Wafer edge morphology parameter determination device and method, control system and wafer series

Abstract

The embodiment of the invention discloses a wafer edge morphology parameter determination device and method, a control system and a wafer series, and the device comprises a memory which is configured to store a group of discrete data points representing the edge contour of a wafer; and a processor in communication connection with the memory, the processor being configured to calculate an initial curvature value of the set of discrete data points and calculate stable curvature values of a plurality of target data points in the set of discrete data points, the stable curvature value is determined based on a weighted combination of an initial curvature value of the target data point and initial curvature values of a plurality of adjacent data points of the target data point, and a weight coefficient for the weighted combination is configured to decrease as a distance between the adjacent data points and the target data point increases, the processor is further configured to traverse the stable curvature values of the plurality of target data points to identify a maximum stable curvature value, and take the maximum stable curvature value as an FER parameter value of the wafer.

Description

Technical Field

[0001] This disclosure relates to the field of semiconductor wafer manufacturing, and more particularly to a wafer edge morphology parameter determination apparatus, method, control system, and wafer series. Background Technology

[0002] In modern semiconductor manufacturing, the "Front Edge Radius (FER)," which characterizes the edge morphology of a wafer, has evolved from a geometric detail into a core indicator determining yield. As process nodes approach 14 nm and even more advanced processes, the depth of focus in lithography is compressed to the nanometer scale. Any deviation of FER from the design value will cause local defocusing at the edge, leading to inaccuracies in critical dimensions, coverage errors, and subsequent etching / deposition defects, ultimately resulting in chip scrap. The stringent requirements for edge integrity in immersion lithography make FER control the primary aspect of yield management.

[0003] Currently, the industry commonly uses CCD cameras to acquire discrete points at the wafer edge, then calculates the curvature point by point, and uses the point with the maximum curvature as the FER output. However, curvature calculation involves second-order differentiation, and high-frequency errors such as shot noise, mechanical vibration, thermal drift, and imaging distortion from optical sensors are significantly amplified at this stage, resulting in large random fluctuations in the FER value, a large standard deviation for repeated measurements, and difficulty for process engineers to distinguish between real variations and measurement artifacts.

[0004] Due to the instability in FER caused by the aforementioned noise, the existing process can only passively detect it after the chamfering process is completed. Once the exceedance is detected, the process must be scrapped or reworked, resulting in high costs and long cycles. At the same time, unreliable measurement data makes process engineers hesitant to adjust upstream process parameters. This vicious cycle seriously hinders the implementation of advanced process control. Summary of the Invention

[0005] To address the aforementioned technical problems, this disclosure aims to provide a wafer edge topography parameter determination apparatus, method, control system, and wafer series, thereby resolving the technical issues of unstable, poor repeatability, and low accuracy in the calculation results of semiconductor wafer edge topography parameter FER.

[0006] The technical solution disclosed herein is implemented as follows: In a first aspect, embodiments of this disclosure provide a wafer edge morphology parameter determination apparatus, the apparatus comprising: The memory is configured to store a set of discrete data points representing the edge contours of the wafer; and A processor, communicatively connected to the memory, is configured to calculate initial curvature values ​​for the set of discrete data points and stable curvature values ​​for a plurality of target data points within the set of discrete data points. The stable curvature values ​​are determined based on a weighted combination of the initial curvature values ​​of the target data points and the initial curvature values ​​of a plurality of neighboring data points. The weighting coefficients used for the weighted combination are configured to decrease as the distance between the neighboring data points and the target data points increases. The processor is further configured to iterate through the stable curvature values ​​of the plurality of target data points to identify the maximum stable curvature value and use the maximum stable curvature value as the FER parameter value of the wafer.

[0007] Secondly, embodiments of this disclosure provide a method for determining wafer edge topography parameters, the method comprising: Store a set of discrete data points representing the edge contour of the wafer; Calculate the initial curvature values ​​of the set of discrete data points; Calculate stable curvature values ​​for multiple target data points in the set of discrete data points, wherein the stable curvature values ​​are determined based on a weighted combination of the initial curvature values ​​of the target data points and the initial curvature values ​​of multiple neighboring data points, and the weighting coefficients used for the weighted combination are configured to decrease as the distance between the neighboring data points and the target data point increases; and The stable curvature values ​​of the multiple target data points are traversed to identify the maximum stable curvature value, and the maximum stable curvature value is used as the FER parameter value of the wafer.

[0008] Thirdly, embodiments of this disclosure provide a wafer edge topography control system, the system comprising: The wafer edge morphology parameter determination device according to the first aspect; and A process control device is communicatively connected to the processor of the wafer edge topography parameter determination device. The process control device is configured to generate and send adjustment instructions for adjusting at least one process parameter of the wafer manufacturing tool based on the deviation between the FER parameter value determined by the device and the target value.

[0009] Fourthly, embodiments of this disclosure provide a wafer series, each wafer in which is produced by means of a wafer edge morphology control system according to the third aspect, wherein the coefficient of variation of the FER parameter values ​​of all wafers in the wafer series is less than 2%.

[0010] This disclosure provides an apparatus, method, control system, and wafer series for determining wafer edge topography parameters. First, by introducing a distance-attenuation weighted smoothing algorithm, the accuracy, repeatability, and robustness to noise in FER parameter measurement are significantly improved. Second, by implementing feedforward active process control, the limitations of traditional reactive control modes are overcome, effectively compensating for process variations and reducing wafer scrap or rework caused by unqualified edge topography, thereby lowering production costs. Finally, the overall consistency and stability of semiconductor wafer edge topography are improved, ensuring compliance with the stringent requirements of key processes such as photolithography at advanced technology nodes, ultimately contributing to improved overall chip manufacturing yield. Attached Figure Description

[0011] Figure 1 A schematic diagram of a wafer edge topography parameter determination apparatus according to an embodiment of the present disclosure is shown, illustrating the connection relationship between the memory and the processor and their interaction with the wafer.

[0012] Figure 2 A schematic diagram of the internal structure of a processor according to an embodiment of the present disclosure is shown, further illustrating the role of the data preprocessing unit in processing discrete data points.

[0013] Figure 3 A flowchart of a method for determining wafer edge morphology parameters according to an embodiment of the present disclosure is shown, which sequentially illustrates the steps of acquiring discrete data points, calculating initial curvature values, calculating stable curvature values, and identifying the maximum stable curvature value as the FER parameter value.

[0014] Figure 4 A schematic diagram of the wafer edge topography control system according to an embodiment of the present disclosure is shown, illustrating the communication connection between the wafer edge topography parameter determination device and the process control device, as well as the control relationship between the process control device and the wafer manufacturing tool. Detailed Implementation

[0015] The technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0016] Current processes typically involve first using a CCD camera to capture discrete points at the wafer edge, then calculating the curvature point by point and taking the maximum value as the FER (Functional Ergonomics). However, curvature calculation is extremely sensitive to noise. High-frequency errors such as CCD shot noise and mechanical vibration are amplified in the second-order derivative, resulting in random fluctuations and poor repeatability of the FER value. Process engineers find it difficult to distinguish between true variations and measurement artifacts, and can only passively detect them after the chamfering is completed. If the values ​​exceed the limits, the wafer is scrapped and reworked, resulting in high costs and long cycles, making advanced process control impossible.

[0017] To address this, this disclosure proposes an acquisition method that incorporates a "distance attenuation weighted" smoothing algorithm after data acquisition. This method suppresses noise at the source and moves the control node forward, thereby achieving stable FER measurement and real-time feedforward compensation.

[0018] See Figure 1 This disclosure provides a wafer edge topography parameter determination device 100, which may include a memory 120 and a processor 140.

[0019] The memory 120 is configured to store a set of discrete data points representing the edge contour of the wafer W.

[0020] Processor 140 is communicatively connected to memory 120. Processor 140 is configured to calculate the initial curvature values ​​of a set of discrete data points and the stable curvature values ​​of multiple target data points in the set of discrete data points. The stable curvature values ​​are determined based on a weighted combination of the initial curvature values ​​of the target data points and the initial curvature values ​​of multiple neighboring data points of the target data points. The weighting coefficients used for the weighting combination are configured to decrease as the distance between the neighboring data points and the target data points increases. Processor 140 is also configured to traverse the stable curvature values ​​of the multiple target data points to identify the maximum stable curvature value and use the maximum stable curvature value as the FER parameter value of wafer W.

[0021] Specifically, for each target data point Using the initial curvature values ​​of itself and its neighboring data points that have already been calculated. Calculate a self-stable and robust curvature value. The calculation formula is as follows:

[0022] in, The range of the neighborhood is defined, that is, within Take from both the left and right Calculate each point. It is the normalization coefficient, and its calculation method is as follows: This is to ensure that the curvature value after weighted summation has the correct dimensions and physical meaning.

[0023] The core idea of ​​this disclosure is embodied in the weighting coefficient. In terms of design, it differs from a simple moving average (where all...) (All are equal), this disclosure requires weighting coefficients. It must satisfy a key property: it depends on the neighboring points and the current target point. The distance between them (by index difference) The distance attenuation decreases monotonically with the increase of noise. This "distance attenuation weighting" concept is the technical advantage of this disclosure, which is able to effectively smooth noise while preserving the true peak curvature (i.e., FER characteristics) to the maximum extent.

[0024] Because noise is spatially distributed at high frequency and randomly, data points near the target data point are more affected by noise. After weighting by distance, the contribution of these high-noise samples to the final curvature is significantly weakened, thus significantly suppressing overall fluctuations. The true curvature peak (FER feature) usually exhibits a local continuous convexity, with its energy concentrated in the target data point and its immediate vicinity. Distance attenuation weighting locks the calculation focus in the "signal region," preventing the estimation of the peak position from being skewed by distant noise. Traditional methods apply equal weights or simple averages to all neighboring points, which is equivalent to superimposing noise and signal equally, causing the FER parameter value to fluctuate with noise. However, this scheme achieves statistical filtering with "high fidelity for near points and low interference for far points" through a weighting function, preserving the key curvature peaks that determine the process window while significantly reducing the standard deviation of repeated measurements.

[0025] See in some examples Figure 2 The processor 140 may include a data preprocessing unit 142, which is configured to filter a set of discrete data points to suppress noise before calculating the initial curvature value.

[0026] Before entering the core curvature calculation, the original discrete data points are processed by the data preprocessing unit 142. , The processing of the data aims to effectively filter out high-frequency random noise while preserving, to the maximum extent possible, the true, high-frequency features of the edge profiles crucial for determining the FER. This is a critical trade-off: excessive smoothing can blur or even eliminate the curvature peaks we aim to measure, while insufficient smoothing fails to effectively suppress noise. Therefore, an "edge-preserving" filter capable of distinguishing between signal and noise must be used, rather than a simple mean filter. The data preprocessing unit 142 filters discrete data points before curvature calculation, essentially removing the "amplifier" of noise in advance. After preprocessing, the fluctuations in FER parameter values ​​are greatly compressed, ensuring that subsequent "maximum curvature localization" will not be misjudged due to over-smoothing. Because preprocessing significantly reduces the input noise variance of curvature calculation, the sensitivity of the wafer edge morphology parameter determination device 100 to differences in discrete data points stored in the memory 120, vibration environment, or temperature drift is simultaneously reduced.

[0027] In some examples, the data preprocessing unit 142 may be configured to apply at least one edge-preserving filter selected from Savitzky-Golay filters, bilateral filters, and wavelet denoising algorithms.

[0028] The Savitzky-Golay filter is a powerful tool for smoothing and differentiating in the time or spatial domains. Its core idea is to perform local least-squares fitting of data points with a low-order polynomial within a moving window. The smoothed value at the center of the window is then replaced with the value of the fitted polynomial at that point. Mathematically, this is equivalent to performing a convolution operation on the original data; the coefficients of the convolution kernel can be pre-calculated by solving a system of linear equations based on the Vandermonde matrix. A key advantage of the Savitzky-Golay filter (using second- or higher-order polynomials) compared to simple moving averages (i.e., zero-order polynomial fitting) is its superior preservation of the height, width, and shape of peaks in the signal, which is crucial for retaining the curvature peaks corresponding to the FER parameters.

[0029] A bilateral filter is an advanced nonlinear filter that considers two dimensions of information simultaneously when calculating the smoothed value of a point: spatial proximity and numerical similarity. Its weights are determined by the product of two kernel functions: a "domain kernel" (usually a Gaussian function) that takes spatial distance as the variable, and a "range kernel" (also usually a Gaussian function) that takes the difference in pixel values ​​(coordinate values ​​in this scenario) as the variable. This dual-weighting mechanism allows the filter to effectively smooth spatially adjacent and numerically similar points (i.e., flat regions), while assigning very low weights to spatially adjacent but numerically different points (i.e., edges), thus smoothing noise while preserving the sharp edges of the contour.

[0030] Wavelet denoising is a powerful technique for separating signal and noise in the transform domain. First, a multi-level wavelet transform is performed on a one-dimensional contour signal using a specific wavelet basis function. This transform decomposes the signal into coefficients at different scales (frequency), including a low-frequency approximation coefficient and multiple high-frequency detail coefficients. The main energy of the signal is usually concentrated in a few coefficients with large amplitudes. Since noise typically manifests as small-amplitude coefficients distributed across all scales, a threshold can be set to process the detail coefficients at each level. Common processing methods include "hard thresholding" (setting coefficients with absolute values ​​less than the threshold directly to zero) and "soft thresholding" (after setting them to zero, further shrinking coefficients greater than the threshold towards zero by a threshold amount). Finally, an inverse wavelet transform is performed using the thresholded coefficients (including the unprocessed approximation coefficients and the processed detail coefficients) to reconstruct the denoised signal. The advantage of wavelet denoising lies in its multi-resolution analysis capability, which can very effectively remove noise at different frequencies while preserving sharp local features such as FER peaks.

[0031] See in some examples Figure 2The wafer edge morphology parameter determination device 100 may further include a data acquisition unit 160 for acquiring the set of discrete data points. The core task of the data acquisition unit 160 is to physically interact with the wafer W, precisely scanning the edge region of the wafer W in a non-contact manner. It typically includes a high-precision wafer stage for supporting, rotating, and precisely positioning the wafer W, a multi-axis actuator driven by piezoelectric ceramics or a servo motor, and an advanced optical measurement head. This optical measurement head can be a standard imaging system equipped with a high-pixel-density CCD or CMOS sensor. The goal of the data acquisition unit 160 is to acquire a sequence of discrete data points, P1, P2, ..., P3, that can accurately represent the edge contour of the wafer W. m Each point is represented by its coordinates ( , (Definition). To achieve this goal, a variety of advanced optical measurement techniques can be employed, each with its own emphasis on accuracy, speed, and noise immunity. The data acquisition unit 160 may include at least one of the following measurement devices: a confocal microscope, a white light interferometer, and a laser scanner.

[0032] Confocal microscopy utilizes the confocal principle to achieve superior optical sectioning capabilities and noise suppression. Its core lies in a spatial pinhole strategically placed on the conjugate focal plane in front of the detector. When a laser beam is focused on the focal plane of the sample, light reflected or emitted from that plane can pass smoothly through the pinhole and be received by the detector. However, stray light from above or below the focal plane, whose focal point is off-center from the pinhole, is largely blocked by the pinhole's edges. This mechanism significantly improves the signal-to-noise ratio and axial resolution, providing true optical sections and directly addressing the noise and blurring problems prevalent in traditional wide-field imaging. Confocal microscopy can achieve a lateral resolution of approximately 0.2 µm and an axial resolution of approximately 0.6 µm, making it ideal for high-precision reconstruction of microscopic edge morphology.

[0033] A white-light interferometer is a powerful non-contact 3D surface topography measuring instrument. Its working principle is based on a Michelson interferometer structure, using a broadband light source (i.e., white light). The beam is split into two paths: one is directed towards a high-precision reference mirror, and the other towards the edge surface of the wafer being measured. The two reflected beams re-merge to produce interference. Due to the extremely short coherence length of white light, interference fringes with maximum contrast are only produced when the optical path difference between the reference and measurement paths is close to zero. By precisely scanning the sample or reference mirror along the axial direction, a curve (called a cross-correlation plot) showing the interference signal intensity changing with axial position can be recorded for each pixel. The peak position of this curve precisely corresponds to the surface height represented by that pixel. White-light interferometers can achieve sub-nanometer vertical resolution, making them ideal for accurately mapping minute curvature variations on wafer edges.

[0034] A laser scanner is a contour measuring instrument that balances speed and accuracy, making it particularly suitable for high-throughput industrial production environments. Its basic principle involves projecting a laser beam or point onto the edge of a wafer at a known angle, while a camera with a position-sensitive detector (such as a CCD or CMOS array) observes the reflected laser spot on the surface from another known angle. Due to trigonometric geometry, the imaging position of the reflected spot on the detector shifts with changes in the height of the measured surface. By precisely calibrating the relative position and angle of the camera and laser, the system can calculate the coordinates of the corresponding point in real time based on the pixel position of the laser spot on the detector.

[0035] In some examples, processor 140 can be configured to calculate initial curvature values ​​by at least one method selected from: the close circle method, the finite difference method, and the local polynomial spline fitting method. Thus, processor 140 can calculate initial curvature values ​​for each data point. Calculate an initial curvature value, which may still contain noise. .

[0036] The method of closely spaced circles is a geometrically intuitive approach. For any three adjacent, non-collinear points on a curve (e.g., ...), ... , , A circle passing through all three points can be uniquely determined; this circle is called the "closely circumcircle" or "circumcircle" of that point. curvature at It is then estimated as the reciprocal of the radius R of the circle, i.e. =1 / R. Based on geometric relationships, the curvature can be directly calculated from the coordinates of these three points.

[0037] The finite difference method treats the edge profile as a parameterized curve and obtains the curvature value using an analytical formula based on the curve equation. To apply this formula to discrete data points, finite differences are needed to approximate the first and second derivatives. For example, the central difference formula can be used to obtain second-order accuracy. Substituting the approximate values ​​of the first and second derivatives into the analytical formula yields the estimate of the curvature. This method is computationally efficient, but as described in the background section, it is highly sensitive to noise because the difference operations amplify high-frequency noise.

[0038] In the local polynomial spline fitting method, to obtain a smoother derivative estimate, a point... Within a local neighborhood, a low-order polynomial (e.g., a quadratic or cubic B-spline curve) is fitted. This fitted polynomial can then be analytically computed in [the local area]. The first and second derivatives of the point are taken and substituted into the analytical formula for curvature. This method provides a degree of intrinsic smoothness by fitting locally, similar to applying a specific form of low-pass filter.

[0039] In some examples, the weighting coefficients can be obtained using a Gaussian kernel function. It is confirmed that, among them, The distance between neighboring data points and the target data point. The standard deviation is used to control the rate of weight decay.

[0040] The Gaussian kernel provides a smooth, bell-shaped weighted distribution, giving the highest weight to the nearest neighbor, while the weights decay smoothly and exponentially with increasing distance. The Gaussian kernel possesses many excellent mathematical properties. In the frequency domain, the Fourier transform of the Gaussian function remains a Gaussian function, meaning that a Gaussian filter is an ideal low-pass filter that smoothly attenuates high-frequency noise without producing oscillations or "ringing" artifacts in the frequency domain—a drawback inherent in simple box filters (i.e., moving averages). Furthermore, the Gaussian filter achieves an optimal balance between spatial and frequency domain localization; that is, it satisfies the lower bound of the uncertainty principle, enabling effective frequency filtering while minimizing spatial ambiguity.

[0041] Standard deviation This is a key adjustable parameter of the kernel function; it controls the rate of weight decay and is also known as the "bandwidth parameter." A smaller... A higher value will concentrate the weights more on the nearest neighbor points, providing less smoothing but better preserving local details; while a higher value will... The value then takes into account the influence of more distant neighbors, providing a stronger smoothing effect. In practical applications, and neighborhood size It can be empirically fixed based on historical data of a specific process, or it can be designed to be adaptive, for example, dynamically adjusted based on real-time noise level estimates (such as local variance) of the acquired data. A practical option. The guidance is to correlate it with the full width at half maximum (FWHM) of the feature to be preserved.

[0042] While the Gaussian kernel is preferred, other kernel functions that satisfy the distance decay property can also be used to implement the ideas of this disclosure, such as the Epanechnikov kernel (which is parabolic in shape) or the triangular kernel. These kernel functions may provide computational convenience in certain situations, while still adhering to the core principle of "contributing more to nearby objects and less to distant objects".

[0043] See Figure 3This disclosure also provides a method for determining wafer edge morphology parameters, which may include the following steps S301, S302, S303 and S304.

[0044] Step S301: Store a set of discrete data points representing the edge profile of wafer W.

[0045] Step S302: Calculate the initial curvature values ​​of a set of discrete data points.

[0046] Step S303: Calculate the stable curvature values ​​of multiple target data points in a set of discrete data points, wherein the stable curvature values ​​are determined based on the initial curvature values ​​of the target data points and the weighted combination of the initial curvature values ​​of multiple neighboring data points of the target data points, and the weight coefficients used for the weighted combination are configured to decrease as the distance between the neighboring data points and the target data points increases.

[0047] Step S304: Traverse the stable curvature values ​​of multiple target data points to identify the maximum stable curvature value, and use the maximum stable curvature value as the FER parameter value of wafer W.

[0048] After calculating the stable curvature values ​​of multiple target data points in step S303, the final identification step is executed. This step does not simply output the entire curve, but extracts a single parameter with clear physical meaning and process guidance value. Specifically, the processor traverses the entire array of stable curvature values ​​to identify the global maximum value. The spatial location of the data point corresponding to this maximum curvature value is ultimately defined as the position of the "Front Edge Radius (FER)". This maximum curvature value itself is output as the final, robust FER parameter for subsequent process control and quality monitoring. This final identification step makes the technical solution of this disclosure more complete and directly addresses the core technical problem of "obtaining stable and reliable FER parameters" as described in the background art.

[0049] In some examples, the method may also include filtering a set of discrete data points to suppress noise before calculating the initial curvature value.

[0050] In some examples, the weighting coefficients can be obtained using a Gaussian kernel function. It is confirmed that, among them, The distance between neighboring data points and the target data point. The standard deviation is used to control the rate of weight decay.

[0051] See Figure 4The present disclosure also provides a wafer edge topography control system 10, which may include a wafer edge topography parameter determination device 100 and a process control device 200 according to the embodiments described in the present disclosure.

[0052] The process control device 200 is communicatively connected to the processor 140 of the wafer edge topography parameter determination device 100. The process control device 200 is configured to generate and send adjustment instructions for adjusting at least one process parameter of the wafer manufacturing tool 20 based on the deviation between the FER parameter value determined by the wafer edge topography parameter determination device 100 and the target value.

[0053] Therefore, a novel process flow can be implemented using the aforementioned robust FER parameter values ​​to achieve proactive process control. This flow is a specific batch-to-batch control implementation within the advanced process control framework in semiconductor manufacturing. Its core lies in the introduction of a "feedforward" control mechanism, fundamentally changing the traditional "posterior" control mode. The following will describe this novel process flow in detail.

[0054] In addition, this modular system architecture not only clearly defines the functions of each part, but also provides a high degree of flexibility and scalability for deployment and upgrades in actual production environments.

[0055] Before a critical final shaping step (e.g., edge etching or edge polishing), but after a preliminary shaping step (e.g., edge grinding), a wafer edge morphology parameter determination method according to an example of this disclosure is performed to obtain an accurate, pre-polishing intermediate FER parameter value.

[0056] The process control unit 200 receives the intermediate FER parameter value and compares it with an ideal target value predefined for that specific product and process layer. The process control unit 200 then calculates the deviation between the two. This deviation precisely quantifies the degree to which the edge morphology of the current wafer deviates from the ideal state after the upstream process is completed.

[0057] Based on the calculated deviation, the process control unit 200 executes its core feedforward control logic, which involves selecting or dynamically calculating adjustments to the subsequent polishing process formulation. This is a classic feedforward control behavior because it utilizes upstream measurement information to proactively control the downstream process, aiming to eliminate defects before they form. This adjustment logic can be based on a formulation database / lookup table model. Specifically, the process control unit 200 internally stores a pre-established database or lookup table. This table maps different ranges of incoming material deviation values ​​to a set of pre-optimized, specific polishing formulation adjustment parameters. This table can be built based on extensive engineering experimental data or process simulations.

[0058] Adjustments to the formulation can modify various polishing process parameters to controllably and differentiate the amount of material removed from the wafer edges. These parameters include, but are not limited to, mechanical and chemical parameters. Mechanical parameters may include, for example, total polishing time, wafer carrier rotation speed, pressure applied to the wafer, and polishing slurry flow rate. Chemical parameters may include, for example, the composition of the polishing slurry (e.g., pH value, oxidant concentration, etchant concentration, abrasive particle size distribution or concentration) and the temperature of the polishing process.

[0059] The new process formulation, after precise calculation and adjustment, can be sent to wafer fabrication tool 20 (e.g., polishing equipment) via, for example, the SECS / GEM protocol. The wafer fabrication tool 20 then processes the wafer using these parameters "tailor-made" for the current wafer.

[0060] After the polishing process is complete, a final FER measurement can be performed on the wafer. This post-processing data serves a dual purpose: firstly, it verifies the effectiveness of the feedforward control and acts as a final quality control gate for this batch of wafers. Secondly, and more importantly, this data can be fed back to the process control unit 200 to continuously update and improve its internal process model over time. This hybrid mode, combining feedforward and feedback control, can automatically compensate for slow process drift caused by equipment aging, consumable replacement, etc., thereby achieving long-term, maximum process stability.

[0061] This disclosure also provides a wafer family, each wafer in which is produced by means of a wafer edge topography control system 10 according to an embodiment of this disclosure, wherein the coefficient of variation of the FER parameter values ​​of all wafers in the wafer family is less than 2%.

[0062] In one example, 10 wafers from each of two series were processed using both the conventional process and the novel process described above, with a target FER parameter value of 85.

[0063] For the first series of 10 wafers, the FER parameters of each wafer measured after processing are shown in Table 1 below, obtained by reducing etching time and polishing speed in the new process flow.

[0064]

[0065] Table 1 As shown in Table 1, the average value of the FER parameter is 85.361, and the standard deviation is 0.92. Therefore, the coefficient of variation is (0.92 / 85.361)×100%=1.08%, indicating that the fluctuation range is relatively small compared with the measurement benchmark. In addition, the 95% confidence interval calculated using the t-distribution is (84.7, 86.0), while the target value is 85, indicating that there is no significant difference between the actual measured value and the target value (at the 95% confidence level).

[0066] In contrast, for the 10 wafers in the second series, conventional methods were used for processing (without reducing etching time or polishing speed). The FER parameters of each wafer measured after processing are shown in Table 2 below.

[0067]

[0068] Table 2 As shown in Table 2, the average value of the FER parameter is 85.38 and the standard deviation is 5.00. Therefore, the coefficient of variation is (5.00 / 85.38)×100%=5.86%, indicating that the fluctuation range is relatively large compared to the measurement benchmark. In addition, the 95% confidence interval calculated using the t-distribution is (81.8, 88.9), which is much wider. This indicates that the estimation error of the FER parameter value is larger, the data volatility is higher, and the data points are more dispersed.

[0069] In another example, 10 wafers from each of the two series were processed using both the conventional process and the novel process described above, with a target FER parameter value of 120.

[0070] For the first series of 10 wafers, in the new process flow, by increasing the etching time and polishing speed, the FER parameters of each wafer measured after processing are shown in Table 3 below.

[0071]

[0072] Table 3 As shown in Table 3, the average value of the FER parameter is 120.44, and the standard deviation is 1.11. Therefore, the coefficient of variation is (1.11 / 120.44)×100%=0.92%, indicating that the fluctuation range is relatively small compared with the measurement benchmark. In addition, the 95% confidence interval calculated using the t-distribution is (119.7, 121.2), while the target value is 120, indicating that there is no significant difference between the actual measured value and the target value (at the 95% confidence level).

[0073] In contrast, for the 10 wafers in the second series, conventional methods were used for processing (without increasing etching time or polishing speed). The FER parameters of each wafer measured after processing are shown in Table 4 below.

[0074]

[0075] Table 4 As shown in Table 4, the average value of the FER parameter is 118.24, and the standard deviation is 3.41. Therefore, the coefficient of variation is (5.00 / 85.38)×100%=2.88%, indicating that the fluctuation range is relatively large compared to the measurement benchmark. In addition, the 95% confidence interval calculated using the t-distribution is (115.8, 120.7), which is much wider. This also indicates that the estimation error of the FER parameter value is larger, the data volatility is higher, and the data points are more dispersed.

[0076] It should be noted that the technical solutions described in the embodiments of this disclosure can be combined arbitrarily without conflict.

[0077] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A device for determining wafer edge morphology parameters, characterized in that, The device includes: The memory is configured to store a set of discrete data points representing the edge contours of the wafer; and A processor, communicatively connected to the memory, is configured to calculate initial curvature values ​​for the set of discrete data points and stable curvature values ​​for a plurality of target data points within the set of discrete data points. The stable curvature values ​​are determined based on a weighted combination of the initial curvature values ​​of the target data points and the initial curvature values ​​of a plurality of neighboring data points. The weighting coefficients used for the weighted combination are configured to decrease as the distance between the neighboring data points and the target data points increases. The processor is further configured to iterate through the stable curvature values ​​of the plurality of target data points to identify the maximum stable curvature value and use the maximum stable curvature value as the FER parameter value of the wafer.

2. The wafer edge morphology parameter determination device according to claim 1, characterized in that, The processor includes a data preprocessing unit configured to filter the set of discrete data points to suppress noise before calculating the initial curvature value.

3. The wafer edge morphology parameter determination device according to claim 2, characterized in that, The data preprocessing unit is configured to apply at least one edge-preserving filter selected from the following: Savitzky-Golay filter, bilateral filter, and wavelet denoising algorithm.

4. The wafer edge morphology parameter determination device according to claim 1, characterized in that, The apparatus further includes a data acquisition unit for acquiring the set of discrete data points, the data acquisition unit comprising at least one of the following measurement devices: a confocal microscope, a white light interferometer, and a laser scanner.

5. The wafer edge morphology parameter determination device according to claim 1, characterized in that, The processor is configured to calculate the initial curvature value by at least one of the following methods: close circle method, finite difference method, and local polynomial spline fitting method.

6. The wafer edge morphology parameter determination apparatus according to any one of claims 1 to 5, characterized in that, The weighting coefficients are obtained through a Gaussian kernel function. It is confirmed that, among them, The distance between neighboring data points and the target data point. The standard deviation is used to control the rate of weight decay.

7. A method for determining wafer edge morphology parameters, characterized in that, The method includes: Store a set of discrete data points representing the edge contour of the wafer; Calculate the initial curvature values ​​of the set of discrete data points; Calculate stable curvature values ​​for multiple target data points in the set of discrete data points, wherein the stable curvature values ​​are determined based on a weighted combination of the initial curvature values ​​of the target data points and the initial curvature values ​​of multiple neighboring data points, and the weighting coefficients used for the weighted combination are configured to decrease as the distance between the neighboring data points and the target data point increases; and The stable curvature values ​​of the multiple target data points are traversed to identify the maximum stable curvature value, and the maximum stable curvature value is used as the FER parameter value of the wafer.

8. The method for determining wafer edge morphology parameters according to claim 7, characterized in that, The method further includes filtering the set of discrete data points to suppress noise before calculating the initial curvature value.

9. A wafer edge topography control system, characterized in that, The system includes: The wafer edge morphology parameter determination apparatus according to any one of claims 1 to 6; and A process control device is communicatively connected to the processor of the wafer edge topography parameter determination device. The process control device is configured to generate and send adjustment instructions for adjusting at least one process parameter of the wafer manufacturing tool based on the deviation between the FER parameter value determined by the device and the target value.

10. A wafer series, characterized in that, Each wafer in the wafer series is produced by means of the wafer edge morphology control system according to claim 9, wherein the coefficient of variation of the FER parameter values ​​of all wafers in the wafer series is less than 2%.

Images