light penetration depth and processing depth vary depending on the thickness, adaptive correction of the processing parameters is required.

[0243] In one embodiment, the substrate thickness measuring unit (1570) may be implemented by placing a spectral interferometer, an ultrasonic thickness gauge, or two distance sensors on the upper and lower surfaces of the circuit board (O) respectively to perform differential measurement. The measured thickness information is used as one of the input information for the parameter prediction member (1650a) of the AI ​​parameter control unit (1650) described later, and an optimal processing parameter adapted to the thickness is calculated.

[0244] The stage unit (1500) is mechanically and optically connected to the optical focusing unit (1200) to position the circuit board (O) on the optical axis of the optical focusing unit (1200), and is electrically connected to the processing sequence control unit (1610) and the AI ​​parameter control unit (1650) of the control unit (1600) to transmit and receive position control signals and parameter correction information.

[0245] Control Unit

[0246] The control unit (1600) is connected to the laser irradiation unit (1100), optical focusing unit (1200), vision unit (1300), lighting unit (1400), and stage unit (1500), respectively, and identifies circuit pattern defects based on fluorescent images and reflection images acquired from the vision unit (1300), and controls the laser irradiation unit (1100) and stage unit (1500) to perform a processing sequence to remove circuit pattern defects. The control unit (1600) acts as the brain of the device of the present invention and integrally performs automatic detection of defects, adaptive determination of processing parameters, automatic evaluation of results after processing, and closed-loop reprocessing.

[0247] The control unit (1600) includes a processing sequence control unit (1610), a synchronization control unit (1620), an AI vision control unit (1640), an AI parameter control unit (1650), an environment monitoring unit (1670), a process history storage unit (1680), an interlock safety unit (1690), a substrate identification unit (1695), and a beam alignment correction unit (1696).

[0248] Below, the processing sequence control unit (1610), synchronization control unit (1620), environment monitoring unit (1670), process history storage unit (1680), interlock safety unit (1690), substrate identification unit (1695), and beam alignment correction unit (1696) will be described first, and the AI ​​vision control unit (1640) and AI parameter control unit (1650) will be described later in a separate section because their configurations are complex.

[0249] Processing Sequence Control Unit

[0250] The processing sequence control unit (1610) performs a closed-loop processing sequence of shooting-processing-shooting-reprocessing, which is the core operation of the present invention. That is, the processing sequence control unit (1610) causes the vision unit (1300) to acquire fluorescent images and reflection images, and then performs processing by the laser irradiation unit (1100) based on the location and shape of the circuit pattern defect identified by the image processing unit (1330) or the AI ​​vision control unit (1640). After performing processing, it acquires fluorescent images and reflection images again to determine whether the circuit pattern defect remains, and if it is determined that the circuit pattern defect remains, it repeats the processing.

[0251] The processing sequence control unit (1610) includes a closed loop rigger member (1610a), a residual defect determination member (1610b), a repetition count upper limit setting member (1610c), and a sequence termination condition determination member (1610d).

[0252] The closed loop rigger member (1610a) triggers a reprocessing sequence depending on whether there are residual circuit pattern defects. When the closed loop rigger member (1610a) receives residual defect information from the residual defect determination member (1610b) described later or the reprocessing sequence member (1640j) of the AI ​​vision control unit (1640), it immediately generates a trigger signal to start the next processing cycle and transmits it to the laser irradiation unit (1100), lighting unit (1400), and stage unit (1500).

[0253] The residual defect determination unit (1610b) determines whether a circuit pattern defect remains based on the fluorescent image and reflection image acquired after processing. In one embodiment, the residual defect determination unit (1610b) receives a defect detection result from the rule-based defect area extraction unit (1330c) of the image processing unit (1330) or the processing completion determination unit (1640i) of the AI ​​vision control unit (1640), and outputs a binary determination result (remaining / removal) depending on whether a defect is detected.

[0254] The repetition count upper limit setting member (1610c) sets an upper limit value for the number of repetitions of the closed-loop processing sequence. In principle, the closed-loop processing sequence of the present invention is repeated until the defect is completely removed, but an upper limit value is required to prevent infinite repetition when the defect is essentially impossible to remove (e.g., when the circuit pattern is damaged and cannot be normalized). In one embodiment, the repetition count upper limit setting member (1610c) can set an upper limit value in the range of 5 to 100 times, and an appropriate value is selected according to the type of circuit pattern defect and the material of the circuit board (O). According to the inventor's review, the closed-loop processing sequence of the present invention can typically completely remove the defect with tens of repetitions.

[0255] The sequence termination condition determination unit (1610d) determines the termination of the closed-loop processing sequence when no circuit pattern defects remain or when the number of repetitions reaches an upper limit. When termination is determined by the sequence termination condition determination unit (1610d), the processing sequence control unit (1610) does not perform further processing, performs post-processing such as generating a diagnostic report by the comprehensive diagnostic unit (1640o) and saving processing history by the process history storage unit (1680), and then generates a signal to unload the circuit board (O).

[0256] Synchronization Control Unit

[0257] The synchronization control unit (1620) aligns the operation triggers of the laser irradiation unit (1100), vision unit (1300), and lighting unit (1400) on the time axis. In the present invention, since various operations such as time-division acquisition of fluorescent images and reflection images, pulse emission of processing laser light, and change of RF signal of the AOD unit must be performed with precise time synchronization, integrated time synchronization by the synchronization control unit (1620) is essential.

[0258] The synchronization control unit (1620) includes a master clock member (1620a) and a trigger distribution member (1620b).

[0259] The master clock element (1620a) generates a reference clock signal. In one embodiment, the master clock element (1620a) may be implemented as a crystal oscillator, an OCXO (Oven-Controlled Crystal Oscillator), or an atomic clock and provides a reference time for the entire system. In one embodiment, the master clock element (1620a) outputs a high-frequency reference signal in the range of 10 MHz to 100 MHz to perform trigger distribution with nanosecond (ns) level time resolution.

[0260] The trigger distribution member (1620b) distributes a reference clock signal generated from the master clock member (1620a) as a trigger signal to each of the laser irradiation unit (1100), vision unit (1300), and lighting unit (1400). In one embodiment, the trigger distribution member (1620b) is implemented as an FPGA or a dedicated trigger distributor IC and outputs a synchronized trigger pulse by applying a preset delay and division ratio for each unit. By the trigger distribution member (1620b), the pulse firing timing of the laser irradiation unit (1100), the RF change timing of the AOD unit (1130), the lighting switching timing of the lighting controller (1440), and the exposure timing of the camera unit (1310) are all synchronized to the reference time of the master clock member (1620a).

[0261] Environmental Monitoring Department

[0262] The environment monitoring unit (1670) monitors at least one of the temperature, humidity, and cleanliness of the environment in which the PCB and RDL circuit laser repair device (1000) is installed in real time, and outputs the monitored environment information to the environment correction unit (1650p). Since precision optical processing equipment such as the present invention may experience thermal deformation of the optical system or fluctuation in laser output due to temperature changes in the working environment, condensation may occur on the optical member due to humidity changes, and optical performance may deteriorate due to dust adhering to the optical member in environments with low clean class, environmental monitoring by the environment monitoring unit (1670) is essential for ensuring long-term stability of processing quality.

[0263] In one embodiment, the environment monitoring unit (1670) includes a temperature sensor of the RTD (Resistance Temperature Detector) or thermocouple type, a capacitive humidity sensor, and a light scattering type particle counter, and continuously measures the temperature (e.g., range of 20°C to 25°C), relative humidity (e.g., range of 40% to 60%) and cleanliness (e.g., ISO Class 5 to 8 level) of the working environment, and transmits the measured values ​​to the environment correction member (1650p) of the AI ​​parameter control unit (1650) described later.

[0264] Process History Storage Unit

[0265] The process history storage unit (1680) stores the execution history of the processing sequence, processing parameters, the distribution of circuit pattern defects, and the judgment results of the processing completion judgment unit (1640i) for each circuit board (O) in a database, and provides the stored information as learning data in response to requests from the online learning unit (1640n) of the AI ​​vision control unit (1640) and the batch learning unit (1650t) of the AI ​​parameter control unit (1650). The process history storage unit (1680) serves as the core data infrastructure for the AI-based learning function of the present invention, enabling the continuous improvement of the performance of the AI ​​model by utilizing accumulated processing data.

[0266] In one embodiment, the process history storage unit (1680) may be implemented as a relational database (RDBMS), a NoSQL database, or a time-series database, and systematically manages processing history by day, week, month, and year. The information stored may include identification information of the circuit board (O) (obtained from the board identification unit (1695)), processing time and date, location, type, and size of the processed defect, processing parameters applied to each defect, processing results (complete removal / residual), processing trace analysis results, etc.

[0267] Interlock Safety Department

[0268] The interlock safety unit (1690) detects the door opening / closing status, emergency stop signal, and worker intrusion of the PCB and RDL circuit laser repair device (1000), and stops the operation of the laser irradiation unit (1100) when an abnormal state is detected. Since the present invention uses a high-power ps / fs pulse laser, ensuring worker safety is very important, and laser safety standards such as the Industrial Safety and Health Act and IEC 60825-1 must be complied with.

[0269] In one embodiment, the interlock safety unit (1690) is electrically connected to safety sensors such as a micro switch for detecting door opening / closing, an emergency stop button, a light curtain, or a safety mat, and if any of the safety sensors detect an abnormal state, it immediately outputs a stop signal to the processing laser unit (1110) and the shutter unit (1160) of the laser irradiation unit (1100). Upon receiving the stop signal, the shutter driving member (1160b) of the shutter unit (1160) moves the beam blocking member (1160a) into the optical path to immediately block the processing laser light, and at the same time, the pulse control member (1110b) of the processing laser unit (1110) stops the pulse oscillation.

[0270] <Board Identification Unit>

[0271] The substrate identification unit (1695) recognizes an identification code attached to the circuit board (O) to obtain identification information for the circuit board (O), and based on the identification information, calls a processing recipe corresponding to the circuit board (O). In one embodiment, a barcode, a QR code, or an RFID tag may be attached to the circuit board (O), and the substrate identification unit (1695) includes an optical scanner or an RFID reader for recognizing the corresponding code.

[0272] In one embodiment, the identification information obtained from the substrate identification unit (1695) may include the product number, lot number, manufacturing date, circuit design ID, etc. of the circuit board (O), and a processing recipe corresponding to the circuit board (O) is called from a database within the control unit (1600) or an external MES (Manufacturing Execution System) using the identification information as a key. The processing recipe includes the type, material, circuit pattern design information, recommended processing parameters, etc. of the circuit board (O), and a processing sequence is automatically set based on the called recipe.

[0273] Beam Alignment Correction Unit

[0274] The beam alignment correction unit (1696) periodically inspects the alignment status of the processing laser light on the optical path and, if misalignment of the optical path is detected, corrects the position or angle of the optical component on the optical path so that the processing laser light is accurately irradiated at the intended position on the circuit board (O). Since the optical path of the optical system may gradually become misaligned due to micro-movement, vibration, thermal deformation, etc. of the optical component resulting from long-term use, periodic correction by the beam alignment correction unit (1696) is essential for ensuring long-term stability of the processing precision of the present invention.

[0275] In one embodiment, the beam alignment correction unit (1696) irradiates a processing laser beam onto a specific reference position (e.g., the center of an alignment mark) on a circuit board (O), then verifies the actual irradiation position through the vision unit (1300) and calculates the error from the intended position. If the calculated error exceeds a preset allowable range (e.g., ±1 μm), the beam alignment correction unit (1696) outputs a correction signal to the mirror holder member (1140b) of the DM unit (1140), the galvanometer driving member (1150c) of the galvanometer auxiliary unit (1150), or the AOD correction member (1130e) of the AOD unit (1130) to automatically correct the optical path.

[0276] <AI비전제어부>

[0277] The AI ​​vision control unit (1640) takes fluorescent images and reflection images acquired from the vision unit (1300) as input to identify the location, shape, and type of circuit pattern defects, and determines whether to perform processing on the identified circuit pattern defects and the result after processing. The AI ​​vision control unit (1640) performs defect detection based on a pre-trained neural network model in parallel or independently, separate from the rule-based defect detection of the image processing unit (1330) of the present invention, thereby providing the ability to detect complex defect patterns or new types of defects that are difficult to detect with the image processing unit (1330) alone.

[0278] The AI ​​vision control unit (1640) includes an image preprocessing member (1640a), a multimodal fusion member (1640b), a defect pattern recognition member (1640c), a neural network defect classification member (1640d), a segmentation member (1640e), a defect importance evaluation member (1640f), a processing priority setting member (1640g), a processing difficulty prediction member (1640h), a processing completion determination member (1640i), a reprocessing sequence member (1640j), a processing trace analysis member (1640k), a model selection member (1640l), a transfer learning member (1640m), an online learning member (1640n), and a comprehensive diagnosis member (1640o).

[0279] <Lack of Image Preprocessing>

[0280] The image preprocessing unit (1640a) performs noise removal, contrast correction, and pixel matching on the fluorescent image and the reflection image to align the fluorescent image and the reflection image into the same coordinate system. In one embodiment, the image preprocessing unit (1640a) sequentially applies noise removal by a Gaussian filter, a median filter, or a bilateral filter, contrast correction by histogram equalization or adaptive histogram equalization, and pixel matching by phase correlation or feature point matching. The processing result of the image preprocessing unit (1640a) is transmitted as an input to the multimodal fusion unit (1640b) described later.

[0281] <Multimodal Fusion Absence>

[0282] The multimodal fusion member (1640b) combines the fluorescent image and the reflection image aligned in the image preprocessing member (1640a) using a channel-wise fusion method or a late fusion method to generate a single multimodal input data. The channel-wise fusion method is a method of combining the fluorescent image and the reflection image into a multi-channel image by considering each as a single channel, and the late fusion method is a method of processing the two images with separate neural networks and then integrating their output features. The channel-wise fusion method has the advantage of a simple structure and excellent learning efficiency, while the late fusion method has the advantage of enabling the complementary utilization of features extracted from the two images. In one embodiment, the present invention may use the two methods selectively or in combination depending on the characteristics of the application environment and the circuit board (O) to be processed.

[0283] <Lack of defect pattern recognition>

[0284] The defect pattern recognition member (1640c) extracts the location and shape of the area where circuit pattern defects exist by a neural network pre-trained from multimodal input data generated from the multimodal fusion member (1640b).

[0285] The defect pattern recognition member (1640c) includes a feature extraction body (1640ca), a candidate region detection body (1640cb), and a region determination body (1640cc).

[0286] The feature extractor (1640ca) extracts a feature map from multimodal input data by a pre-trained Convolutional Neural Network (CNN) that includes a plurality of convolution layers. In one embodiment, the feature extractor (1640ca) may adopt a CNN architecture such as ResNet, VGGNet, DenseNet, or EfficientNet as a backbone.

[0287] The candidate region detector (1640cb) detects candidate regions of circuit pattern defects from the feature map extracted from the feature extractor (1640ca). In one embodiment, the candidate region detector (1640cb) adopts a Region Proposal Network (RPN) or an anchor-free detection method to efficiently generate candidate regions of various sizes.

[0288] The region determination body (1640cc) determines whether a circuit pattern defect exists for each candidate region detected by the candidate region detection body (1640cb) and outputs the location coordinates of the circuit pattern defect. In one embodiment, the region determination body (1640cc) applies an additional convolution or fully-connected layer to the feature map of the candidate region to output a defect probability score and location coordinates (e.g., the center coordinates and size of the bounding box).

[0289] <Lack of Neural Network Defect Classification>

[0290] The neural network defect classification member (1640d) classifies a circuit pattern defect extracted from the defect pattern recognition member (1640c) into one of a short defect, a protrusion defect, and a residue defect by a pre-trained neural network. In one embodiment, the neural network defect classification member (1640d) takes the output (defect location and defect area feature map) of the defect pattern recognition member (1640c) as input and determines the type of defect by a multi-class classification neural network.

[0291] The neural network defect classification unit (1640d) can operate in parallel with the rule-based defect classification unit (1330d) of the image processing unit (1330) to mutually verify the defect classification results. If the two classification results match, it means that the classification reliability is very high, and if they do not match, it means that the characteristics of the defect are not typical or there is a possibility that it is a new type of defect, so additional review by an operator or a separate processing procedure may be applied.

[0292] <Absence of Segmentation>

[0293] The segmentation member (1640e) divides the boundaries of circuit pattern defects extracted from the defect pattern recognition member (1640c) into pixel units and sets them as a processing area (ROI: Region Of Interest).

[0294] The segmentation member (1640e) includes a pixel labeling body (1640ea) and a boundary calculation body (1640eb).

[0295] The pixel labeling body (1640ea) uses a pre-trained semantic segmentation neural network with an encoder-decoder structure to label each pixel of multimodal input data as corresponding to a circuit pattern defect. In one embodiment, the pixel labeling body (1640ea) may adopt a semantic segmentation neural network architecture such as U-Net, DeepLab, SegNet, or Mask R-CNN.

[0296] The boundary calculation body (1640eb) calculates the boundary coordinates of the processing area based on the results labeled by the pixel labeling body (1640ea). In one embodiment, the boundary calculation body (1640eb) expresses the boundary of the processing area as polygonal or curved coordinates by a contour tracing algorithm or a polygon approximation algorithm of the labeled pixel group, and transmits the processing area information to the AOD scan pattern control member (1650c) described later.

[0297] <Lack of Defect Importance Assessment>

[0298] The defect importance evaluation member (1640f) calculates a risk score by quantifying the degree of influence of circuit pattern defects on the circuit function of the circuit board (O) based on the type of circuit pattern defect classified in the neural network defect classification member (1640d) and the size and location of the processing area set in the segmentation member (1640e).

[0299] In one embodiment, the defect importance evaluation member (1640f) calculates a risk score by combining the weights for each type of defect (short defects have the highest weight as they directly cause a circuit short, protruding defects have a medium weight, and residue defects have a relatively low weight), the size of the defect (area or length of the processing area), and the location of the defect (e.g., defects adjacent to the critical path of the circuit pattern have an additional weight). The risk score is expressed as a quantitative value in the range of, for example, 0 to 100, and is used as an automatic switching standard for the processing priority setting member (1640g) and the lens exchange member (1210b) of the optical focusing unit (1200) described later.

[0300] <Absence of Processing Priority Setting>

[0301] When multiple circuit pattern defects are identified on a circuit board (O), the processing priority setting member (1640g) determines the processing order of multiple circuit pattern defects based on the risk score calculated by the defect importance evaluation member (1640f). In one embodiment, the processing priority setting member (1640g) configures a priority queue to prioritize the processing of defects with high risk scores, and simultaneously cooperates with the processing path optimization member (1650f) of the AI ​​parameter control unit (1650) described later to ensure that the processing order also considers the minimization of the travel distance between defects.

[0302] <Lack of prediction of processing difficulty>

[0303] The processing difficulty prediction member (1640h) predicts the processing difficulty in advance based on the type and shape of the circuit pattern defect and the material information of the circuit board (O). In one embodiment, the processing difficulty prediction member (1640h) outputs the processing difficulty as a score in the range of 0 to 100 by a pre-trained regression model, and the output difficulty score is used as one of the input information of the parameter prediction member (1650a) of the AI ​​parameter control unit (1650) and is reflected in the determination of processing parameters. Specifically, the regression model of the processing difficulty prediction member (1640h) is pre-trained using processing history data accumulated for the same type of defect as training data, with input features such as the size of the defect, the fineness of the circuit pattern where the defect is located (e.g., line-space (L / S) size), the distance from the defect to an adjacent pattern, and a processing threshold value according to the material of the circuit board (O). As the processing difficulty increases, a greater number of pulses, more precise AOD control, and stronger safety margin settings are applied. Additionally, in one embodiment, the output of the processing difficulty prediction member (1640h) is reflected in the automatic switching decision of the lens exchange member (1210b) of the optical focusing unit (1200), so that high-difficulty defects can be processed after being automatically switched to a high-magnification objective lens.

[0304] Processing Completion Determination Incomplete

[0305] The processing completion determination member (1640i) compares and analyzes the fluorescent image and reflection image before processing and the fluorescent image and reflection image after processing to determine whether circuit pattern defects remain in the processing area.

[0306] The processing completion judgment component (1640i) includes a difference image output component (1640ia), a threshold judgment component (1640ib), and a judgment result output component (1640ic).

[0307] The difference image output unit (1640ia) generates a difference image by calculating the difference between the fluorescent image and reflection image before processing and the fluorescent image and reflection image after processing in pixel units. In one embodiment, the difference image output unit (1640ia) applies an absolute difference or a normalized difference to each pair of images before and after processing to highlight the pixel area changed by processing.

[0308] The threshold determination body (1640ib) determines whether there exists an area where the intensity of the difference image generated by the difference image output body (1640ia) exceeds a preset threshold. In one embodiment, the threshold determination body (1640ib) uses a threshold that is adaptively set according to the type of circuit board (O) and the type of defect to distinguish between normal changes caused by processing (e.g., changes in brightness due to metal removal in the processed area) and residual defects.

[0309] The judgment result output body (1640ic) outputs a final judgment result regarding whether processing is completed based on the judgment result of the threshold judgment body (1640ib). The output judgment result is transmitted to the residual defect judgment member (1610b) and sequence termination condition judgment member (1610d) of the processing sequence control unit (1610), and to the reprocessing sequence member (1640j) described later.

[0310] <Absence of reprocessing sequence>

[0311] When the reprocessing sequence member (1640j) determines that a circuit pattern defect remains in the processing completion judgment member (1640i), it automatically triggers a reprocessing sequence for the remaining circuit pattern defect and transmits a reprocessing command to the closed loop rigger member (1610a) of the processing sequence control unit (1610). The reprocessing triggered by the reprocessing sequence member (1640j) is not simply repeated with the same processing parameters, but is performed with new processing parameters that are adaptively adjusted by reflecting information on the remaining defect by the adaptive control member (1650l) of the AI ​​parameter control unit (1650) described later.

[0312] <Absence of processing trace analysis>

[0313] The processing trace analysis member (1640k) analyzes the surface condition around the processing area from fluorescent images and reflection images after processing is performed to detect damage or residues that have occurred on the substrate portion of the circuit board (O). Non-thermal ablation by the ps / fs pulse laser of the present invention does not, in principle, cause damage to the substrate portion, but if excessive pulse energy or inappropriate processing parameters are applied, there is a possibility that minor damage may occur to the substrate portion. The processing trace analysis member (1640k) detects such processing traces and transmits the analysis results to the safety control member (1650m) and comprehensive diagnostic member (1640o) described later.

[0314] <Lack of Model Selection>

[0315] The model selection member (1640l) receives information regarding the type of circuit board (O), the material of the circuit board (O), and the thickness of the metal pattern formed on the circuit board (O), selects a model corresponding to the input information from among a plurality of previously stored pre-trained models, and applies it to the defect pattern recognition member (1640c) and the neural network defect classification member (1640d).

[0316] Since the characteristics of the fluorescence / reflection image vary significantly depending on the type of circuit board (O) (e.g., general PCB, FCBGA, glass interposer, organic interposer), material, and metal pattern thickness, it is difficult to accurately detect defects on all types of boards with a single pre-trained model. The model selection member (1640l) is a type of model library manager that automatically selects the model most suitable for the current processing target from among multiple models optimized for each type of board.

[0317] <Lack of Transfer Learning>

[0318] The transfer learning member (1640m) fine-tunes the parameters of a pre-trained model selected from the model selection member (1640l) to match the process conditions of the circuit board (O) currently being worked on. In one embodiment, the transfer learning member (1640m) may adopt a fine-tuning method that retrains only the weights of the last few layers of the pre-trained model using a small amount of data acquired from the current circuit board (O). Through transfer learning, the existing model can be rapidly adapted without the need to train the model from scratch for a new type of circuit board (O), thereby shortening the initial setup time when applying the present invention to mass production.

[0319] <Lack of Online Learning>

[0320] The online learning member (1640n) accumulates the fluorescent images, reflection images, and judgment results of the processing completion judgment member (1640i) collected during the processing sequence as learning data, and updates the pre-trained model in real time. In one embodiment, the online learning member (1640n) accumulates data in mini-batch units and updates the model parameters at a certain period (e.g., once after processing 100 sheets) or when a certain amount of data is reached. The updated model is immediately applied to the defect pattern recognition member (1640c) and the neural network defect classification member (1640d), and is gradually optimized for the data characteristics of the current mass production line.

[0321] <Lack of Comprehensive Diagnosis>

[0322] The comprehensive diagnostic component (1640o) generates a diagnostic report for the circuit board (O) by synthesizing the distribution of multiple circuit pattern defects identified on the circuit board (O) after the completion of the processing sequence, the judgment result of the processing completion judgment component (1640i), and the analysis result of the processing trace analysis component (1640k). The diagnostic report is generated for each circuit board (O), stored in the process history storage unit (1680), and utilized for operator review or mass production quality control. In one embodiment, the diagnostic report may include a map of the distribution of defect locations and types, a processing success rate, the degree of processing traces, the distribution of residual defects, recommended follow-up processing, etc.

[0323] <AI파라미터제어부>

[0324] The AI ​​parameter control unit (1650) adaptively determines the processing parameters of the laser irradiation unit (1100) and the AOD unit (1130) based on the location, shape, and type of circuit pattern defects identified by the AI ​​vision control unit (1640), and dynamically readjusts the processing parameters according to the results after processing. The AI ​​parameter control unit (1650) is the core of the closed-loop adaptive control of the present invention and learns the characteristics of the defects and feedback on the processing results to automatically determine the optimal processing parameters for each processing cycle.

[0325] The AI ​​parameter control unit (1650) includes a parameter prediction member (1650a), a laser parameter control member (1650b), an AOD scan pattern control member (1650c), a beam shape control member (1650d), a pulse timing control member (1650e), a processing path optimization member (1650f), a reinforcement learning member (1650h), a Bayesian optimization member (1650i), a model prediction control member (1650j), a hybrid control member (1650k), an adaptive control member (1650l), a safety control member (1650m), a multi-purpose optimization member (1650n), an energy distribution optimization member (1650o), an environment correction member (1650p), an output correction member (1650q), a digital twin member (1650r), and a parameter database member (1650s). It includes a batch learning member (1650t), a quality-based learning member (1650u), and a federation learning member (1650v).

[0326] <Lack of Parameter Prediction>

[0327] The parameter prediction member (1650a) takes as input information regarding the type, shape, and location of circuit pattern defects, the material of the circuit board (O), and the thickness of the metal pattern formed on the circuit board (O), and outputs processing parameters of the laser irradiation unit (1100) and scan parameters of the AOD unit (1130) by a pre-trained deep neural network (DNN) model.

[0328] The parameter prediction member (1650a) includes an input transformation body (1650aa), a prediction output body (1650ab), and a constraint verification body (1650ac).

[0329] The input converter (1650aa) converts information about circuit pattern defects and information about the circuit board (O) into an input vector. In one embodiment, the input converter (1650aa) normalizes information such as the type of defect (one-hot encoded categorical variable), the size of the defect (continuous variable), the location coordinates of the defect (continuous variable), the material of the circuit board (O) (one-hot encoded), the metal pattern thickness (continuous variable), and environmental conditions (temperature, humidity) to convert it into a fixed-dimensional input vector.

[0330] The prediction output body (1650ab) inputs an input vector into a pre-trained deep neural network model and outputs predicted values ​​of processing parameters and scan parameters. In one embodiment, the deep neural network model of the prediction output body (1650ab) may be implemented as a Multi-Layer Perceptron (MLP) or Transformer-based regression model and outputs parameter values ​​such as pulse width, pulse repetition frequency, pulse energy, number of pulses, AOD scan trajectory, scan speed, scan area, and scan pattern.

[0331] The constraint verification body (1650ac) determines the final parameter value by verifying whether the predicted value output from the prediction output body (1650ab) is within a physically feasible range. That is, the constraint verification body (1650ac) checks whether the predicted value exceeds the operating range of the laser irradiation unit (1100) or the AOD unit (1130), whether it exceeds the damage threshold of the circuit board (O), and whether the processing time is within an allowable range. If there is a violation, it determines the final parameter value by clipping the predicted value to the nearest value within the allowable range or by applying a safety margin.

[0332] Laser Parameter Control Member

[0333] The laser parameter control member (1650b) variably controls the pulse width, pulse repetition frequency, pulse energy, and number of pulses of the processing laser unit (1110) of the laser irradiation unit (1100) based on the result output from the parameter prediction member (1650a). In one embodiment, the laser parameter control member (1650b) transmits the determined parameter value as a digital or analog signal to the pulse control member (1110b) of the processing laser unit (1110), so that the pulse control member (1110b) sets the oscillation condition of the laser light source member (1110a) to the corresponding value.

[0334] <AOD스캔패턴제어부재>

[0335] The AOD scan pattern control member (1650c) variably controls the scan trajectory, scan speed, scan area, and scan pattern of the AOD unit (1130) based on the result output from the parameter prediction member (1650a). In one embodiment, the AOD scan pattern control member (1650c) selects a suitable scan pattern (point scan, line scan, raster scan, concentric circle scan, etc.) according to the shape of the defect (point shape, line shape, surface shape) and outputs an RF frequency sequence corresponding to the selected pattern to the RF driving member (1130c) of the AOD unit (1130).

[0336] Beam shape control member

[0337] The beam shape control member (1650d) controls the beam diameter and beam profile of the processing laser light by variably controlling the beam magnification ratio of the beam expander unit (1120) based on the result output from the parameter prediction member (1650a). In one embodiment, the beam shape control member (1650d) controls the position of the second lens member (1120b) of the beam expander unit (1120) in the direction of the optical axis to vary the beam diameter and divergence angle.

[0338] Pulse Timing Control Member

[0339] The pulse timing control member (1650e) synchronizes the pulse emission timing of the processing laser unit (1110) with the acoustic frequency change timing of the AOD unit (1130) to control the processing laser light so that it is accurately irradiated at the intended location. Since it takes a settling time of several microseconds for the actual acoustic wave to form in a normal state inside the crystal after the RF frequency of the AOD unit (1130) is changed, precise timing control is required so that the pulse emission occurs after this settling time. The pulse timing control member (1650e) is synchronized with the reference signal of the master clock member (1620a) of the synchronization control unit (1620) to align the pulse emission timing and the RF change timing with nanosecond (ns) level precision.

[0340] Processing Path Optimization Member

[0341] The processing path optimization member (1650f) receives the location coordinates of a plurality of circuit pattern defects identified on the circuit board (O) and calculates a processing order and a processing path such that the total travel distance of the processing path between the plurality of circuit pattern defects is minimized. In one embodiment, the processing path optimization member (1650f) calculates the optimal processing order by dynamic programming, branch and bound, genetic algorithm, or history-based heuristic, which are solutions to the Traveling Salesman Problem (TSP). By minimizing the total travel distance of the processing path, the travel time of the stage unit (1500) is shortened, and the total processing time is significantly reduced.

[0342] <Lack of reinforcement learning>

[0343] The reinforcement learning component (1650h) learns the correlation between the processing parameters applied during the processing of circuit pattern defects and the result after processing based on reinforcement learning, and updates the policy of the processing parameters to be applied in the next processing sequence.

[0344] The reinforcement learning body (1650h) includes a state observer (1650ha), a behavior selector (1650hb), and a policy updater (1650hc).

[0345] A state observer (1650ha) observes the current state of a processing sequence and generates a state vector. In one embodiment, the state vector includes defect information (type, size, location), the result of the previous processing (complete removal / residual), the degree of trace after processing, environmental conditions (temperature, humidity), the current processing cycle number, etc.

[0346] The action selector (1650hb) selects a processing parameter to be applied as an action based on the state vector. In one embodiment, the action selector (1650hb) selects an optimal action according to the currently learned policy or explores a new action with a certain probability by means of an ε-greedy policy or a softmax policy.

[0347] The policy updater (1650hc) updates the policy using the degree of improvement in processing quality measured after the performance of an action and the degree of damage to the substrate portion of the circuit board (O) as a reward function. In one embodiment, the policy updater (1650hc) may adopt a reinforcement learning algorithm such as Q-learning, Actor-Critic, Proximal Policy Optimization (PPO), or Deep Deterministic Policy Gradient (DDPG).

[0348] Absence of Bayesian Optimization

[0349] When there is a lack of collected data for a new type of circuit board (O), the Bayesian optimization member (1650i) uses a Bayesian optimization algorithm to search for the optimal value of the processing parameter with a small number of trials. In one embodiment, the Bayesian optimization member (1650i) uses a Gaussian process as a surrogate model to efficiently search for the optimal value in the parameter space, and can converge to the optimal solution with a much smaller number of trials compared to a conventional grid search or random search.

[0350] <Lack of Model Predictive Control>

[0351] The model predictive control member (1650j) determines the machining parameters to be applied in the next iteration step of the closed-loop machining sequence based on a model predictive control (MPC) algorithm. In one embodiment, the model predictive control member (1650j) predicts the machining results of N future cycles based on a dynamic model of the machining system and optimizes the parameters of the current cycle so that the predicted results satisfy the target specifications.

[0352] Hybrid Control Member

[0353] The hybrid control member (1650k) secures control stability of the processing parameters by generating a final control value through weighted combination of the output of the parameter prediction member (1650a) and the output of the PID (Proportional-Integral-Derivative) control algorithm. In one embodiment, neural network-based prediction has the advantage of being able to learn complex non-linear relationships, but the prediction accuracy may be reduced for inputs not included in the training data; and while PID control has excellent stability and interpretability, it has the limitation of being limited in its response to non-linearity. Therefore, the hybrid control, which combines the two methods through weighted combination, can utilize the advantages of both.

[0354] <Lack of adaptive control>

[0355] The adaptive control member (1650l) receives residual defect information as feedback from the processing completion judgment member (1640i) of the AI ​​vision control unit (1640) as a result of the previous processing sequence, and adaptively corrects the processing parameters to be applied to the next processing sequence based on the feedback information.

[0356] The adaptive control member (1650l) includes a feedback receiving body (1650la), an adjustment amount calculation body (1650lb), and a parameter updating body (1650lc).

[0357] The feedback receiver (1650la) receives residual defect information from the processing completion determination member (1640i). The residual defect information includes the location, size, and residual amount (residual ratio compared to before processing) of the residual defect.

[0358] The adjustment amount calculator (1650lb) calculates the adjustment amount of the processing parameters by comparing and analyzing residual defect information and the processing parameters applied in the previous processing sequence. In one embodiment, the adjustment amount calculator (1650lb) calculates the adjustment amount by increasing the pulse energy or increasing the number of pulses when the residual amount is large, and applies only a change in the fine adjustment level when the residual amount is small.

[0359] The parameter update body (1650lc) applies the adjustment amount calculated by the adjustment amount calculation body (1650lb) to the laser parameter control member (1650b) and the AOD scan pattern control member (1650c) to update the processing parameters of the next processing sequence.

[0360] Safety Control Lack

[0361] The safety control member (1650m) dynamically limits the upper limit of the processing parameters so that the processing parameters of the processing laser unit (1110) do not exceed a threshold value that causes damage to the substrate portion of the circuit board (O). The safety control member (1650m) is one of the most critical protection functions of the present invention and performs final verification on the output of all other parameter determination members (parameter prediction member (1650a), reinforcement learning member (1650h), adaptive control member (1650l), etc.) to fundamentally prevent damage to the circuit board (O).

[0362] In one embodiment, the safety control member (1650m) maintains a database of damage threshold values ​​for each material of the circuit board (O), and when any combination of processing parameters is input, calculates the processing effect (e.g., absorbed energy, size of the heat-affected zone) predicted to occur due to the parameters and compares it with the damage threshold values. If the threshold values ​​are exceeded, the parameters are immediately rejected and replaced with the nearest safe parameter below the threshold values.

[0363] Multipurpose Optimization Material

[0364] The multi-objective optimization member (1650n) searches for a Pareto optimum for a multi-objective function that uses the speed of the processing sequence, processing quality, and the degree of damage to the substrate portion of the circuit board (O) as evaluation indicators. In the present invention, since it is required to simultaneously satisfy multiple goals that may conflict with each other, such as processing speed (throughput), processing quality (defect removal rate), and minimization of substrate damage, it is difficult to find an appropriate solution through single-objective optimization. In one embodiment, the multi-objective optimization member (1650n) derives various solution candidates on the Pareto frontier by a multi-objective evolutionary algorithm such as NSGA-II (Non-dominated Sorting Genetic Algorithm II) and selects one of them according to the operator's requirements.

[0365] Energy Distribution Optimization Material

[0366] The energy distribution optimization member (1650o) learns the energy distribution formed on the circuit board (O) according to the scan pattern of the AOD unit (1130) and provides a correction value to the AOD scan pattern control member (1650c) so that a uniform energy distribution is formed within the processing area. The energy reaching each pixel within the processing area by the AOD scan may vary depending on the scan speed, pulse repetition frequency, overlap rate of the scan pattern, etc., and the energy distribution optimization member (1650o) compensates for these variations to ensure uniform processing quality throughout the entire processing area. Specifically, the energy distribution optimization member (1650o) simulates the pulse arrival time and the accumulated pulse energy amount at each scan position to calculate the accumulated energy distribution per unit area within the processing area, and uses the non-uniformity of the calculated energy distribution (e.g., distribution variance, difference between maximum and minimum values) as a quantitative evaluation indicator. If the evaluation indicator exceeds a preset allowable range, the energy distribution optimization member (1650o) calculates a correction value, such as variable scan speed, relocation of the scan path, or positional adjustment of pulse energy, and provides it to the AOD scan pattern control member (1650c). Through this uniformization of energy distribution, the problem of some pixels within the processing area being over-ablated or under-ablated is prevented, thereby improving the uniformity of processing quality.

[0367] Environmental Protection

[0368] The environment correction member (1650p) receives the temperature and humidity of the working environment, calculates an environment correction value for the processing parameters, and applies it to the laser parameter control member (1650b) and the AOD scan pattern control member (1650c). In one embodiment, the environment correction member (1650p) receives real-time environment information from the environment monitoring unit (1670) and calculates a correction value using a pre-learned environment-parameter correction model.

[0369] Output correction component

[0370] The output correction member (1650q) measures the change in the output light quantity of the processing laser unit (1110) over time, calculates an output correction value to compensate for the measured change in light quantity, and applies it to the laser parameter control member (1650b). In one embodiment, the output correction member (1650q) receives a real-time light quantity measurement value from the power monitor member (1110c) of the processing laser unit (1110), and corrects the pulse control parameter when the light quantity deviates from a set value by more than a certain range so that a constant processing energy is always delivered to the circuit board (O).

[0371] <Digital Twin Absence>

[0372] The digital twin component (1650r) generates a virtual model of the circuit board (O) and circuit pattern defects, and performs a processing simulation on the virtual model to pre-verify the processing parameters before actual processing. In one embodiment, the digital twin component (1650r) predicts the result of performing processing with any combination of processing parameters based on a 3D model of the circuit board (O) and a processing laser-metal interaction model. Through this, the suitability of the parameters is verified before actual processing, thereby preventing the discarding of the circuit board (O) due to trial and error.

[0373] <Absence of Parameter Database>

[0374] The parameter database member (1650s) stores optimal parameter information corresponding to a combination of the type of circuit board (O), defect type, and processing parameters in advance, and outputs the optimal parameter information upon the request of the parameter prediction member (1650a). In one embodiment, the parameter database member (1650s) is implemented as a lookup table or a relational database, enabling the rapid retrieval of optimal parameters based on accumulated processing experience. Specifically, the parameter database member (1650s) stores a set of optimal processing parameters (pulse width, pulse repetition frequency, pulse energy, number of pulses, AOD scan pattern, etc.) as values ​​using (circuit board type, defect type, defect size range, metal pattern thickness range) as a key, and the parameter prediction member (1650a) first queries the parameter database member (1650s) before performing neural network inference to determine processing parameters for a new defect. If processing history under similar conditions exists in the database, the corresponding optimal parameter can be applied as an initial value to accelerate neural network inference, or the response time can be shortened by omitting neural network inference and applying the database value as is. The parameter database unit (1650s) can be configured to be linked with the process history storage unit (1680) so that new processing results are automatically registered in the database.

[0375] <Lack of Placement Learning>

[0376] The batch learning member (1650t) analyzes processing result data collected in batch units consisting of a plurality of circuit boards (O) and derives an improved profile of processing parameters to be applied to the next batch. In one embodiment, the batch learning member (1650t) performs statistical analysis (mean, variance, trend) of the batch unit and derives an optimized parameter profile in the batch unit through the results of the statistical analysis and applies it to the next batch.

[0377] <Lack of Quality-Based Learning>

[0378] The quality-based learning member (1650u) updates the policy of the reinforcement learning member (1650h) and the weights of the parameter prediction member (1650a) using the processing quality evaluation result measured after the processing sequence is completed as a compensation function. In one embodiment, the processing quality evaluation is performed by receiving the result of a circuit function test (electrical test) in a post-process or the result of a precision inspection by external inspection equipment as input, and by utilizing this result as a learning signal, the AI ​​model of the present invention is trained to consider not only simple defect removal but also the quality of the final product.

[0379] <Lack of Federation Learning>

[0380] The federated learning unit (1650v) integrates learning data collected from each of the plurality of PCB and RDL circuit laser repair devices (1000) to update a pre-trained deep neural network model, and operates in a federated learning manner in which the raw data of each device is preserved within each device and only the update information of the model parameters is integrated.

[0381] Federation learning is a useful learning paradigm in environments where it is difficult to transmit raw data externally due to data privacy or trade secret concerns. It operates by transmitting only the weight update information (gradient or weight delta) of locally trained models from each device to a central server for aggregation, and then distributing the aggregated update information back to each device. Through this, the processing experience accumulated by multiple devices of the present invention at the mass production site is shared among all devices, while sensitive raw data, such as design information of each circuit board (O), is not leaked externally.

[0382] The AI ​​parameter control unit (1650) receives defect information from the AI ​​vision control unit (1640), outputs processing parameter control signals to the laser irradiation unit (1100) and the AOD unit (1130), and controls the entire closed-loop processing sequence of the present invention in cooperation with the processing sequence control unit (1610).

[0383] The control unit (1600) acts as the brain that integrates and controls all units of the device of the present invention. Specifically, the control unit (1600) operates in the following integrated control flow. First, the substrate identification unit (1695) calls the identification information and processing recipe of the circuit board (O) to set the initial conditions of the processing sequence. Subsequently, when the work safety is confirmed by the interlock safety unit (1690), the synchronization control unit (1620) sets the reference time for the entire system, and the environment monitoring unit (1670) measures the temperature, humidity, and cleanliness of the working environment and transmits them to the environment correction member (1650p) of the AI ​​parameter control unit (1650). When a processing cycle is initiated by the closed loop rigger member (1610a) of the processing sequence control unit (1610), the vision unit (1300) acquires a fluorescent image and a reflection image by time-division switching control of the lighting controller (1440), the AI ​​vision control unit (1640) analyzes the two images to identify circuit pattern defects, and the AI ​​parameter control unit (1650) determines the optimal processing parameters for the identified defects. According to the determined parameters, the laser irradiation unit (1100) irradiates processing laser light onto the circuit board (O) to remove defects, and after processing is performed, the processing completion judgment member (1640i) of the AI ​​vision control unit (1640) determines whether defects remain. If residual defects exist, the adaptive control member (1650l) of the AI ​​parameter control unit (1650) adaptively corrects the processing parameters to trigger a reprocessing cycle, and if there are no residual defects or the number of repetitions reaches an upper limit, the sequence termination condition determination member (1610d) determines the sequence termination, and the comprehensive diagnosis member (1640o) generates a diagnosis report and stores it in the process history storage unit (1680). In addition, the beam alignment correction unit (1696) periodically checks the optical path alignment status in the background of the processing sequence and performs automatic correction if an optical path misalignment is detected.By means of the integrated control operation of such control unit (1600), the PCB and RDL circuit laser repair device (1000) of the present invention automatically performs the entire cycle of defect detection, processing, judgment, and reprocessing on the circuit board (O) without human intervention.

[0385] Although embodiments of the present invention have been described in more detail with reference to the attached drawings, the present invention is not necessarily limited to these embodiments and may be modified in various ways within the scope of the technical spirit of the present invention. Accordingly, the embodiments disclosed in the present invention are intended to explain, not limit, the technical spirit of the present invention, and the scope of the technical spirit of the present invention is not limited by these embodiments. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. The scope of protection of the present invention shall be interpreted by the claims below, and all technical spirits within an equivalent scope shall be interpreted as being included within the scope of rights of the present invention.

[0386] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below. Explanation of the symbols

[0387] 1000: PCB and RDL Circuit Laser Repair Device 1100: Laser Irradiation Unit 1110: Processing Laser Section 1110a: Laser light source component 1110b: Pulse control member 1110c: Power monitor missing 1110d: Laser cooling element 1120: Beam Expander Section 1120a: First lens member 1120b: Second lens component 1130: AOD Department 1130a: X-axis AOD member 1130b: Y-axis AOD member 1130c: RF driving member 1130d: 1st-order diffraction selection element 1130e: AOD correction material 1140: DM Department 1140a: Dichroic mirror absence 1140b: Mirror holder absence 1150: Galvanobo Jobu 1150a: X-galvanometer mirror component 1150b: Y-galvanometer mirror component 1150c: Galvano driving member 1160: Shutter section 1160a: Beam blocking member 1160b: Shutter driving member 1200: Optical focusing unit 1210: Objective lens section 1210a: Objective lens component 1210b: Lens replacement part 1220: Tube lens part 1220a: Tube lens component 1230: BS Department 1230a: Beam splitter member 1230b: Polarizing beam splitter member 1300: Vision Unit 1310: Camera Department 1310a: Image sensor absence 1310b: Global shutter control component 1310c: Exposure time control element 1320: Fluorescent blocking filter section 1320a: Notch filter member 1320c: Filter holder component 1330: Image processing unit 1330a: Image matching component 1330b: Difference image generation component 1330c: Rule-based defect area extraction component 1330d: Absence of rule-based defect classification 1400: Lighting unit 1410: Fluorescent lighting section 1410a: NUV light source component 1410b: Polarizer component (p) 1420: Coaxial illumination unit 1420a: RB light source component 1420b: Polarizer component(s) 1420c: Beam alignment member 1430: Ring light section 1430a: Ring LED array component 1430b: Angle adjustment member 1440: Lighting Controller 1440a: Switching timing control element 1440b: Synchronous pulse generating element 1440c: Light intensity control element 1500: Stage Unit 1510: X-axis stage section 1510a: X-axis linear motor component 1510b: X-axis position sensor component 1520: Y-axis stage section 1520a: Y-axis linear motor component 1520b: Y-axis position sensor component 1530: Optical axis position adjustment unit 1530a: Optical axis driving member 1530b: Distance sensor missing 1540: Board fixing part 1540a: Vacuum chuck member 1550: Alignment Mark Recognition Unit 1550a: Mark imaging component 1550b: Coordinate alignment member 1560: Autofocus unit 1560a: Distance measuring member 1560b: Optical axis direction correction member 1570: Substrate thickness measuring unit 1600: Control Unit 1610: Processing Sequence Control Unit 1610a: Absence of closed loop tractor 1610b: Residual defect determination component 1610c: Absence of upper limit setting for repetitions 1610d: Sequence termination condition determination missing 1620: Synchronization control unit 1620a: Master clock missing 1620b: Lack of trigger distribution 1640: AI Vision Control Unit 1640a: Image preprocessing component 1640b: Multimodal Fusion Member 1640c: Lack of defect pattern recognition 1640ca: Feature extractor 1640cb: Candidate region detector 1640cc: Area determination body 1640d: Neural network defect classification missing 1640e: Absence of segmentation 1640ea: Pixel Labeling Font 1640eb: Boundary output 1640f: Lack of defect severity assessment 1640g: Processing priority setting component 1640h: Lack of processing difficulty prediction 1640i: Processing Completion Judgment Member 1640ia: Difference image output 1640ib: Threshold determination body 1640ic: Judgment result output body 1640j: Reprocessing sequence component 1640k: Machining trace analysis component 1640l: Lack of model selection 1640m: Lack of transfer learning 1640n: Lack of online learning 1640o: Lack of comprehensive diagnosis 1650: AI Parameter Control Unit 1650a: Lack of parameter prediction 1650aa: Input converter 1650ab: Predicted output 1650ac: Constraint Verification Body 1650b: Laser parameter control member 1650c: AOD scan pattern control component 1650d: Beam shape control member 1650e: Pulse timing control element 1650f: Machining path optimization component 1650h: Lack of reinforcement learning 1650ha: State Observer 1650hb: Action Selector 1650hc: Policy Updater 1650i: Bayesian Optimization Absence 1650j: Model predictive control absence 1650k: Hybrid control component 1650l: Adaptive control component 1650la: Feedback receiving body 1650lb: Adjustment amount calculation 1650lc: Parameter updater 1650m: Safety control component 1650n: Multipurpose Optimization Member 1650o: Energy distribution optimization component 1650p: Environmental protection component 1650q: Output correction component 1650r: Digital twin absence 1650s: Absence of parameter database 1650t: Lack of batch learning 1650u: Lack of quality-based learning 1650v: Lack of federation learning 1670: Environmental Monitoring Department 1680: Process History Storage Unit 1690: Interlock Safety Section 1695: Board Identification Unit 1696: Beam alignment correction unit O: Circuit board

Claims

Claim 1 A PCB and RDL circuit laser repair device for removing circuit pattern defects of a circuit board by laser, comprising: a laser irradiation unit that generates processing laser light for removing circuit pattern defects of the circuit board and irradiates it toward the circuit board; an optical focusing unit that focuses the processing laser light emitted from the laser irradiation unit onto the surface of the circuit board and receives reflected light and emitted light from the circuit board; a vision unit that acquires an image of the circuit board via the optical focusing unit; an illumination unit that selectively irradiates fluorescent illumination and reflected illumination onto the circuit board and acquires a fluorescent image and a reflected image of the circuit board through the vision unit; and a stage unit that supports the circuit board, moves the circuit board within a horizontal plane, and adjusts the position of the circuit board in the direction of the optical axis of the optical focusing unit. and a control unit each connected to the laser irradiation unit, the optical focusing unit, the vision unit, the illumination unit, and the stage unit, and identifying the circuit pattern defect based on the fluorescence image and the reflection image acquired from the vision unit, and controlling the laser irradiation unit and the stage unit to perform a processing sequence to remove the circuit pattern defect; wherein the laser irradiation unit comprises: a processing laser unit that generates pulsed laser light with a femtosecond (fs) or picosecond (ps) pulse width as the processing laser light; a beam expander unit that expands the beam diameter of the processing laser light generated from the processing laser unit; an AOD unit that receives the processing laser light emitted from the beam expander unit and changes the deflection direction of the processing laser light by an acousto-optical effect; and a DM unit having wavelength selectivity that reflects the processing laser light emitted from the AOD unit and transmits light in the visible light region. and a galvano-auxiliary unit disposed on the optical path between the AOD unit and the DM unit, which further changes the deflection direction of the processing laser light emitted from the AOD unit;The processing laser unit comprises: a laser light source member that emits pulsed laser light with a femtosecond (fs) or picosecond (ps) pulse width; and a pulse control member connected to the laser light source member and variablely controlling the pulse width, pulse repetition frequency, and pulse energy of the pulsed laser light; and the AOD unit comprises: an X-axis AOD member that deflects the processing laser light in the X-axis direction; a Y-axis AOD member that deflects the processing laser light in the Y-axis direction; and an RF driving member that controls the acoustic frequency of the X-axis AOD member and the Y-axis AOD member by applying a variable frequency RF signal to the X-axis AOD member and the Y-axis AOD member, respectively. and a first-order diffraction selection member that blocks zero-order diffracted light emitted from the X-axis AOD member and the Y-axis AOD member and passes only first-order diffracted light; wherein the processing laser light is configured to be deflected in a two-dimensional direction, and the galvano auxiliary member comprises: an X-galvano mirror member that deflects the processing laser light in the X-axis direction; and a Y-galvano mirror member that deflects the processing laser light in the Y-axis direction; and a galvanometer driving member that controls the rotation angle of the X galvanometer mirror member and the Y galvanometer mirror member; wherein the galvanometer auxiliary member is responsible for scanning a wide field of view area on the circuit board, and the AOD member is responsible for fine precision scanning within the wide field of view area, and the galvanometer auxiliary member and the AOD member are configured to cooperate in series to deflect the processing laser light; and the optical focusing unit comprises: an objective lens member disposed on the optical path between the DM member and the stage unit to focus the processing laser light onto the surface of the circuit board and receive the reflected light and the emitted light from the circuit board; a tube lens member that forms an image of the reflected light and the emitted light received from the objective lens member onto an image sensor of the camera member of the vision unit; and a BS member disposed on the optical path between the tube lens member and the objective lens member to branch or combine the optical path of the processing laser light and the optical path of the light incident on the vision unit.The vision unit comprises: a camera unit equipped with a single image sensor to acquire both the fluorescent image and the reflection image; a fluorescent blocking filter unit coupled to the incident end of the camera unit, which blocks the output wavelength of the fluorescent illumination of the illumination unit while allowing the wavelength of the fluorescent emission from the circuit board and the wavelength of the reflection illumination to pass through; and an image processing unit that combines or compares the fluorescent image and the reflection image acquired by the camera unit to identify the location and shape of the circuit pattern defect; wherein the camera unit comprises a global shutter control member that performs simultaneous exposure and simultaneous reading for all pixels of the single image sensor; and wherein the illumination unit comprises: a fluorescent illumination unit that irradiates light of an ultraviolet or near-ultraviolet wavelength onto the circuit board to induce fluorescent emission on a substrate portion of the circuit board; a coaxial illumination unit that irradiates light in the visible light region onto the circuit board in a direction coaxial with the optical axis of the optical focusing unit; and a ring illumination unit that irradiates light in the visible light region onto the circuit board from a lateral direction. A lighting controller electrically connected to each of the fluorescent lighting unit, the coaxial lighting unit, and the ring lighting unit, and performing time-division switching control such that at least one of the fluorescent lighting unit, the coaxial lighting unit, and the ring lighting unit is mutually exclusively turned ON / OFF, and outputting a trigger signal synchronized with the exposure timing of the camera unit;The lighting controller is configured such that the camera unit acquires the fluorescent image in the section where the fluorescent lighting unit is ON, and the camera unit acquires the reflection image in the section where the coaxial lighting unit or the ring lighting unit is ON, so that the single image sensor alternately acquires the fluorescent image and the reflection image in time division; the control unit is configured such that the vision unit acquires the fluorescent image and the reflection image, then performs processing by the laser irradiation unit based on the location and shape of the circuit pattern defect identified by the image processing unit, and after performing the processing, acquires the fluorescent image and the reflection image again to determine whether the circuit pattern defect remains, and if it is determined that the circuit pattern defect remains, the processing is repeated; the processing sequence control unit performs a closed-loop processing sequence of shooting-processing-shooting-reprocessing; the laser irradiation unit, the vision unit, and the lighting unit each have their operation triggers aligned on the time axis; and the fluorescent acquired by the vision unit An AI vision control unit that takes an image and the reflected image as inputs to identify the location, shape, and type of the circuit pattern defect and determines whether to perform processing on the identified circuit pattern defect and the result after processing; and an AI parameter control unit that adaptively determines processing parameters of the laser irradiation unit and the AOD unit based on the location, shape, and type of the circuit pattern defect identified by the AI ​​vision control unit, and dynamically readjusts the processing parameters according to the result after processing; wherein the synchronization control unit comprises: a master clock member that generates a reference clock signal; and a trigger distribution member that distributes the reference clock signal generated from the master clock member as a trigger signal to each of the laser irradiation unit, the vision unit, and the illumination unit.The AI ​​vision control unit comprises: an image preprocessing member that performs noise removal, contrast correction, and pixel alignment on the fluorescent image and the reflection image to align the fluorescent image and the reflection image into the same coordinate system; a multimodal fusion member that combines the fluorescent image and the reflection image aligned by the image preprocessing member using a channel-wise fusion method or a post-fusion method to generate a single multimodal input data; a defect pattern recognition member that extracts the location and shape of the region where the circuit pattern defect exists from the multimodal input data generated by the multimodal fusion member using a pre-trained neural network; a neural network defect classification member that classifies the circuit pattern defect extracted by the defect pattern recognition member into one of a short defect, a protrusion defect, and a residue defect using a pre-trained neural network; a segmentation member that divides the boundary of the circuit pattern defect extracted by the defect pattern recognition member into pixel units and sets it as a processing area (ROI: Region Of Interest); and the fluorescent image and the reflection image before processing. A processing completion determination member that determines whether the circuit pattern defect remains in the processing area by comparing and analyzing the image with the fluorescent image and the reflection image after the processing is performed; and a reprocessing sequence member that, when the processing completion determination member determines that the circuit pattern defect remains, automatically triggers a reprocessing sequence for the remaining circuit pattern defect and transmits a reprocessing command to the processing sequence control member; wherein the AI ​​parameter control member includes a parameter prediction member that takes as input the type, shape, and location information of the circuit pattern defect, the material of the circuit board, and the thickness information of the metal pattern formed on the circuit board, and outputs processing parameters of the laser irradiation unit and scan parameters of the AOD unit by a pre-trained deep neural network model;A laser parameter control member that variably controls the pulse width, pulse repetition frequency, pulse energy, and number of pulses of the processing laser part of the laser irradiation unit based on the result output from the parameter prediction member; an AOD scan pattern control member that variably controls the scan trajectory, scan speed, scan area, and scan pattern of the AOD part based on the result output from the parameter prediction member; a pulse timing control member that synchronizes the pulse emission timing of the processing laser part with the acoustic frequency change timing of the AOD part to control the processing laser light so that it is accurately irradiated at an intended location; an adaptive control member that receives residual defect information as feedback from the processing completion judgment member of the AI ​​vision control unit as a result of the previous processing sequence and adaptively corrects the processing parameters to be applied to the next processing sequence based on the feedback information; and a safety control member that dynamically limits the upper limit of the processing parameters so that the processing parameters of the processing laser part do not exceed a threshold value that causes damage to the substrate portion of the circuit board; characterized by comprising: a PCB and RDL circuit laser repair device. Claim 2 In claim 1, the beam expander unit comprises: a first lens member that expands the diameter of the processing laser light incident from the processing laser unit; and a second lens member that shapes the processing laser light emitted from the first lens member into parallel light; the DM unit comprises: a dichroic mirror member having wavelength selectivity; and a mirror holder member that finely adjusts the angle with respect to the optical axis of the dichroic mirror member; the objective lens unit of the optical focusing unit comprises an objective lens member that focuses the processing laser light onto the circuit board; the tube lens unit comprises a tube lens member that forms the reflected light and the emitted light onto an image sensor of the camera unit of the vision unit; the BS unit comprises a beam splitter member that branches or combines the optical path of the processing laser light and the optical path of the light incident on the vision unit; and the camera unit of the vision unit comprises a CCD or CMOS type image sensor member; and further comprising an exposure time adjustment member that variably adjusts the exposure time of the image sensor member; and the fluorescence blocking filter member comprises: a notch filter member that blocks the output wavelength band of the fluorescence illumination member; and a filter holder member that mounts the notch filter member at the incident end of the camera member; and the image processing member comprises: an image matching member that aligns pixels of the fluorescence image and the reflection image; a difference image generating member that calculates the difference between the fluorescence image and the reflection image aligned by the image matching member to generate a difference image; and a rule-based defect region extraction member that applies a rule-based image processing algorithm to the difference image generated by the difference image generating member to extract the region where the circuit pattern defect exists. and a rule-based defect classification member that classifies the circuit pattern defect extracted from the rule-based defect region extraction member into any one of a short defect, a protrusion defect, and a residue defect; wherein the fluorescent lighting part of the lighting unit comprises: an NUV light source member that generates light of an ultraviolet or near-ultraviolet wavelength; and a polarizer member (p) that shapes the light emitted from the NUV light source member into p-polarization;The coaxial illumination unit comprises: an RB light source member that generates red and blue visible light; a polarizer member (s) that shapes the light emitted from the RB light source member into s-polarization; and a beam alignment member that directs the light emitted from the RB light source member to the surface of the circuit board in a coaxial direction through the BS unit; and the ring illumination unit comprises a ring LED array member arranged annularly on the outer side of the objective lens unit of the optical focusing unit to irradiate visible light onto the circuit board from a lateral direction. The lighting controller comprises: an angle adjustment member for variably adjusting the illumination angle of the ring LED array member; and the lighting controller comprises: a switching timing control member that performs time-division switching control so that only one of the fluorescent lighting unit, the coaxial lighting unit, and the ring lighting unit is turned ON at a single point in time; a synchronization pulse generation member that generates a trigger pulse and outputs it to the camera unit so that the switching operation of the switching timing control member and the exposure operation of the camera unit are synchronized; and a light intensity control member that variably adjusts the illumination light intensity of each of the fluorescent lighting unit, the coaxial lighting unit, and the ring lighting unit; and the processing sequence control unit of the control unit comprises: a closed-loop trigger member that triggers a reprocessing sequence based on whether the circuit pattern defect remains; a residual defect determination member that determines whether the circuit pattern defect remains based on the fluorescent image and the reflection image acquired after processing; and a repetition count upper limit setting member that sets an upper limit value for the number of repetitions of the closed-loop processing sequence. A PCB and RDL circuit laser repair device characterized by including: a sequence termination condition determination member that determines the termination of the closed-loop processing sequence when no circuit pattern defects remain or when the number of repetitions reaches the upper limit value. Claim 3 In claim 1, the AI ​​vision control unit comprises: a defect importance evaluation member that calculates a risk score by quantifying the degree of influence of the circuit pattern defect on the circuit function of the circuit board based on the type of the circuit pattern defect classified by the neural network defect classification member and the size and location of the processing area set by the segmentation member; a processing priority setting member that determines the processing order of the plurality of circuit pattern defects based on the risk score calculated by the defect importance evaluation member when a plurality of circuit pattern defects are identified on the circuit board; a processing difficulty prediction member that predicts the difficulty of processing in advance based on the type and shape of the circuit pattern defect and material information of the circuit board; a processing trace analysis member that detects damage or residues generated on the substrate portion of the circuit board by analyzing the surface condition around the processing area from the fluorescent image and the reflection image after the processing is performed; and a model corresponding to the input information among a plurality of pre-trained models stored in advance, which receives input information on the type of the circuit board, the material of the circuit board, and the thickness of the metal pattern formed on the circuit board. A model selection member applied to the defect pattern recognition member and the neural network defect classification member; a transfer learning member that fine-tunes the parameters of the pre-trained model selected by the model selection member to match the process conditions of the circuit board currently being worked on; an online learning member that accumulates the fluorescent image, the reflection image, and the judgment result of the processing completion judgment member collected during the processing sequence as learning data to update the pre-trained model in real time; and a comprehensive diagnostic member that, after the completion of the processing sequence, synthesizes the distribution of the plurality of circuit pattern defects identified on the circuit board, the judgment result of the processing completion judgment member, and the analysis result of the processing trace analysis member to generate a diagnostic report for the circuit board.The defect pattern recognition member further comprises: a feature extractor that extracts a feature map from the multimodal input data by a pre-trained convolutional neural network including a plurality of convolutional layers; a candidate region detector that detects candidate regions of the circuit pattern defect from the feature map extracted by the feature extractor; and a region determination member that determines the existence of the circuit pattern defect for each of the candidate regions detected by the candidate region detector and outputs the location coordinates of the circuit pattern defect; and the segmentation member comprises a pixel labeling member that labels whether each pixel of the multimodal input data corresponds to the circuit pattern defect using a pre-trained semantic segmentation neural network of an encoder-decoder structure. The boundary calculation unit calculates the boundary coordinates of the processing area based on the results labeled by the pixel labeling unit; and the processing completion determination unit includes: a difference image calculation unit that generates a difference image by calculating the difference between the fluorescent image and the reflection image before processing and the fluorescent image and the reflection image after processing in pixel units; a threshold determination unit that determines whether there exists an area where the intensity of the difference image generated by the difference image calculation unit exceeds a preset threshold; and a determination result output unit that outputs a final determination result regarding whether the processing is completed based on the determination result of the threshold determination unit; and the AI ​​parameter control unit includes: a beam shape control unit that adjusts the beam diameter and beam profile of the processing laser light by variably controlling the beam magnification ratio of the beam expander unit based on the result output from the parameter prediction unit; and a processing path optimization unit that receives the location coordinates of a plurality of circuit pattern defects identified on the circuit board and calculates a processing sequence and a processing path such that the total travel distance of the processing path between the plurality of circuit pattern defects is minimized.A reinforcement learning component that learns, based on reinforcement learning, the correlation between the processing parameters applied during the processing of the circuit pattern defect and the result after processing, and updates the policy of the processing parameters to be applied in the next processing sequence; a Bayesian optimization component that searches for the optimal value of the processing parameters with a small number of trials using a Bayesian optimization algorithm when there is insufficient collected data for a new type of circuit board; a model predictive control component that determines the processing parameters to be applied in the next iteration step of the closed-loop processing sequence based on a model predictive control algorithm; a hybrid control component that secures the control stability of the processing parameters by generating a final control value through a weighted combination of the output of the parameter prediction component and the output of the PID (Proportional-Integral-Derivative) control algorithm; a multi-objective optimization component that searches for a Pareto optimum for a multi-objective function using the speed of the processing sequence, processing quality, and the degree of damage to the substrate portion of the circuit board as evaluation indicators; and the circuit according to the scan pattern of the AOD unit An energy distribution optimization member that learns the energy distribution formed on a substrate and provides a correction value to the AOD scan pattern control member so that a uniform energy distribution is formed within the processing area; an environment correction member that receives the temperature and humidity of the working environment, calculates an environment correction value for the processing parameter, and applies it to the laser parameter control member and the AOD scan pattern control member; and an output correction member that measures the change in the output light quantity of the processing laser unit over time, calculates an output correction value to compensate for the measured change in light quantity, and applies it to the laser parameter control member.A digital twin component that generates a virtual model of the circuit board and the circuit pattern defect and performs a processing simulation on the virtual model to pre-verify the processing parameters before actual processing; a parameter database component that stores optimal parameter information corresponding to a combination of the type of circuit board, defect type, and processing parameters in advance and outputs the optimal parameter information upon a request from the parameter prediction component; a batch learning component that analyzes processing result data collected in batch units consisting of multiple circuit boards to derive an improved profile of the processing parameters to be applied to the next batch; and a quality-based learning component that updates the policy of the reinforcement learning component and the weight of the parameter prediction component using the processing quality evaluation result measured after the completion of the processing sequence as a compensation function. The apparatus further includes a federation learning member that operates in a distributed learning manner by integrating learning data collected from each of a plurality of PCB and RDL circuit laser repair devices to update the pre-trained deep neural network model, wherein the raw data of each device is preserved within each device and only update information of model parameters is integrated; wherein the parameter prediction member includes: an input converter that converts information on the circuit pattern defect and information on the circuit board into an input vector; a prediction output unit that inputs the input vector into the pre-trained deep neural network model and outputs predicted values ​​of the processing parameters and the scan parameters; and a constraint verification unit that determines the final parameter value by verifying whether the predicted value output from the prediction output unit is within a physically feasible range; and wherein the adaptive control member includes: a feedback receiver that receives residual defect information from the processing completion determination member; and an adjustment amount calculation unit that calculates the adjustment amount of the processing parameters by comparing and analyzing the residual defect information and the processing parameters applied in the previous processing sequence.A PCB and RDL circuit laser repair device characterized by including: a parameter update body that applies the adjustment amount calculated by the adjustment amount calculation body to the laser parameter control member and the AOD scan pattern control member to update the processing parameters of the next processing sequence;

Images