Semiconductor chip defect on-line detection system based on optical physical characteristics

Abstract

The invention relates to the technical field of material detection, and discloses a semiconductor chip defect online detection system based on optical physical characteristics, which comprises a coherent light projection module, a reference light projection module, a displacement driving feedback unit, a discrete photoelectric detection array, a material support body and a logic processing unit, the displacement driving feedback unit acquires a rotating speed pulse signal, the logic processing unit calculates a laser sampling frequency and adjusts a pulse triggering time sequence, so that the axial displacement of the strip to be measured is constantly within a preset resolution range, the coherent light projection module and the reference light projection module project light beams at different angles, and a differential optical path is constructed; the discrete photoelectric detection array collects composite optical signals, the logic processing unit counteracts background phase noise by means of signal differential logic and extracts phase features of a speckle pattern to determine defect space coordinates, mechanical interference is stripped at a sensing source by means of a physical hedging mechanism, and optical dynamic ambiguity under the high-speed working condition is effectively eliminated.

Description

Technical Field

[0001] This invention belongs to the field of materials testing technology, and in particular relates to an online defect detection system for semiconductor chips based on optical physical properties. Background Technology

[0002] In the current precision metal strip processing of semiconductor leadframes, the uniformity of the material's surface microstructure and internal stress distribution directly determines the reliability of chip packaging. Real-time analysis of material physical properties using optical detection methods is the mainstream solution to ensure production line quality. Under ultra-high-speed rolling conditions, the strip linear speed is usually higher than 5000 mm / s. The high-frequency mechanical vibration generated by the rolling mill and the dynamic fluctuation of the strip surface are intertwined, resulting in a large amount of background displacement noise mixed in the signal collected by the sensor. To suppress noise, conventional solutions use low-pass filters, but this produces physical phase lag, causing the detection point and the rolling gap to become disconnected in the spatiotemporal domain. Since the control system cannot capture the transient stress changes within the rolling gap, its detection is essentially a post-hoc remedy for defects that have already occurred, which is difficult to meet the leadframe material's standards for zero surface defects and stress isotropy.

[0003] Besides hardware limitations, the control methods also have shortcomings. For example, Chinese invention patent CN116030053B discloses a method, device, equipment, and medium for detecting connector pin defects. It acquires images through top-down and side-rotational perspectives and uses circular or rectangular geometric features for binarization to determine the pin position and length. Such methods are based on static or quasi-static imaging and rely on discrete image feature comparison. They cannot adapt to the optical motion blur caused by ultra-high-speed rolling. Under strong vibration backgrounds, there is no physical means to remove dynamic phase noise. They cannot distinguish between the background vibration of the equipment and the microscopic evolution of internal stress in the material. The sensing rate is mismatched with the rheological scale of the strip. The lag detection mechanism that relies on geometric shape scanning cannot perform closed-loop intervention at the defect initiation stage. Simply increasing the sampling frequency or increasing the stiffness of the measuring support cannot eliminate the phase shift caused by physical limits. This inherent trade-off between the sensing signal-to-noise ratio and the real-time feedback restricts the further improvement of the production accuracy of ultra-thin metal strips.

[0004] Therefore, how to accurately sense and synchronously compensate for transient stress inside the strip under strong background noise and high-speed motion environment has become the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides an online defect detection system for semiconductor chips based on optical physical properties, used for defect detection of the strip material to be tested on a material support. The system includes a coherent light projection module, a reference light projection module, a displacement drive feedback unit, a discrete photoelectric detection array, a material support, and a logic processing unit.

[0006] The displacement drive feedback unit is used to acquire the rotational speed pulse signal of the drive end of the strip under test;

[0007] The logic processing unit is used to calculate the laser sampling frequency based on the rotation speed pulse signal. And according to the laser sampling frequency Adjusting the pulse triggering timing of the coherent light projection module and the reference light projection module, by adjusting the laser sampling frequency... Synchronize with the rotational speed pulse signal to ensure that the axial displacement of the strip under test within a single pulse cycle does not exceed the preset feature resolution;

[0008] A coherent beam projection module is used to project a coherent main beam onto the surface of the strip under test.

[0009] The reference beam projection module is used to project an incoherent reference beam onto the projection area of ​​the coherent main beam on the surface of the strip under test. The incident angles of the coherent main beam and the incoherent reference beam are different, so as to construct a differential optical path.

[0010] Discrete photoelectric detection array is used to acquire composite optical signals, including mechanical vibration noise and micro-deformation signals, reflected from the surface of the strip under test;

[0011] The logic processing unit is used to execute the deformation calculation program, which includes: performing signal differential processing on the composite optical signal, using the common-mode component in the composite optical signal to cancel the background phase noise generated by mechanical vibration, extracting the phase features of the coherent pattern modulated by the micro-stress of the strip under test, and determining the spatial coordinates of the defects in the strip under test based on the phase distribution discontinuities in the phase features of the coherent pattern.

[0012] Preferably, the logic processing unit is also used to perform sampling power synchronization compensation: the logic processing unit is based on the laser sampling frequency. The reciprocal of the peak compensation gain of the driving current is determined, and the coherent light projection module is driven to adjust the laser emission power according to the peak compensation gain so as to maintain the reference light intensity value of the detection area constant when the rotation speed pulse signal changes.

[0013] Preferably, it also includes an active excitation calibration module; the active excitation calibration module is used to apply a pulse force disturbance signal of a preset frequency to the material support; the logic processing unit is used to collect the dynamic phase response characteristics of the discrete photoelectric detection array to the pulse force disturbance signal, invert the surface normal stiffness of the material support, and update the preset reference stress feature vector accordingly.

[0014] Preferably, the logic processing unit is used to execute a weight iteration algorithm to correct the reference stress eigenvector, and the calculation formula of the weight iteration algorithm is expressed as: ,in, This is the corrected reference stress eigenvector. The first calculation based on surface normal stiffness Adjustment weights, It is a positive integer. These are reference components pre-stored in the logic processing unit and used to characterize the evolution of the surface morphology of the material support.

[0015] Preferably, the logic processing unit is also used to extract the high-frequency scintillation envelope from the phase features of the coherence pattern and perform spectral density analysis on the high-frequency scintillation envelope to calculate the fluid load pressure component on the surface of the strip under test.

[0016] Preferably, the logic processing unit introduces the fluid load pressure component as a phase delay correction factor into the deformation calculation program to counteract the optical path difference signal interference caused by interfacial fluid fluctuations.

[0017] Preferably, the logic processing unit is also used to monitor the temporal coherence intensity attenuation ratio of the phase characteristics of the coherence pattern, and when the intensity attenuation ratio exceeds a preset threshold, output technical parameters characterizing the fatigue state of the surface of the strip under test.

[0018] Preferably, the discrete photoelectric detection array includes multiple sets of photoelectric sensors arranged in a matrix, and the spatial distance between two adjacent sets of photoelectric sensors is no greater than 0.5 mm.

[0019] Preferably, the coherent light projection module includes a distributed feedback laser and a collimation and beam expansion assembly, with the output of the distributed feedback laser connected to a frequency modulation circuit driven by a logic processing unit.

[0020] Preferably, the reference light projection module includes a superluminescent diode controlled by a logic processing unit, and the optical axis of the reference light projection module and the optical axis of the coherent light projection module coincide at the intersection point on the surface of the strip under test.

[0021] Compared with existing technologies, the online defect detection system for semiconductor chips based on optical physical properties of this invention has the following advantages:

[0022] 1. In online defect detection of semiconductor chips, the system's sensing logic is based on the spatial phase modulation mechanism of coherent light generated during the metal plastic deformation process. A coherent light projection module projects a laser onto the strip surface at the roll exit end, utilizing the coherent pattern generated by the metal's micro-roughness to represent stress characteristics. The feature extraction module calculates the contrast distribution gradient of the pattern. and statistical characteristic values This transforms the abstract internal stress distribution of a material into a quantifiable optical field feature vector. This approach enables direct characterization of the microscopic physical properties of materials. Compared to traditional post-detection methods that rely on geometric dimension scanning, this solution utilizes the interaction between light-speed-level feature sensing and millisecond-level hydraulic compensation to allow the hydraulic actuator to counteract local stress concentrations before the strip enters the defect initiation stage, thereby completing adaptive shape repair before defects are formed.

[0023] 2. To address the common mechanical-dynamic coupling problem in precision rolling environments, the system employs a dual-path optical structure. An incoherent reference beam directed towards the same region is introduced alongside the main coherent beam to acquire the background vector reflecting the geometric runout of the strip. The feature extraction module extracts coherent feature vectors. With background vector By performing spatial phase difference calculations, geometric displacement noise caused by high-frequency resonance of the rolling mill is removed at the sensing source, thereby extracting a pure stress mapping vector. This denoising mechanism based on physical offset avoids feedback phase lag caused by software filtering, ensuring that the system can still accurately identify the micro-stress evolution inside the strip under strong vibration conditions, thus improving the signal-to-noise ratio of the measurement signal.

[0024] 3. During high linear speed operation, the system obtains the speed encoder signal of the roll motor in real time through the speed feedback module. Based on this, the pulse width and emission frequency of the coherent light projection module are dynamically adjusted. This hard synchronization mechanism ensures that the physical displacement of the strip within a single sampling period is always limited to the preset feature resolution range, effectively eliminating the optical motion blur caused by high flow rate. At the same time, the driving circuit compensates for the peak power of the laser based on the reciprocal of the sampling frequency, maintaining a constant calculation benchmark for the optical field entropy value, and ensuring that the detection sensitivity of the system remains consistent throughout the entire process of startup acceleration, high-speed stable operation, and deceleration and stopping. Attached Figure Description

[0025] Figure 1 This is a diagram showing the overall hardware architecture and photoelectric signal transmission link of the detection system of this invention;

[0026] Figure 2 This is a block diagram of the closed-loop control architecture and signal processing interaction of the logic processing unit of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0028] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0029] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0030] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0031] An online defect detection system for semiconductor chips based on optical physics characteristics comprises a coherent light projection module, a reference light projection module, a displacement drive feedback unit, a discrete photodetector array, a material support, and a logic processing unit. The coherent light projection module and the reference light projection module are located in the critical deformation zone at the exit end of the rolling mill rolls, used to project light beams onto the surface of the moving strip under test to construct a differential optical path. The discrete photodetector array collects composite optical signals reflected from the surface of the strip under test, including mechanical vibration noise and microscopic deformation signals. The displacement drive feedback unit acquires the rotational speed pulse signal at the driving end of the strip under test and transmits it to the logic processing unit. The logic processing unit adjusts the pulse triggering timing according to the rotational speed pulse signal and executes a deformation calculation program to extract the phase features of the coherent pattern. The logic processing unit executes a coordinate mapping procedure to determine the spatial coordinates of the defect, which indexes the pixels of the discrete photodetector array. Converted to physical coordinates of the strip surface The conversion is based on the object-side magnification of the optical system. Physical spacing between pixels Horizontal physical coordinates Horizontal indexing of pixels Physical spacing between pixels The product divided by the object magnification Determine the vertical physical coordinates. Vertical indexing by cell Physical spacing between pixels The product divided by the object magnification The absolute position coordinates of the strip in the direction of travel are determined by the rotational speed pulse signal acquired in real time by the displacement drive feedback unit. Accumulated displacement It is determined that the entire system utilizes the spatial phase modulation mechanism of coherent light generated by the plastic deformation process of the material to convert the internal stress distribution of the material into a quantized optical field feature vector. .

[0032] In the rolling process of high-precision metal strips for semiconductor lead frames, the metal rheology within the rolling deformation zone exhibits transient instability due to the strip thickness being on the micrometer scale. Traditional detection methods rely on external optical scanning after rolling, which is essentially a post-processing measure and cannot capture transient stress changes within the rolling gap. To overcome this obstacle, the system uses a coherent light projection module to project coherent laser light onto the surface of the strip under test at the roll exit. It utilizes the microscopic roughness of the metal surface to generate a coherent pattern of stress characteristics. The logic processing unit executes a deformation calculation program to extract the contrast distribution gradient of the coherent pattern. and statistical characteristic values The logic processing unit maps the intensity distribution of the pattern acquired by the discrete photodetector array into a pixel matrix, and determines the contrast distribution gradient by calculating the second moment of the pixel intensity within a local window. And determine the statistical characteristic values ​​according to the following formula. : ,in, This represents the maximum light intensity value within the patch. The minimum light intensity value, This is the average light intensity value. The standard deviation of the light intensity distribution is used by the logic processing unit to establish a mapping relationship between the stress field and the light field distribution. The magnitude of the light field feature vector is calculated by the following logic: multiplying the contrast distribution gradient by a gain coefficient of 0.7, and adding the statistical feature value by a gain coefficient of 0.3. This proportional coefficient is obtained by looking up and comparing the static test results of a known 3.5MPa load sample in the early stage of system deployment, thereby ensuring that the feature quantity and the physical stress state have a one-to-one numerical mapping relationship. The system transforms the detection object from static geometric defects to dynamic stress evolution, enabling it to perform intervention in a quasi-static state before the defect has formed.

[0033] Using object magnification Physical spacing between pixels Establish a physical space coordinate system and determine the gradient based on the contrast distribution. With internal stress Inter-proportional coefficient Quantitatively analyze the stress distribution characteristics. A discrete photoelectric detection array acquires composite optical signals and converts them into a 512x512 pixel intensity matrix. Spatial gradient convolution is performed to extract horizontal and vertical gradient components. The second moment of pixel intensity within an 8x8 local sampling window is calculated to determine the intensity distribution. light field feature vector Module length corresponding With statistical characteristic values Product, statistical characteristic value The maximum light intensity value in the pixel intensity matrix With minimum light intensity value The difference in relative average light intensity with standard deviation The ratio of the product is determined; during the rolling process, the high-frequency mechanical vibration generated by the rolling mill system and the physical jitter between the strip under test and the rolling process cause random displacement vector shifts in the coherent pattern. Traditional software filtering often produces physical phase lag, thus creating a sensing blind zone. The system projects incoherent reference light onto the projection area of ​​the coherent main beam on the surface of the strip under test through the reference light projection module. Since the incident angles of the coherent main beam and the incoherent reference light are different, the system constructs a differential optical path with physical offset. The discrete photoelectric detection array collects the composite optical signal, and the logic processing unit executes a signal differential program on the composite optical signal. By extracting the common-mode component of the incoherent reference light in the reflected signal, the background vector caused by the physical jitter of the strip under test is determined. The logic processing unit will process the optical field feature vector that reflects the stress state of the strip. With background vector Spatial phase difference calculations are performed to remove background phase noise generated by mechanical vibration, thus eliminating the masking of defect characteristic signals by high-frequency resonance of the rolling mill.

[0034] Under high-speed rolling conditions, the rapid displacement of the strip surface under test causes dynamic overlap of the light field information captured by the sensor, i.e., optical blurring, which leads to changes in statistical characteristic values. The decrease in signal-to-noise ratio causes random jitter in the reduction correction command; to overcome this obstacle, the system uses a displacement drive feedback unit to acquire the rotational speed pulse signal of the roll motor in real time, and the logic processing unit calculates the laser sampling frequency based on the rotational speed pulse signal. Based on this, the pulse triggering timing of the coherent light projection module and the reference light projection module is dynamically adjusted, and the logic processing unit obtains the speed encoder signal of the motor. Calculate the real-time linear velocity of the strip under test. By shortening the width of a single coherent pulse and increasing the sampling frequency, the physical displacement of the strip under test within a single sampling period is ensured to be no greater than the preset feature resolution. This synchronization mechanism eliminates optical feature degradation caused by flow velocity; real-time linear velocity When the speed is above 5000 mm / s, the logic processing unit determines the speed based on the rotational speed pulse signal. Adjusting the coherent pulse width ,make The critical exposure time, not exceeding the axial displacement within a preset resolution range, compensates for energy attenuation caused by shortened exposure, based on the laser sampling frequency. Reciprocal calculation of peak compensation gain The driving frequency modulation circuit adjusts the driving current of the distributed feedback laser to achieve the reference light intensity value in the detection area. Maintaining a constant rotational speed eliminates the standard deviation of light intensity distribution. Nonlinear fluctuations cause statistical eigenvalues The sensitivity to micro-stress modulation response remains consistent at different rates.

[0035] The heat generated during continuous rolling causes thermal expansion of the rolls, known as thermal crown. This slow physical deformation causes a shift in the static geometry of the rolling gap, resulting in temperature drift of the initial stress reference established by the system. The system employs an active vibration calibration module to apply pulsed force disturbance signals of a preset frequency to the material support. The perturbation frequency is set between 10Hz and 50Hz, and the logic processing unit acquires the pulse force perturbation signal from the discrete photoelectric detection array. The dynamic phase response characteristics are used to invert the effective stiffness and physical morphology of the material support surface, and a weighted iterative algorithm is executed to correct the reference stress eigenvector. The corrected formula is expressed as: ,in, This is the corrected reference stress eigenvector. The first is determined based on the normal stiffness of the material support surface. Adjustment weights, It is a positive integer. As a reference component pre-stored in the logic processing unit to characterize the evolution of surface morphology, this mechanism automatically cancels the detection reference offset caused by micro-aging or thermal deformation of the roll surface; the driving material support rotates at a constant speed under no-load for at least a full cycle, and the discrete photoelectric detection array collects data. Frame pixel intensity matrix, before performing principal component analysis on the matrix. The eigenvectors of each term are stored in non-volatile memory as reference components. The active excitation calibration module applies a pulsed force disturbance signal. The logic processing unit monitors the coherence speckle pattern in Dynamic phase response characteristic inversion surface normal stiffness under excitation Calculate each item The adjustment weight is determined based on the linear mapping relationship between preset stiffness and phase offset for the change in response amplitude. The reference stress eigenvector is updated and corrected after quantization. .

[0036] Under high-speed finishing rolling conditions, local thickness fluctuations in the lubricating oil film between the rolls and the strip can generate additional optical path phase delays, creating false defect detection signals. To address this challenge, the logic processing unit extracts the high-frequency scintillation envelope from the phase characteristics of the coherent pattern and performs a spectral density analysis program on it. The system utilizes the causal relationship between the turbulence intensity generated by the interfacial fluid flow and the pattern contrast to calculate the fluid load pressure component on the surface of the strip under test. The logic processing unit will process the fluid load pressure component. Introduced as a phase delay correction factor into the deformation calculation program to counteract optical path difference signal interference caused by interfacial fluid fluctuations, this scheme achieves a deep understanding of the liquid-solid interface state in the rolling deformation zone and reduces the risk of false alarms caused by hydrodynamic interference. The system also includes an arbitration module for processing the optical field feature vector. Perform spatial autocorrelation analysis along the direction of travel of the strip under test; when the optical field eigenvector autocorrelation function When the displacement is equal to the roll circumference When a peak occurs, the arbitration module determines that the signal originates from local peeling or physical damage on the roll surface. At this time, the system automatically suppresses the weight of this component in the reduction compensation operator and outputs an actuator morphology warning signal. By using the physical law of periodic contact in strip rolling as a filter, the decoupling analysis of material defect characteristics and actuator fault characteristics is realized.

[0037] Example 1: At a thickness of 150 In the continuous finishing rolling process of high-precision copper alloy strip for semiconductor lead frames, the mill line speed is maintained at 5000 mm / s. Mechanical resonance with a frequency of 200 Hz to 500 Hz exists in the critical deformation zone at the roll exit. Traditional detection methods suffer from feature ambiguity due to the mismatch between the optical sampling frequency and the strip displacement step size. Furthermore, the mechanical vibration component interferes with the phase distribution of the coherent pattern, affecting the statistical characteristic values. The signal-to-noise ratio drops below the preset threshold; the system obtains the speed encoder signal of the roll motor through the displacement drive feedback unit. The logic processing unit is based on the speed encoder signal. Calculate real-time linear velocity And based on real-time linear velocity Determine the laser sampling frequency laser sampling frequency With real-time linear velocity To maintain linear synchronization, the axial displacement within a single pulse cycle is kept at 1 by adjusting the coherent pulse trigger timing. The resolution is on the order of m; the reference light projection module projects incoherent reference light onto the strip surface and constructs a differential optical path; the discrete photodetector array collects composite optical signals containing background noise; and the logic processing unit extracts the background vector generated by the physical fluctuations of the strip. And the light field feature vector that reflects deformation information With background vector Perform phase offsetting.

[0038] laser sampling frequency Synchronous sampling eliminates feature attenuation caused by motion blur, while the physical differential path removes background vectors. Enhance the light field eigenvector The extraction accuracy is improved, and the synergistic effect of the two enables the system to identify the pattern phase distortion caused by stress change under the working condition of 5000mm / s linear speed, realize the spatial coordinate of the defect on the strip surface and drive the hydraulic actuator to perform the reduction compensation, and convert the random physical fluctuation under high speed condition into a definite feature vector offset.

[0039] Example 2: The experiment was conducted on a physical testing platform simulating high-precision cold rolling conditions. Data was sourced from the platform's real-time acquisition system. The physical testing platform included a support structure for the strip material under test and a servo motor that drove the support structure to rotate. The servo motor was equipped with a speed encoder with a resolution of no less than 65,536 lines per revolution to provide speed pulse signals. The test environment introduced 50Hz power frequency electromagnetic interference and 20dB background vibration noise. The laser sampling frequency was [not specified]. The value is based on the real-time linear velocity. With preset axial resolution The ratio is determined by increasing the sampling frequency to suppress optical blurring caused by high-speed displacement of the strip under test, while avoiding data processing load exceeding the hardware computing power limit of the logic processing unit, for a real-time linear velocity of 5000 mm / s. Set the laser sampling frequency To ensure that the single pulse period covers a constant displacement step size of 1 μm, the frequency is 5 MHz. This is a rotational speed pulse signal, measured in lines per revolution. This refers to the laser sampling frequency, measured in Hz. Real-time linear velocity, in mm / s, Deltax res Preset axial resolution, unit: m.

[0040] The experiment was conducted using three control groups: one without a physical differential path and sampling synchronization mechanism, another lacking a displacement-driven feedback unit, and the third lacking a reference light projection module. The test group employed the scheme described in this invention. The test was conducted on a strip with a depth of 5 μm. Real-time linear velocity under mum surface defect conditions When the speed is increased to 3000 mm / s, the contrast distribution gradient of control group 2 is generated due to optical motion blur. The diffusion and defect extraction accuracy was 72.1%, while the control group 3 suffered from optical field feature vector distortion due to 500Hz mechanical resonance interference. The signal-to-noise ratio dropped to 4.5 dB. The experimental group utilized the laser sampling frequency. With speed pulse signal The synchronization mechanism maintains feature resolution while simultaneously using the background vector. With light field feature vector Spatial phase offset stripping noise maintains a signal-to-noise ratio of 16.8 dB at a speed of 5000 mm / s. The contrast distribution gradient is dimensionless. For light field feature vectors, This is the background vector.

[0041] To verify the correlation between performance boundaries and gradients, real-time linear velocity was tested while keeping environmental disturbances constant. The system's ability to capture 1μm-level defects was tested by increasing the speed from 1000mm / s to 6000mm / s. Measurement data showed that under the gradient conditions of 1000mm / s, 2000mm / s, 3000mm / s, 4000mm / s, and 5000mm / s, the defect extraction accuracy of the experimental group was 99.8%, 99.7%, 99.5%, 99.3%, and 99.1%, respectively. (Real-time linear velocity...) When the speed is increased to 5500 mm / s, the defect extraction accuracy is 95.2%, and the real-time linear speed is [missing information]. After exceeding the upper limit of 6000 mm / s, the defect extraction accuracy drops to 81.6%. At this point, the integration time within a single sampling period approaches the integration limit of the discrete photoelectric detection array, leading to an increase in the standard deviation of the light intensity distribution. Nonlinear saturation occurs, where The standard deviation of light intensity distribution is expressed in candela. Experimental results show that the displacement-driven feedback unit and the differential optical path achieve physical synergy at a real-time linear velocity of 5000 mm / s. Realizing background vectors under strong mechanical interference environments With light field feature vector Decoupling was used to solve the measurement distortion problem caused by the superposition of dynamic fuzziness and environmental vibration in the online inspection of high-precision metal strips, and the system's detection sensitivity and engineering reliability in the face of transient rheological stress changes were verified.

[0042] Example 3: This example combines Figures 1 to 2 A description of an online defect detection system for semiconductor chips based on optical physical properties, such as... Figure 1 As shown, it mainly consists of a displacement drive feedback unit, a logic processing unit, a coherent light projection module, a reference light projection module, a discrete photodetector array, and a material support. The displacement drive feedback unit is responsible for acquiring the rotational speed pulse signal of the drive end of the strip under test and transmitting the rotational speed pulse signal to the logic processing unit. The logic processing unit, as the core control center, performs tasks such as adjusting the pulse trigger timing, signal differential processing, and determining the defect coordinates. It is connected to the coherent light projection module through the pulse trigger timing and sampling frequency control circuit to control the projection of the coherent main beam, and connected to the reference light projection module through the timing control circuit to control it to project incoherent reference light at different angles. The coherent main beam and the incoherent reference light converge on the surface of the strip under test on the material support. The strip under test reflects a composite optical signal containing noise and deformation, which is transmitted to the discrete photodetector array through the differential optical path. After the discrete photodetector array collects the composite optical signal, it sends a digital signal feedback to the logic processing unit. The logic processing unit finally maps the phase characteristics of the coherent pattern and outputs the spatial coordinates of the defect.

[0043] like Figure 2 As shown, the logic processing unit, displacement drive feedback unit, discrete photoelectric detection array, coherent light projection module, reference light projection module, and active excitation calibration module constitute a closed-loop control architecture. At the control level, the logic processing unit calculates the laser sampling frequency and adjusts the pulse triggering sequence based on the input from the displacement drive feedback unit, thereby controlling the actions of the coherent light projection module and the reference light projection module. Simultaneously, it performs sampling power synchronization compensation to drive the coherent light projection module to adjust the laser emission power. At the signal processing level, the logic processing unit receives data from the discrete photoelectric detection array and executes a deformation calculation program. This program specifically includes performing signal differential processing, extracting coherent pattern phase features, and determining the spatial coordinates of the defect. In addition, the logic processing unit also performs tasks in parallel to calculate the fluid load pressure component and output surface fatigue state technical parameters. It can also invert the surface normal stiffness based on the interactive signal from the active excitation calibration module to update the reference stress feature vector.

[0044] Example 4: At a thickness of 10 In the aluminum foil rolling process, the roll bearing position experiences micron-level thermal elongation displacement. The logic processing unit uses a 3×3 operator to process the 512×512 pixel intensity matrix. Perform convolution to generate horizontal gradient components With vertical gradient components Calculate the local area according to the following formula Contrast distribution gradient within the window : ,in, The contrast distribution gradient, For horizontal gradient components, For vertical gradient components, The average pixel intensity within the window; the active excitation calibration module applies a 30Hz pulsed force perturbation signal to the material support. The logic processing unit acquires the displacement response amplitude fed back by the discrete photoelectric detection array. When the displacement response amplitude When the weight is increased by 15% compared to the baseline, the adjustment weight The value was reduced from 1.0 to 0.85 to correct the reference stress eigenvector. .

[0045] The logic processing unit extracts frequency components from the high-frequency envelope of the coherence pattern and inverts the fluid load pressure components. The phase correction factor is determined based on the preset linear mapping table. ,in The phase pressure coefficient, For the fluid load pressure component, when the fluid load pressure component 2MPa and phase pressure coefficient At a pressure of 0.1 rad / MPa, the logic processing unit injects a phase initial offset of -0.2 rad into the deformation calculation program to compensate for the spurious signal caused by oil film fluctuations, thus ensuring the optical field eigenvector... The absolute deviation is maintained within 0.05, thus achieving a tolerance of 0.5. Feature extraction of m-level spalling defects.

[0046] Example 5: At a thickness of 10 In the deployment and commissioning phase of the m-level strip under test, the logic processing unit establishes the functional mapping relationship between the physical phase delay generated by the lubricating oil film and the fluid load pressure component PL. The system adjusts the rolling oil pressure in the closed cavity through the pressure control valve, gradually changing the pressure value in 0.5MPa increments within the pressure range of 0.5MPa to 10MPa. The coherent light projection module records the phase shift at each pressure point. The logic processing unit performs linear fitting operations on multiple sets of measured data points to determine the phase pressure coefficient k. When the measured pressure increases from 2MPa to 4MPa and the observed phase change is 0.2rad, the logic processing unit determines the phase pressure coefficient k to be 0.1rad / MPa and stores the phase pressure coefficient k in non-volatile memory for use in subsequent operation to perform phase offset compensation based on the real-time value of the fluid load pressure component PL.

[0047] When the system is applied to a production line that includes new roll assemblies, the arbitration module performs a pre-calibration procedure to determine the autocorrelation decision threshold. Under the condition that the rolls are rotating unloaded and their surfaces are uncovered, the arbitration module continuously acquires 10 cycles of optical field feature vectors. And calculate the corresponding autocorrelation function. Extracting the periodic reference amplitude generated by the inherent morphology of the roll surface. The autocorrelation threshold will be determined. Set as reference amplitude 1.5 times that, used to distinguish material defects from inherent equipment morphology features in actual testing, when the autocorrelation function is calculated in real time. When the displacement is equal to the roll circumference The value exceeds the autocorrelation threshold. At that time, the system automatically reduces the weight of the current periodic characteristic component in the hydraulic control command, and uses the physical law of periodic contact in strip rolling to physically suppress the interference signal caused by component wear.

[0048] Example 6: Initialization calibration procedure for different batches of test strips with reflectivity ranging from 60% to 85%, the difference in incident angle between the coherent light projection module and the reference light projection module. Based on the reference roughness of the surface of the strip to be tested The logic processing unit acquires pattern data of the discrete photodetector array at different angular step sizes and calculates the contrast distribution gradient. When the angle of incidence is different The contrast distribution varies within the scanning range of 15° to 45° and has a gradient. When the maximum output amplitude is reached, the system locks the current incident angle difference. As a phase-sensitive angle, the light field characteristic vector The modulation depth is maintained above 0.8.

[0049] When applied to material supports with different surface textures, the logic processing unit executes a texture stripping procedure, driving the material support to rotate the entire roll circumference under no-load conditions. Discrete photoelectric detection array acquires 1024 frames of pixel intensity matrix The benchmark feature matrix is ​​generated by performing time-domain averaging operations by the logic processing unit. During the detection process, the logic processing unit will acquire the pixel intensity matrix in real time. With the benchmark feature matrix Perform pixel-by-pixel subtraction to eliminate constant phase shifts caused by surface roughness of the material support, thus improving the statistical eigenvalues. In response to transient stress changes in the strip under test, the system achieves zero-point adaptive calibration for different deployment environments; when performing compression compensation, the system calculates the control variables of the hydraulic actuator through the logic processing unit. It is based on the light field feature vector Real-time modulus and preset stress reference vector The discrete deviations are used to perform transfer function calculations, specifically by utilizing the longitudinal elastic modulus of the material. Establish a reduction correction model and control variables. Set as a scaling factor When the product of the characteristic deviation and the detected characteristic vector deviation reflecting stress fluctuation reaches 1.2 times the preset alarm threshold, the logic processing unit outputs a pulse width modulation signal to drive the hydraulic servo valve to achieve micron-level dynamic compensation for the roll gap.

[0050] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A system for on-line inspection of semiconductor chips for defects based on optical physical properties, for inspecting a web of material on a material support for defects, characterized in that, The application relates to a coherent light projection module, a reference light projection module, a displacement driving feedback unit, a discrete photoelectric detection array, a material support body and a logic processing unit. The displacement driving feedback unit is used for acquiring a rotating speed pulse signal of a driving end of the measured strip. A logic processing unit is configured to calculate a laser sampling frequency according to the rotation speed pulse signal , and to adjust the pulse trigger timing of the coherent light projection module and the reference light projection module according to the laser sampling frequency , so that the laser sampling frequency is kept synchronized with the rotation speed pulse signal, and the axial displacement of the measured strip in a single pulse period is not greater than a preset characteristic resolution. The coherent light projection module is used for projecting a coherent main light beam to the surface of the measured strip. The reference light projection module is used for projecting incoherent reference light to the projection area of the coherent main light beam on the surface of the measured strip, and the incident angles of the coherent main light beam and the incoherent reference light are different from each other, so as to construct a differential optical path. The discrete photoelectric detection array is used for collecting a composite optical signal containing mechanical vibration noise and micro deformation signals reflected by the surface of the measured strip. The logic processing unit is used for executing a deformation calculation program, and the deformation calculation program comprises the following steps: performing signal differential processing on the composite optical signal, offsetting background phase noise generated by mechanical vibration by using common mode components in the composite optical signal, extracting coherent speckle pattern phase characteristics modulated by micro stress of the measured strip, and determining defect space coordinates of the measured strip based on phase distribution discontinuous points in the coherent speckle pattern phase characteristics.

2. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 1, wherein The logic processing unit is also configured to perform a sampling power synchronization compensation: the logic processing unit determines a peak compensation gain of the driving current based on the reciprocal of the laser sampling frequency , and drives the coherent light projection module to adjust the laser emission power according to the peak compensation gain, so as to maintain the reference light intensity value of the detection area constant when the rotation speed pulse signal changes.

3. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 1, wherein The application further comprises an active excitation calibration module. The active excitation calibration module is used for applying a pulse force disturbance signal of a preset frequency to the material support body; the logic processing unit is used for collecting dynamic phase response characteristics of the discrete photoelectric detection array to the pulse force disturbance signal, inverting surface normal rigidity of the material support body, and updating a preset reference stress characteristic vector.

4. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 3, wherein The logic processing unit is used to execute the weight iteration algorithm to correct the reference stress eigenvector. The calculation formula of the weight iteration algorithm is expressed as: ,in, This is the corrected reference stress eigenvector. The first calculation based on surface normal stiffness Adjustment weights, It is a positive integer. These are reference components pre-stored in the logic processing unit and used to characterize the evolution of the surface morphology of the material support.

5. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 1, wherein The logic processing unit is further used for extracting high-frequency flicker envelopes in the coherent speckle pattern phase characteristics, performing spectral density analysis on the high-frequency flicker envelopes, and calculating fluid load pressure components on the surface of the measured strip.

6. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 5, wherein The logic processing unit introduces the fluid load pressure components as a phase delay correction factor into the deformation calculation program, so as to offset optical path difference signal interference caused by interface fluid fluctuation.

7. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 1, wherein The logic processing unit is further used for monitoring time coherence intensity attenuation ratios of the coherent speckle pattern phase characteristics, and outputting technical parameters representing fatigue states of the surface of the measured strip when the intensity attenuation ratios exceed a preset threshold.

8. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 1, wherein, The discrete photoelectric detection array comprises a plurality of groups of photoelectric sensors arranged in a matrix, and the spatial interval between adjacent two groups of photoelectric sensors is not greater than 0.5 mm.

9. The system for on-line detection of defects in semiconductor chips based on optical physical properties according to claim 1, wherein, The coherent light projection module comprises a distributed feedback laser and a collimating and expanding assembly, and the output end of the distributed feedback laser is connected to a frequency modulation circuit driven by the logic processing unit.

Images