Non-normal wafer parameter SPC control method based on quantile method

Abstract

The application relates to a non-normal wafer parameter SPC control method based on a quantile method, belongs to the technical field of semiconductor manufacturing quality control, and comprises the following operation steps: S1: performing normality inspection on collected data, screening out non-normal parameters, establishing exclusive data sets, and completing preprocessing; S2: sorting the preprocessed data sets, adopting an empirical quantile interpolation algorithm to calculate control reference values and upper and lower control limits, and differentiating quantile threshold values according to parameter risk grades; S3: collecting data in real time and comprehensively judging in combination with an SPC discrimination criterion, triggering graded early warning of abnormality; and S4: re-performing data calculation and checking according to a preset period, automatically updating the control limits to realize dynamic optimization. The method can accurately adapt to skewed and discrete wafer quality parameters, effectively reduce the false alarm and missed detection rates, realize unified control standards and dynamic iteration of the control limits, and improve wafer process stability and product yield.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing quality control technology, specifically to a nonnormal wafer parameter SPC control method based on quantile method. Background Technology

[0002] In the mass production of semiconductor wafer materials, statistical process control (SPC) is a core means to ensure process stability, improve product yield, and control process fluctuations. Traditional SPC control systems are based on the normal distribution assumption and use the ±3σ criterion to calculate process control limits. In theory and in early applications, this system is only applicable to process parameters that strictly follow a normal distribution.

[0003] In actual wafer manufacturing scenarios, many core quality parameters exhibit significant skewed distributions, strong dispersion, and marked non-normal characteristics. Typical parameters include the number of particles on the wafer surface, the number of micro-defects, and local thickness deviations. Data for these parameters generally fail normality tests. Forcibly applying the traditional SPC system will result in the following core defects: 1. Poor control accuracy: Traditional 3σ normal control limits may result in excessively large or small control ranges for non-normal parameters, failing to match the actual process fluctuation patterns.

[0004] 2. Serious misjudgments and omissions: Normal process fluctuations are easily misjudged as abnormalities (excessive alarms), or minor hidden process drifts cannot be identified (missed detection), resulting in frequent production downtime, batch misinterpretation, and the retention of potential abnormalities.

[0005] 3. Poor universality: The industry lacks targeted solutions for managing non-normal parameters, and on-site personnel can only forcibly apply normal SPC models or rely on manual experience to make judgments, resulting in low levels of standardization and digitalization in management. Fourth, lagging control: Fixed and unchanging normal control limits cannot adapt to changes in mass production batch process iterations, equipment aging, and environmental fluctuations. Long-term use will continuously accumulate process risks, affecting wafer yield and shipment quality stability. Summary of the Invention

[0006] This invention addresses the shortcomings of existing technologies by providing a non-normal wafer parameter SPC control method based on the quantile method. It overcomes the limitations of traditional SPC methods, which rely on the assumption of normal distribution, leading to control failures of non-normal parameters, high false alarm and false negative rates, and the inability to adaptively update control limits. By employing precise adaptation to skewed and discrete wafer quality parameters, it effectively reduces false alarm and false negative rates, achieves unified control standards and dynamic iteration of control limits, and improves wafer process stability and product yield.

[0007] The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions: A method for SPC control of nonnormal wafer parameters based on quantile method includes the following steps: Step S1: Data Acquisition and Preprocessing: Perform normality testing on the acquired wafer process parameter dataset, screen out non-normal process parameters that do not meet the normal distribution, establish a dedicated control dataset for non-normal parameters, and remove abnormal data to complete data preprocessing.

[0008] Step S2: Control Limit Calculation: Sort the preprocessed non-normal parameter dataset and use the empirical quantile interpolation algorithm to calculate the SPC control baseline and upper and lower control limits.

[0009] Step S3: Real-time monitoring and early warning: Collect process parameter data during wafer mass production in real time and synchronize it to the SPC system. Combine the SPC anomaly judgment criteria to comprehensively judge the process fluctuations of non-normal parameters. If an anomaly rule is triggered, generate graded early warning information and push the anomaly type and parameters.

[0010] Step S4: Dynamic update of control limits: Recalculate and process historical data, perform normality verification and quantile calculation according to a preset period, and automatically update the SPC control limits of the corresponding parameters to achieve adaptive dynamic optimization of control limits.

[0011] As a preferred option, in step S1, the Anderson-Darling test is used to perform a normality test on the collected dataset. If the dataset fails the Anderson-Darling normality test, it is determined to be a non-normal process parameter and included in the dedicated control dataset.

[0012] As a preferred option, in step S2, the preprocessed nonnormal parameter dataset is sorted in ascending order; the 0.5 quantile is selected as the process centerline CL, the 0.135% quantile is selected as the lower control limit LCL, and the 99.865% quantile is selected as the upper control limit UCL, so as to be aligned with the traditional 3σ control confidence level.

[0013] Preferably, in step S2, the quantile threshold is set differently according to the risk level of the wafer process parameters; for high-risk parameters, the tail quantile range is adjusted to improve control accuracy; for low-risk parameters, the tail quantile range is adjusted to take into account production efficiency.

[0014] Preferably, in step S3, the SPC anomaly detection criteria include anomaly detection rules adjusted for non-normal, skewed distributions and discrete data characteristics; the graded early warning information is divided and pushed in a graded manner according to the severity of the triggering anomaly rules, the triggering frequency, and the risk level of the parameters themselves.

[0015] Preferably, in step S4, the preset cycle is set differently according to the parameter characteristics: the preset cycle for key and important characteristic parameters is once per quarter, and the preset cycle for general characteristic parameters is once every six months; the updated control limits are used to adapt to the normal changes in mass production caused by slight aging of equipment, fine-tuning of processes, and fluctuations in environmental temperature and humidity.

[0016] Preferably, in step S4, when automatically updating the SPC control limits of the corresponding parameters, if the distribution pattern of the new data collected within the update cycle changes significantly due to process iteration, then steps S1 and S2 are retried to recalculate the quantile control limits based on the data of the new cycle.

[0017] The present invention can achieve the following effects: This invention provides a nonnormal wafer parameter SPC control method based on the quantile method, which has the following significant advantages compared with the prior art: I. Precisely adapts to non-normal characteristics, significantly improving control accuracy: Abandoning the traditional 3σ calculation method based on the assumption of normal distribution, it adopts an empirical quantile interpolation algorithm (0.135%, 0.5%, 99.865% quantile) to benchmark against the traditional 3σ confidence level, truly reflecting the distribution law of skewed and discrete parameters (such as particle number and micro-defects), fundamentally solving the problem of distortion in the traditional 3σ control limit for non-normal parameters.

[0018] II. Significantly reduces false positives and false negatives, minimizing wasted work: By precisely setting control limits, it effectively avoids problems caused by excessively large control intervals (potential for missed detections) or excessively small control intervals (excessive alarms) due to long-tail distribution. Practical application verification shows that it can reduce the wasted analysis workload of on-site quality personnel by more than 60%, accurately capture the hidden drift and long-tail fluctuations of non-normal parameters, predict potential process defects in advance, and reduce the generation of batch defects.

[0019] 3. Differentiated risk management, balancing quality and capacity: Breaking through the traditional SPC "one-size-fits-all" management model, quantile thresholds are set differently according to the risk level of wafer process parameters. This ensures extremely high control accuracy for high-risk parameters (such as the number of particles that affect yield) while taking into account the production flow efficiency of low-risk parameters, which is highly in line with the actual needs of wafer mass production scenarios.

[0020] IV. Dynamic Adaptive Optimization of Control Limits for Long-Term Consistency: By setting a regular review and update mechanism (quarterly or semi-annually), the system automatically recalculates quantile control limits based on new batch data, perfectly adapting to normal mass production changes such as slight equipment aging, process fine-tuning, and fluctuations in environmental temperature and humidity. This avoids control failures caused by long-term use of fixed control limits, maintains SPC control accuracy over the long term, and ensures product quality consistency and customer compliance.

[0021] V. Achieve full data-driven and standardized management and control: The SPC management and control standards for non-normal parameters have been unified, completely replacing the traditional reliance on human experience judgment. This enables a fully digital closed loop from data acquisition, preprocessing, control limit calculation to real-time early warning, significantly improving the professionalism and standardization of semiconductor manufacturing quality control. Attached Figure Description

[0022] Figure 1 This is a comparison chart of the shortcomings of the traditional normal 3σSPC control method of this invention.

[0023] Figure 2 This is a schematic diagram illustrating the principle of calculating the control limits using the quantile method of the present invention.

[0024] Figure 3 This is a diagram of the non-normal parameter process anomaly graded early warning push architecture of the present invention. Detailed Implementation

[0025] The technical solution of the invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0026] Example: Figure 1 , Figure 2 and Figure 3 As shown, a non-normal wafer parameter SPC control method based on quantile method includes the following operation steps: Step S1: Data Acquisition and Preprocessing. In the initial stage of establishing the SPC project, historical data on the number of wafer surface particles from 500 consecutive batches in a specific wafer cleaning process section were collected. Since particle count is a typical discrete count data point with a right-skewed distribution (long-tailed distribution), the traditional normal distribution assumption is not applicable. The Anderson-Darling (AD) test was used to check the normality of these 500 data points. The calculated p-value was less than 0.05, failing the normality test. Therefore, it was determined to be a non-normal process parameter, and a dedicated control dataset for "wafer surface particle count" was established. Subsequently, this dataset underwent data cleaning to remove extreme outliers caused by sudden equipment failures, test machine malfunctions, etc., completing the data preprocessing.

[0027] Step S2: Control Limit Calculation. The preprocessed non-normal particle quantity dataset is sorted in ascending order. The empirical quantile interpolation algorithm is used to calculate the SPC control baseline and upper and lower control limits: The 0.5 quantile (median) of the sorted dataset is selected as the process centerline (CL), reflecting the true concentration trend of the current process; the 0.135% quantile is selected as the lower control limit (LCL); and the 99.865% quantile is selected as the upper control limit (UCL). These intervals cover 99.73% of the data, accurately aligning with the confidence level of traditional 3σ normal distribution control, while effectively avoiding the problem of excessively large traditional 3σ UCLs due to right-skewed long tails.

[0028] Meanwhile, thresholds were set differently based on the risk level of the parameters. Because the "number of particles on the wafer surface" directly affects the subsequent lithography yield and is considered a high-risk parameter, the tail percentile of its UCL was fine-tuned to the 99.80th percentile on-site to further tighten the control limit and improve the accuracy of controlling abnormal particle growth.

[0029] Step S3: Real-time monitoring and early warning. During wafer mass production, the cleaning machine synchronizes the particle count test data of each batch of wafers to the SPC system in real time. The system compares the real-time data with the quantile control limits calculated in step S2, and makes a comprehensive judgment based on the SPC anomaly criteria adjusted for non-normal characteristics (such as: an upward trend of 7 consecutive test points, single-point breakthrough of UCL, etc.).

[0030] When the particle count of a batch triggers the anomaly detection rule (e.g., the particle count of 3 consecutive batches approaches the UCL and shows an upward trend), the SPC system immediately generates an "orange warning" message. The system automatically pushes the anomaly type (hidden process drift warning), batch number, out-of-range parameters and current values ​​to the terminals of quality engineers and equipment maintenance personnel so that they can intervene in time to investigate the lifespan of filter consumables or the concentration of cleaning fluid and avoid batch defects.

[0031] Step S4: Dynamically update control limits. As cleaning equipment ages slightly with increased production batches, filter cartridge efficiency gradually declines, and the baseline particle count exhibits a slow, consistent upward trend. If control limits remain unchanged, this will lead to a surge in false alarms over the long term.

[0032] Because "wafer surface particle count" is a critical characteristic parameter, the system sets its control limit review and update cycle to "once per quarter". At the end of each quarter, the system automatically extracts newly collected data from the past three months, re-performs normality verification and data preprocessing, and recalculates the 0.5, 0.135%, and 99.865% quantiles, automatically updating the SPC control limits for the next quarter.

[0033] Through this adaptive dynamic optimization, the control limits can keep up with the normal changes in mass production, such as equipment aging, process fine-tuning, and fluctuations in ambient temperature and humidity, maintaining long-term control accuracy. If a major process iteration occurs during the update cycle (such as replacing with a new cleaning machine), the system will trigger a reset mechanism to re-execute steps S1 and S2 based on the new machine data to establish a new baseline.

[0034] In summary, this SPC control method for non-normal wafer parameters based on the quantile method effectively solves the problems of control range distortion, misjudgment, and missed judgment that occur when using traditional 3σ control limits for non-normal parameters such as the number of particles on the wafer surface. Field application shows that this method reduces the amount of ineffective analysis work for on-site quality personnel by approximately 65%, can predict potential performance degradation of equipment filtration systems 1-2 batches in advance, effectively improves the stability of the wafer cleaning process and the final product yield, and simultaneously achieves standardization and full digitalization of non-normal parameter control.

[0035] The above description is only a specific embodiment of the present invention, but the structural features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art within the scope of the present invention are covered by the patent scope of the present invention.

Claims

1. A method for controlling non-normal wafer parameters using SPC based on quantile method, characterized in that, The following steps are included: Step S1: Data Acquisition and Preprocessing: Perform normality testing on the acquired wafer process parameter dataset, screen out non-normal process parameters that do not meet the normal distribution, establish a dedicated control dataset for non-normal parameters, and remove abnormal data to complete data preprocessing; Step S2: Control Limit Calculation: Sort the preprocessed non-normal parameter dataset and use the empirical quantile interpolation algorithm to calculate the SPC control baseline and upper and lower control limits; Step S3: Real-time monitoring and early warning: Collect process parameter data during wafer mass production in real time and synchronize it to the SPC system. Combine the SPC anomaly detection criteria to comprehensively determine the process fluctuations of non-normal parameters. If an anomaly rule is triggered, a tiered early warning message will be generated, and the anomaly type and parameters will be pushed out. Step S4: Dynamic update of control limits: Recalculate and process historical data, perform normality verification and quantile calculation according to a preset period, and automatically update the SPC control limits of the corresponding parameters to achieve adaptive dynamic optimization of control limits.

2. The SPC control method for non-normal wafer parameters based on the quantile method according to claim 1, characterized in that, In step S1, the Anderson-Darling test is used to check the normality of the collected dataset. If the dataset fails the Anderson-Darling normality test, it is determined to be a non-normal process parameter and included in the dedicated control dataset.

3. The SPC control method for non-normal wafer parameters based on the quantile method according to claim 1, characterized in that, In step S2, the preprocessed nonnormal parameter dataset is sorted in ascending order; the 0.5 quantile is selected as the process centerline CL, the 0.135% quantile is selected as the lower control limit LCL, and the 99.865% quantile is selected as the upper control limit UCL, in order to align with the traditional 3σ control confidence level.

4. The SPC control method for non-normal wafer parameters based on the quantile method according to claim 3, characterized in that, In step S2, quantile thresholds are set differently based on the risk level of the wafer process parameters; for high-risk parameters, the tail quantile range is adjusted to improve control accuracy; for low-risk parameters, the tail quantile range is adjusted to balance production efficiency.

5. The SPC control method for non-normal wafer parameters based on the quantile method according to claim 1, characterized in that, In step S3, the SPC anomaly detection criteria include anomaly detection rules adjusted for non-normal, skewed distributions and discrete data characteristics; the graded early warning information is divided and pushed in a graded manner according to the severity of the triggering anomaly rules, the triggering frequency, and the risk level of the parameters themselves.

6. The SPC control method for non-normal wafer parameters based on the quantile method according to claim 1, characterized in that, In step S4, the preset cycle is set differently according to the parameter characteristics: the preset cycle for key and important characteristic parameters is once per quarter, and the preset cycle for general characteristic parameters is once every six months. The updated control limits are designed to accommodate normal production changes caused by slight equipment aging, process fine-tuning, and fluctuations in ambient temperature and humidity.

7. The SPC control method for non-normal wafer parameters based on the quantile method according to claim 1 or 6, characterized in that, In step S4, when automatically updating the SPC control limits of the corresponding parameters, if the distribution pattern of the new data collected within the update cycle changes significantly due to process iteration, then steps S1 and S2 are retried, and the quantile control limits are recalculated based on the data of the new cycle.

Images