Wafer detection spiral scanning motion system and method based on virtual arc length main shaft
By introducing a fully closed-loop control architecture with a virtual arc length master axis and an extended Kalman filter, the problem of balancing scanning efficiency and accuracy in wafer inspection equipment is solved, achieving efficient and accurate defect detection, which is suitable for semiconductor inspection equipment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TAIYUAN UNIVERSITY OF TECHNOLOGY
- Filing Date
- 2026-01-16
- Publication Date
- 2026-04-21
AI Technical Summary
In existing wafer inspection equipment, traditional XY platform linear grid scanning suffers from low efficiency and mechanical vibration due to frequent start-stop cycles, while spiral scanning control technology suffers from uneven spectral acquisition due to trigger signal jitter and delay, making it impossible to achieve high-efficiency and high-precision defect detection.
A wafer inspection spiral scanning motion system based on a virtual arc length master axis is adopted. Combined with extended Kalman filter and EKF state prediction technology, a fully closed-loop analytical control architecture is constructed. Through the virtual arc length master axis generation module, trajectory calculation engine, state observation module, prediction trigger management module and optomechanical collaborative control module, the precise binding of spectral data and physical coordinates and zero-delay start-up are achieved.
It achieves full-coverage detection of the wafer surface, reduces trigger signal delay to sub-millisecond level, and position jitter to sub-pixel level, improving detection efficiency and accuracy. It is suitable for achieving nanosecond-level operation response with low resource consumption in FPGA.
Smart Images
Figure CN121908860A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ultra-precision motion control and semiconductor inspection equipment technology, specifically to a wafer inspection spiral scanning motion system and method based on a virtual arc length spindle. Background Technology
[0002] With the continuous evolution of semiconductor manufacturing processes, integrated circuits place extremely stringent requirements on the fundamental quality of silicon wafers. Tiny defects such as scratches, water stains, and particles generated during wafer manufacturing directly determine the yield and reliability of the final product. Therefore, developing high-precision, high-efficiency automated inspection technologies has become an urgent need for the semiconductor industry.
[0003] In existing wafer inspection equipment, the choice of scanning method directly restricts the efficiency and accuracy of inspection. Traditional XY platform linear grid scanning has significant disadvantages in practical applications: 1. Linear scanning requires deceleration, stopping, reversing, and re-accelerating at the end of each line. Frequent starts and stops not only significantly increase ineffective motion time, limiting throughput, but also easily excite mechanical vibrations, affecting the signal-to-noise ratio of the spectral signal. 2. Rectangular scanning paths are difficult to adapt to the circular contours of silicon wafers, easily forming detection blind spots or overlaps at wafer edges, leading to missed detections. To overcome these shortcomings, the industry is gradually shifting towards using... The helical scanning method of the rotating platform achieves constant linear velocity scanning through continuous rotation and radial feed. However, existing helical scanning control technologies are mostly based on "time domain" parameterization. For point scanning spectral detection, if the trigger signal has jitter or delay, it will cause uneven distribution of spectral acquisition points in physical space, which will lead to geometric distortion in the reconstructed wafer mapping and make it impossible to accurately trace the defect location through coordinates.
[0004] Therefore, there is an urgent need for an innovative control scheme that can combine the high efficiency of helical scanning with a complete solution to the problems of synchronous triggering accuracy and stability, so as to meet the ultimate pursuit of defect detection in advanced manufacturing processes. Summary of the Invention
[0005] Addressing the problems and shortcomings of existing technologies, this invention aims to resolve the contradiction between scanning efficiency and synchronization accuracy in traditional wafer inspection. To meet the stringent requirements of spatiotemporal alignment and energy consistency in spectral detection, this invention overcomes bottlenecks such as trigger delay, first-round blind zone, and speed sensitivity in existing helical scanning. By introducing a virtual arc length principal axis and EKF state prediction technology, this invention constructs a fully closed-loop analytical control architecture. This architecture eliminates the influence of mechanical disturbances while achieving precise binding of spectral data and physical coordinates and zero-delay startup, providing a high-efficiency, high-precision universal solution for semiconductor defect detection and film thickness analysis.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: A wafer inspection spiral scanning motion system based on a virtual arc length master axis is disclosed. The system operates on a heterogeneous embedded platform and is implemented in a field-programmable gate array (FPGA) using a fully fixed-point logic pipeline. It includes a virtual master axis generation module, a slave axis controller, a trajectory calculation engine, a state observation module, a predictive trigger management module, and an optomechanical co-control module. Specifically: the virtual master axis generation module generates a monotonically changing arc length reference signal, which is transmitted to the trajectory calculation engine and the predictive trigger management module; the trajectory calculation engine calculates the dual-axis position and analytical feedforward in real time based on the arc length and outputs the result to the slave axis controller; the slave axis controller receives the position and analytical feedforward values output by the trajectory calculation engine. The feed rate drives the radial (R-axis) and rotational (θ-axis) motors to move in tandem, and transmits the encoder feedback signal from the motion platform to the state observation module. The state observation module has a built-in extended Kalman filter that estimates the actual arc length and linear velocity in real time and outputs them to the predictive trigger management module and the optomechanical co-control module. The predictive trigger management module calculates the remaining time based on the arc length reference signal output by the virtual spindle generation module and the estimated arc length and linear velocity output by the state observation module, and loads it into the timer built into the FPGA. When the timer returns to zero, it generates an equal arc length trigger signal with sub-control cycle accuracy. The optomechanical co-control module adjusts the PWM duty cycle of the light source in real time based on the estimated linear velocity.
[0007] Furthermore, the optomechanical co-control module adjusts the PWM duty cycle of the light source according to the following formula. : in, Indicates the reference brightness. This represents the estimated linear velocity. Indicates the reference speed.
[0008] A wafer inspection spiral scanning motion method based on a virtual arc length principal axis, based on the aforementioned wafer inspection spiral scanning motion system based on a virtual arc length principal axis, compensates for transmission delay and computation delay through an extended Kalman filter prediction correction mechanism to achieve full coverage inspection of the wafer surface, specifically including the following steps: Step 1. Constructing the virtual arc length master axis: The scanning arc length s is the unique virtual master axis. A virtual arc length that monotonically changes with time is generated within the controller of the wafer inspection spiral scanning motion system. As a reference source for axis movement and triggering events; Step 2. Establish a closed parametric model of the arc length domain: This involves aligning the radial (R-axis) and rotational (R-axis) parameters. θ The axis, acting as a slave axis, establishes a positional mapping relationship with the virtual arc length principal axis through a closed parametric model of the arc length domain. Step 3. Analysis of Feedforward and Closed-Loop Control: Based on the arc length domain closed parameterized model, the combinational logic circuit inside the FPGA is used to calculate the velocity feedforward v_R,ff and acceleration feedforward a_R,ff of the R-axis, and the angular velocity feedforward ω_θ,ff and angular acceleration feedforward α_θ,ff of the θ-axis. The position and feedforward are calculated from the axis controller based on the real-time virtual arc length, and the motor is driven by the feedforward to achieve coordinated motion. Step 4. Extended Kalman Filter State Estimation: Construct a state estimation model that includes the actual arc length. and linear velocity The state vector is used to establish state equations and observation equations. The extended Kalman filter algorithm is used to process the encoder feedback signal collected from the axis controller and output the optimal state estimate in real time to suppress measurement noise and obtain a smooth motion state. Step 5. Predictive equal arc length triggering: Read the optimal state estimate output from Step 4, calculate the remaining arc length and remaining time, and predict the remaining time required to reach the next target trigger arc length; Step 6. Multiphysics Coordination and Error Compensation: This includes error compensation and dynamic velocity-intensity coordination. Error compensation employs spatial domain iterative learning control, establishing a system based on arc length. The error mapping table Map_error(s) is used to perform feedforward compensation for the current control based on the trajectory error of the previous scan cycle, eliminating periodic mechanical disturbances; in the speed-light intensity dynamic coordination process, the real-time linear velocity output by the extended Kalman filter in step 4 is used. The estimated value is input into the excitation source control loop, and the light source intensity is dynamically adjusted to ensure that the excitation energy density within the sampling spot per unit scanning arc length remains constant, eliminating the nonlinear distortion of spectral signal intensity caused by scanning speed fluctuations, and realizing dynamic coordination between speed and light intensity.
[0009] Furthermore, in step 2, the parameterized model for arc-length domain closure is as follows: in, The starting radius, The spacing between the spiral lines, symbol It depends on whether the scanning direction is from the outside in or from the inside out.
[0010] Furthermore, in step 5, the arc length of the next trigger target is calculated. The estimated arc length of the current extended Kalman filter output The difference The current linear velocity output by the extended Kalman filter. The estimated value is used to calculate the remaining time. When the remaining time When the time is less than the control cycle, the remaining time is loaded into the high-precision timer built into the FPGA for countdown. At the end of the countdown, a trigger pulse is output to eliminate the quantization error of the control cycle.
[0011] Furthermore, the extended Kalman filter employs a multi-sensor fusion mechanism, in A tangential accelerometer is installed on the axis motion platform; the state vector is expanded to... And the accelerometer measurements are incorporated into the observation equation. This is the actual arc length. For linear velocity, To improve acceleration, the high-frequency response characteristics of the accelerometer are used to compensate for the high-frequency noise amplification caused by the differential of the grating ruler.
[0012] Furthermore, in step 3, the calculation process of the analytical feedforward is as follows: The velocity feedforward of the R-axis is directly calculated using combinational logic circuits. and acceleration feedforward : in The target linear velocity of the virtual spindle. Indicates the spacing between the spiral lines. This represents the current radial position calculated by the arc-length domain closed parameterized model.
[0013] Furthermore, in step 6, the current control is fed forward to compensate based on the trajectory error of the previous scan cycle. For the first scan cycle, the error mapping table Map_error(s) is initialized to all zero values. During the first scan, the trajectory tracking error corresponding to each arc length position is recorded. After the first scan is completed, the error mapping table Map_error(s) is updated.
[0014] Compared with the prior art, the present invention has the following significant advantages: Through the EKF prediction correction mechanism, this invention can compensate for the transmission and computation delays of the wafer inspection spiral scanning motion system, reducing the average delay of the trigger signal from milliseconds to sub-milliseconds in the traditional DPLL scheme, and reducing trigger position jitter to the sub-pixel level, ensuring strict equidistance of spectral sampling points on the wafer surface. This invention does not rely on the physical locking process of a phase-locked loop; the wafer inspection spiral scanning motion system can generate an effective trigger based on the virtual spindle state at startup, completely eliminating the problem of invalid data in the first rotation in traditional schemes. This achieves 100% full coverage inspection of the wafer surface, significantly improving the detection efficiency and real-time performance of the wafer inspection spiral scanning motion system during startup. The control architecture based on "virtual arc length" achieves natural spatiotemporal decoupling. Even if the mechanical platform is disturbed, causing instantaneous speed fluctuations, the trigger signal remains strictly locked at the physical position, exhibiting only a slight temporal offset, ensuring the accuracy of the scanning position. The closed analytical feedforward and extended Kalman filter (EKF) proposed in this invention, after optimization, is suitable for implementation in FPGA using a fully fixed-point pipelined approach. It has extremely low logic resource occupancy and can achieve nanosecond-level computational response without the need for expensive high-end processors. Attached Figure Description
[0015] Figure 1 This is the overall architecture diagram of the system of the present invention; Figure 2 A schematic diagram of the closed parametric geometry principle of the equidistant spiral trajectory in the arc length domain of this invention; Figure 3 This is a block diagram of the signal flow and control topology of the FPGA logic module of this invention; Figure 4 This is a flowchart of the state estimation and prediction triggering process based on extended Kalman filtering in this invention; Figure 5 This is a schematic diagram comparing the predictive triggering timing of the present invention with the traditional DPLL triggering timing; Figure 6 This is a schematic diagram of the principle of spatial domain iterative learning and light intensity coordinated control of the present invention. Detailed Implementation
[0016] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. The present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the present invention.
[0017] like Figures 1-6As shown, this invention discloses a wafer inspection spiral scanning motion system based on a virtual arc length master axis. The wafer inspection spiral scanning motion system runs on a heterogeneous embedded platform and is implemented in a field-programmable gate array (FPGA) using a fully fixed-point logic pipeline. It includes a virtual master axis generation module, a slave axis controller, a trajectory calculation engine, a state observation module, a predictive trigger management module, and an optomechanical cooperative control module. Specifically: the virtual master axis generation module generates a monotonically changing arc length reference signal, which is transmitted to the trajectory calculation engine and the predictive trigger management module; the trajectory calculation engine calculates the dual-axis position and analytical feedforward in real time based on the arc length and outputs it to the slave axis controller; the slave axis controller receives the position and analytical feedforward output from the trajectory calculation engine, drives the radial (R-axis) and rotational (θ-axis) motors to move collaboratively, and transmits the encoder feedback signal from the motion platform to the state observation module; the state observation module incorporates an extended Kalman filter to estimate the actual arc length and linear velocity in real time, and outputs the arc length estimate... and linear velocity estimates The output is sent to the predictive trigger management module and the optomechanical co-control module. The predictive trigger management module calculates the remaining time based on the arc length reference signal output by the virtual spindle generation module and the estimated arc length and linear velocity output by the state observation module. This time is then loaded into the timer built into the FPGA. When the timer returns to zero, a sub-control cycle precision equal arc length trigger signal is generated. The optomechanical co-control module receives the estimated linear velocity output by the state observation module. The PWM duty cycle of the light source is adjusted in real time based on the estimated linear velocity. The correlation between the various modules of the wafer inspection spiral scanning motion system is as follows: Figure 3 As shown, the data flow is as follows: Virtual spindle generation module → trajectory calculation engine → slave axis controller → status observation module → prediction trigger management module / optical-mechanical collaborative control module.
[0018] The optomechanical co-control module adjusts the PWM duty cycle of the light source according to the following formula. : in, Indicates the reference brightness. This represents the estimated linear velocity. Indicates the reference speed.
[0019] This invention also discloses a wafer inspection spiral scanning motion method based on a virtual arc length principal axis. Based on the aforementioned wafer inspection spiral scanning motion system based on a virtual arc length principal axis, the method compensates for transmission delay and computation delay through an extended Kalman filter prediction correction mechanism to achieve full coverage inspection of the wafer surface. Specifically, the method includes the following steps: Step 1. Constructing the virtual arc length master axis: The scanning arc length s is the unique virtual master axis. A virtual arc length that monotonically changes with time is generated within the controller of the wafer inspection spiral scanning motion system. Serves as a reference source for axis movement and triggering events.
[0020] Step 2. Establish a closed parametric model of the arc length domain: This involves aligning the radial (R-axis) and rotational (R-axis) parameters. θ The slave axis, acting as a slave axis, establishes a position mapping relationship with the virtual arc-length master axis through an arc-length domain closed parameterized model. The target position of each slave axis is directly calculated from the current value s(k) of the virtual arc-length master axis using analytical formulas, forming a master-slave synchronized electronic gear coupling structure. The arc-length domain closed parameterized model directly describes the position of each axis. With arc length The nonlinear mapping relationship eliminates the need for intermediate transformations via time parameters, thus ensuring the absolute closure of the physical trajectory. The arc-length domain closure parameterization model is as follows: in, The starting radius, The spacing between the spiral lines, symbol It depends on whether the scanning direction is from the outside in or from the inside out.
[0021] Step 3. Analyzing Feedforward and Closed-Loop Control: Based on the arc-length domain closed-loop parameterized model, the combinational logic circuits inside the FPGA are used to calculate the velocity feedforward v_R,ff and acceleration feedforward a_R,ff for the R-axis, and the angular velocity feedforward ω_θ,ff and angular acceleration feedforward α_θ,ff for the θ-axis. No numerical differentiation is required here. The slave axis controllers calculate the position and feedforward based on the real-time virtual arc length, send the position to the position loop controller, and superimpose the velocity and acceleration feedforwards onto the control inputs of the velocity loop and current loop, respectively, to achieve coordinated motion of the slave axis motors. The coordinated motion is achieved by combining the feedforwards to drive the motors. The calculation process of the analytical feedforward is as follows: The velocity feedforward of the R-axis is directly calculated using combinational logic circuits. and acceleration feedforward : in The target linear velocity of the virtual spindle. Indicates the spacing between the spiral lines. This represents the current radial position calculated by the arc-length domain closed parameterized model.
[0022] Step 4. Extended Kalman Filter State Estimation: Construct a state estimation model that includes the actual arc length. and linear velocity The state vector is used to establish the state equations and observation equations of the wafer inspection helical scanning motion system. An extended Kalman filter algorithm is used to process the encoder feedback signal acquired from the axis controller. The encoder feedback signal originates from a high-precision grating ruler encoder mounted on the R-axis and θ-axis motion platforms. The encoder position feedback value is transmitted to the FPGA in real time via a high-speed serial interface for processing, and the optimal state estimate is output in real time to suppress measurement noise and obtain a smooth motion state. Under the assumption that the noise and observation noise follow a Gaussian distribution, the state estimate output by the extended Kalman filter is the optimal estimate in the sense of conditional mean.
[0023] Step 5. Predictive Equal Arc Length Trigger: Read the optimal state estimate output from Step 4, calculate the remaining arc length and remaining time, predict the remaining time required to reach the next target trigger arc length; calculate the next target arc length. The estimated arc length of the current extended Kalman filter output The difference The current linear velocity output by the extended Kalman filter. The estimated value is used to calculate the remaining time. When the remaining time When the time is less than the control cycle, the remaining time is loaded into the high-precision timer built into the FPGA for countdown. At the end of the countdown, a trigger pulse is output to eliminate the quantization error of the control cycle.
[0024] The extended Kalman filter employs a multi-sensor fusion mechanism, in A tangential accelerometer is installed on the axis motion platform; the state vector is expanded to... And the accelerometer measurements are incorporated into the observation equation. This is the actual arc length. For linear velocity, To improve acceleration, the high-frequency response characteristics of the accelerometer are used to compensate for the high-frequency noise amplification caused by the differential of the grating ruler.
[0025] Step 6. Multiphysics Coordination and Error Compensation: This includes error compensation and dynamic velocity-intensity coordination. Error compensation employs spatial domain iterative learning control, establishing a system based on arc length. The error mapping table Map_error(s) is used to perform feedforward compensation on the current control based on the trajectory error of the previous scan cycle, eliminating periodic mechanical disturbances. For the first scan cycle, the error mapping table Map_error(s) is initialized to all zeros. During the first scan, the trajectory tracking error corresponding to each arc length position is recorded. After the first scan is completed, the error mapping table Map_error(s) is updated. Spatial domain iterative learning begins from the second scan cycle. Spatial domain iterative learning control: records the trajectory tracking error during the previous complete scan process. Establish the error with respect to arc length Spatial mapping table During the current scanning process, based on the real-time virtual arc length Index or interpolate the spatial mapping table to obtain the corresponding compensation amount and add it to the slave axis control loop to suppress position-related periodic mechanical disturbances.
[0026] In the dynamic coordination process of velocity and light intensity, the real-time linear velocity output by the extended Kalman filter in step 4 is used. The estimated value is input into the excitation source control loop, dynamically adjusting the source intensity to ensure a constant excitation energy density within the sampling spot per unit scanning arc length. The real-time linear velocity estimate output by the extended Kalman filter is then used. The control system of the light source transmits the data to establish a dynamic correlation function between the light source intensity and the linear velocity. When linear velocity fluctuations are detected, the driving current of the light source or the PWM duty cycle is adjusted in real time to keep the exposure energy density within a unit scanning arc length constant, thereby eliminating the nonlinear distortion of spectral signal intensity caused by scanning speed fluctuations and achieving dynamic coordination between speed and light intensity.
[0027] Based on the radial position calculated in real time Online estimation of the changing moment of inertia and resonant frequency of the motion platform The center frequency of the notch filter or input shaper in the control loop is dynamically adjusted to track and suppress mechanical resonance that varies with the radius in real time.
[0028] Example: Synchronous control system for a 300mm wafer bright / dark field defect detection equipment: This embodiment demonstrates a precision motion control system for detecting surface defects on 300mm semiconductor wafers. The helical scanning motion system is based on a high-performance embedded controller (integrating a dual-core ARM processor and FPGA), which implements the virtual arc length spindle control and predictive triggering functions described in this invention.
[0029] like Figure 1 As shown, the wafer inspection spiral scanning motion system mainly includes the following physical components: Motion platform: Air-float type Rotary worktable.
[0030] Axis (rotary axis): Direct drive torque motor, equipped with a high-precision linear scale, with a resolution superior to [previous value]. .
[0031] R-axis (radial axis): Driven by a voice coil motor or linear motor, equipped with a grating ruler with nanometer-level resolution.
[0032] Detection system: High-speed fiber optic spectrometer with a sampling frequency of up to 100 kHz, and the sampling spot diameter corresponds to the object-space accuracy.
[0033] Light source system: The light source is excited by adjustable intensity, and the intensity is controlled by PWM or analog voltage generated by FPGA.
[0034] Scan parameter settings: Target constant linear velocity .
[0035] Helical scan radial step .
[0036] Scan range: starting radius (Wafer edge), termination radius (Wafer center), using an outside-to-in scanning method.
[0037] The internal logic clock of the field-programmable gate array (FPGA) controller operates at 100 MHz (control cycle). The control loop update frequency is 20 kHz. The connection relationships of the main functional modules inside the FPGA are as follows: Figure 3 As shown, the specific implementation is as follows: (1) Virtual arc length spindle generation module In the FPGA, allocate a 64-bit fixed-point accumulator (32-bit integer, 32-bit decimal) as the virtual spindle position. The register.
[0038] The accumulator is used in each servo cycle. It is unconditionally monotonically increasing and serves as the sole spatiotemporal reference for the entire wafer inspection spiral scanning motion system, unaffected by any mechanical disturbances.
[0039] (2) Trajectory calculation engine A fully parallel pipeline architecture is employed to calculate the target position command. To avoid the latency of complex square root calculations, the CORDIC IP core (coordinate rotation digital computer algorithm) inside the FPGA is used for real-time calculation. : in, For the first Each control cycle Axis target position command, for Axis target position command, This is an intermediate result from the square root operation. The computation delay of the arc length domain closed trajectory calculation module is fixed at 20 clock cycles (200). ), for 50 The control loop can be ignored.
[0040] (3) Analysis of feedforward and spatial domain ILC compensation module R-axis and Shaft velocity feedforward With acceleration feedforward As shown in formulas (4) to (7): The above formula is directly implemented on the FPGA using the DSP48 multiplier unit. Building upon this, this embodiment introduces spatial domain iterative learning control: an error correction table is pre-stored in the FPGA's Block RAM. The index is the arc length. The high-order address.
[0041] Wafer inspection spiral scanning motion system reads current Corresponding historical error value and multiplied by the learning law gain This, when added to the current, can effectively compensate for one rotation of the air-bearing turntable. The inherent periodic flatness error in ).
[0042] Extended Kalman Filter (EKF) and Predictive Triggering Process as follows Figure 4 As shown.
[0043] Step A: State Prediction in, and These are the prior estimates (predictions) of the arc length and linear velocity in the k-th period, respectively. and For the first The posterior estimate of the period.
[0044] Step B: Observation Update collection Shaft encoder position Calculate the observation residuals .
[0045] Calculate Kalman gain Status Update: Through this step, the wafer inspection spiral scanning motion system achieves a smooth actual physical arc length with minimal hysteresis. and actual linear velocity .
[0046] Step C: Predictive Hardware Triggering The FPGA has a built-in trigger management state machine. Let the current target trigger arc length be... .
[0047] Calculate the remaining arc length from the next trigger point: .
[0048] Predicting remaining time: when When (it is about to be triggered in the next control cycle), start the nanosecond-level hardware countdown timer.
[0049] Countdown value: This is the FPGA logic clock frequency. The moment the timer returns to zero, the FPGA digital I / O port flips, outputting a trigger pulse to the external trigger input of the spectral detector.
[0050] Meanwhile, the FPGA internally writes the current virtual arc length coordinates into a FIFO buffer, waiting to align them with the spectral data uploaded by the spectrometer.
[0051] Step D: Speed-Intensity Coordinated Control The FPGA will output the real-time speed estimate from the EKF. Simultaneously, the signal is sent to the lighting control module.
[0052] Set the reference brightness Corresponding reference speed .
[0053] The formula for adjusting the real-time PWM duty cycle is: When the mechanical system decelerates momentarily due to minute friction, the intensity of the excitation light automatically decreases to prevent oversaturation or drift of the spectral integrated intensity caused by prolonged irradiation time, thus ensuring the comparability of the data.
[0054] This embodiment demonstrates that by introducing a virtual arc length spindle and EKF prediction technology, and supplementing it with spatial domain ILC and light intensity coordinated control, sub-microsecond synchronization accuracy can be achieved on conventional industrial hardware.
[0055] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. A wafer inspection spiral scanning motion system based on a virtual arc length master axis, characterized in that, The wafer inspection spiral scanning motion system runs on a heterogeneous embedded platform and is implemented in a field-programmable gate array (FPGA) using a fully fixed-point logic pipeline. It includes a virtual spindle generation module, a slave axis controller, a trajectory calculation engine, a state observation module, a prediction trigger management module, and an optomechanical collaborative control module, wherein: The virtual spindle generation module generates a monotonically changing arc length reference signal, which is then transmitted to the trajectory calculation engine and the prediction trigger management module. The trajectory calculation engine calculates the dual-axis position and analyzes the feedforward amount in real time based on the arc length, and outputs it to the slave axis controller; The axis controller receives the position and analytical feedforward output from the trajectory calculation engine, drives the radial R-axis and the rotational θ-axis motors to move in coordination, and transmits the encoder feedback signal on the motion platform to the state observation module. The state observation module has a built-in extended Kalman filter that estimates the actual arc length and linear velocity in real time and outputs them to the prediction trigger management module and the optomechanical collaborative control module. The predictive trigger management module calculates the remaining time based on the arc length reference signal output by the virtual spindle generation module and the estimated values of arc length and linear velocity output by the state observation module, and loads it into the timer built into the field programmable gate array (FPGA). When the timer returns to zero, it generates an equal arc length trigger signal with sub-control cycle precision. The optomechanical co-control module adjusts the PWM duty cycle of the light source in real time based on the estimated linear velocity.
2. The wafer inspection spiral scanning motion system based on a virtual arc length principal axis according to claim 1, characterized in that, The optomechanical co-control module adjusts the PWM duty cycle of the light source according to the following formula. : in, Indicates the reference brightness. This represents the estimated linear velocity. Indicates the reference speed.
3. A wafer inspection spiral scanning motion method based on a virtual arc length principal axis, based on the wafer inspection spiral scanning motion system based on a virtual arc length principal axis as described in claim 1 or 2, characterized in that, By compensating for transmission and computation delays through an extended Kalman filter prediction correction mechanism, full-coverage detection of the wafer surface is achieved. This includes the following steps: Step 1. Constructing the virtual arc length master axis: The scanning arc length s is the unique virtual master axis. A virtual arc length that monotonically changes with time is generated within the controller of the wafer inspection spiral scanning motion system. As a reference source for axis movement and triggering events; Step 2. Establish a closed parametric model of the arc length domain: This involves aligning the radial (R-axis) and rotational (R-axis) parameters. θ The axis, acting as a slave axis, establishes a positional mapping relationship with the virtual arc length principal axis through a closed parametric model of the arc length domain. Step 3. Analysis of Feedforward and Closed-Loop Control: Based on the arc length domain closed parameterized model, the combinational logic circuit inside the FPGA is used to calculate the velocity feedforward v_R,ff and acceleration feedforward a_R,ff of the R-axis, and the angular velocity feedforward ω_θ,ff and angular acceleration feedforward α_θ,ff of the θ-axis. The position and feedforward are calculated from the axis controller based on the real-time virtual arc length, and the motor is driven by the feedforward to achieve coordinated motion. Step 4. Extended Kalman Filter State Estimation: Construct a state estimation model that includes the actual arc length. and linear velocity The state vector is used to establish state equations and observation equations. The extended Kalman filter algorithm is used to process the encoder feedback signal collected from the axis controller and output the optimal state estimate in real time to suppress measurement noise and obtain a smooth motion state. Step 5. Predictive equal arc length triggering: Read the optimal state estimate output from Step 4, calculate the remaining arc length and remaining time, and predict the remaining time required to reach the next target trigger arc length; Step 6. Multiphysics Coordination and Error Compensation: This includes error compensation and dynamic velocity-intensity coordination. Error compensation employs spatial domain iterative learning control, establishing a system based on arc length. The error mapping table Map_error(s) is used to perform feedforward compensation for the current control based on the trajectory error of the previous scan cycle, eliminating periodic mechanical disturbances; in the speed-light intensity dynamic coordination process, the real-time linear velocity output by the extended Kalman filter in step 4 is used. The estimated value is input into the excitation source control loop, and the light source intensity is dynamically adjusted to ensure that the excitation energy density within the sampling spot per unit scanning arc length remains constant, eliminating the nonlinear distortion of spectral signal intensity caused by scanning speed fluctuations, and realizing dynamic coordination between speed and light intensity.
4. The wafer inspection spiral scanning motion method based on a virtual arc length principal axis according to claim 3, characterized in that, In step 2, the arc-length domain closure parameterization model is as follows: in, The starting radius, The spacing between the spiral lines, symbol It depends on whether the scanning direction is from the outside in or from the inside out.
5. The wafer inspection spiral scanning motion method based on a virtual arc length principal axis according to claim 3, characterized in that, In step 5, the arc length of the next trigger target is calculated. The estimated arc length of the current extended Kalman filter output The difference ; Current linear velocity output using the extended Kalman filter The estimated value is used to calculate the remaining time. ; When the remaining time When the time is less than the control cycle, the remaining time is loaded into the high-precision timer built into the FPGA for countdown. At the end of the countdown, a trigger pulse is output to eliminate the quantization error of the control cycle.
6. The wafer inspection spiral scanning motion method based on a virtual arc length principal axis according to claim 5, characterized in that, The extended Kalman filter employs a multi-sensor fusion mechanism. A tangential accelerometer is installed on the axis motion platform; Expand the state vector to And the accelerometer measurements are incorporated into the observation equation. This is the actual arc length. Linear velocity, To improve acceleration, the high-frequency response characteristics of the accelerometer are used to compensate for the high-frequency noise amplification caused by the differential of the grating ruler.
7. The wafer inspection spiral scanning motion method based on a virtual arc length principal axis according to claim 3, characterized in that, In step 3, the calculation process of the analytical feedforward is as follows: The velocity feedforward of the R-axis is directly calculated using combinational logic circuits. and acceleration feedforward : in The target linear velocity of the virtual spindle. Indicates the spacing between the spiral lines. This represents the current radial position calculated by the arc-length domain closed parameterized model.
8. The wafer inspection spiral scanning motion method based on a virtual arc length principal axis according to claim 3, characterized in that, In step 6, the current control is fed forward to compensate for the trajectory error of the previous scan cycle. For the first scan cycle, the error mapping table Map_error(s) is initialized to all zero values. During the first scan, the trajectory tracking error corresponding to each arc length position is recorded. After the first scan is completed, the error mapping table Map_error(s) is updated.