Chip processing monitoring system and method based on integrated circuit optimization

Abstract

The invention discloses a chip processing monitoring system and method based on integrated circuit optimization, and relates to the technical field of chip processing.The method comprises the steps that historical qualified chips are collected, wafers of all batches are sampled, abnormal values are removed, and a historical qualified CD data set is constructed; calculating a batch CD mean value sequence to obtain a global reference statistical magnitude, setting a judgment threshold interval by combining standard error and quantile, calculating CD offset of each batch, distinguishing randomly fluctuating tiny offset and significant offset of process drift, setting time windows to construct CD offset vectors, calculating Euclidean distance of adjacent time windows, and when trend mutation is monitored, judging whether the sequence is abnormal or not; starting an offset prediction model to pre-judge the next batch of offset; through the Pearson coefficient, key process parameters are screened, a mapping model is constructed, the parameter adjustment amount is solved, and adjustment is executed, so that precise monitoring, active early warning and process optimization of the chip CD value are realized, bad batches are effectively reduced, and the chip processing quality is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of chip manufacturing technology, specifically to a chip manufacturing monitoring system and method based on integrated circuit optimization. Background Technology

[0002] In the integrated circuit chip manufacturing process, the CD (Chip Dryness) value directly determines the chip's electrical performance, structural precision, and final product yield. Accurate monitoring and process control of the CD value are crucial for ensuring chip manufacturing quality. However, existing CD value monitoring and process adjustment technologies have significant shortcomings: First, historical CD data acquisition often relies on random sampling, failing to specifically cover critical areas such as the wafer center and edges, and ineffectively removing outliers, resulting in poor data quality and an inability to provide reliable support for subsequent analysis. Second, there is a lack of unified global benchmark statistics; judgment thresholds depend on empirical settings without rationality verification, and the technology cannot distinguish between minor shifts caused by random fluctuations and significant shifts caused by process drift, easily leading to misjudgments or missed judgments. Third, most monitoring methods are passive, making it difficult to predict CD value shifts in advance through trend analysis, and the lack of a quantitative mapping relationship between CD value and process parameters results in blind process adjustments, making it impossible to accurately determine adjustment parameters and amounts, ultimately leading to an increase in defective batches and processing costs, failing to meet the high-precision, high-yield processing requirements of chips. Summary of the Invention

[0003] The purpose of this invention is to provide a chip processing monitoring system and method based on integrated circuit optimization to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a chip fabrication monitoring method based on integrated circuit optimization, the method comprising the following steps: Step S1: Collect historical chips. Sampling is performed on each batch of wafers to collect key data. The key data is preprocessed to construct a historical qualified CD dataset. Step S1-1: Collect qualified chips from N historical batches. For each batch of chips, randomly select x sampling points on the chip wafer in a uniform distribution manner, with the sampling points covering the center and edge positions of the wafer. Collect the CD value and batch number of each sampling point. The key data are CD value and batch number. Step S1-2: Use the 3σ principle to remove outliers from the collected CD values, specifically as follows: ; In the formula, x is the CD value. The CD mean of the original historical data. The CD standard deviation of the original historical data; The CD values ​​processed by the 3σ principle are integrated to construct a historical qualified CD dataset.

[0005] By sampling to cover key locations on the wafer and eliminating outliers using the 3σ principle, the integrity and accuracy of historical qualified CD datasets are ensured, providing reliable data support for subsequent benchmarking and model building.

[0006] Step S2: Based on the preprocessed data, calculate the global baseline statistic, set the judgment threshold, and verify the reasonableness of the threshold. Step S2-1: Calculate the mean CD value for each batch, specifically as follows: ; In the formula, Let m be the mean CD value of batch t, ​​where t is the batch number and m is the number of CD sampling points in batch t. The CD measurement value of the i-th sampling point in the t-th batch; Construct a batch mean sequence of historical qualified CD datasets ; Step S2-2: Calculate the mean and standard deviation of the batch mean sequence of the historical qualified CD dataset, specifically as follows: ; In the formula, Let N be the global baseline mean of historically qualified CDs, N be the total number of historically qualified batches, and t be the sequence number of the historically qualified batch. Let CD be the average value of batch t, ​​and N be the total number of qualified batches in history; ; In the formula, The standard deviation of the historical average of qualified CD batches. The summation symbol indicates that the corresponding data from the 1st to the Nth historical qualified batches are summed. The mean and standard deviation of the batch mean sequence of the historical qualified CD dataset are set as global benchmark statistics; Step S2-3: Calculate the standard error based on the mean and standard deviation of the batch mean sequence of the historical qualified CD dataset. Then, calculate the fluctuation range based on the standard error and quantiles. Specifically: ; In the formula, SE is the standard error. is the standard deviation of the historical average of qualified CD batches, and N is the total number of historical qualified batches; The fluctuation range is: In the formula, SE is the standard error, and z 0.995 It is the 99.5th percentile of the normal distribution; Step S2-4: The decision threshold is calculated based on the mean of the batch mean sequence of the historical qualified CD dataset, combined with the standard error. Specifically: Upper control limit: ; In the formula, UCL is the upper control limit. Let SE be the global baseline mean of historical qualified CDs, and z be the standard error. 0.995 It is the 99.5th percentile of the normal distribution; Lower control limit: ; In the formula, LCL is the lower control limit. Let SE be the global baseline mean of historical qualified CDs, and z be the standard error. 0.995 It is the 99.5th percentile of the normal distribution; Determine the threshold range: ; Step S2-5: Compare the batch mean of the batch mean sequence of the historical qualified CD dataset with the judgment threshold interval. When the batch mean of the batch mean sequence of the historical qualified CD dataset is above J% and falls within the judgment threshold interval, the judgment threshold interval is qualified. J% is set by professionals according to the business scenario.

[0007] Establish global benchmark statistics and reasonable judgment thresholds, clarify the normal fluctuation range of CD values, avoid misjudgments caused by ambiguous judgment criteria, and improve the scientific nature of subsequent offset judgments.

[0008] Step S3: Set the CD offset type. Calculate the CD offset for each batch based on the average CD value of each batch and the average of the batch average sequence of the historical qualified CD dataset. Determine the type of the CD offset. Step S3-1: Set the CD offset type to small offset and significant offset. Small offset is caused by random fluctuations, and significant drift is caused by process drift. Step S3-2: Calculate the CD offset for each batch based on the average CD value of each batch and the average of the batch average sequence of the historical qualified CD dataset. Specifically: ; In the formula, This is the CD offset. The CD mean for each batch, where μ is the mean of the batch mean sequence of the historical qualified CD dataset; Step S3-3, when When >0, the CD value is too large and offset. When <0, the CD value is too small and offset; when | When the price is within the fluctuation range, it is considered a minor offset, and no trend analysis is performed. If the value is not within the threshold range, it is considered a significant shift and should be included in trend analysis.

[0009] Distinguish between minor and significant offsets to accurately pinpoint the problem type, avoid over-processing random fluctuations that have no impact, and focus on process drift issues that require attention.

[0010] Step S4: Set a time window, calculate the Euclidean distance between the offset vectors of two adjacent time windows, quantify the offset trend change, and construct an offset prediction model based on the quantified offset trend change. Step S4-1: Plot an offset trend chart based on the batch number and CD offset, and fit the offset trend, specifically as follows: ; In the formula, Let a0, a1, a2, ..., a be the CD offset for batch t, ​​where t is the batch number. p For polynomial coefficients, This is the random error term; Step S4-2: Solve for the polynomial coefficients using the least squares method, and then substitute the polynomial coefficients into the offset trend fitting formula to construct the offset prediction model. Step S4-3: Set the time window size to m, and construct the CD offset vector using the sliding window method, specifically as follows: ; In the formula, Let be the CD offset vector for the k-th sliding time window, where k is the index of the sliding window. This represents the CD offset for the kth batch. Let m be the CD offset for the (k+1)th batch, and m be the size of the sliding time window. This is the CD offset for the last batch within this time window; Step S4-4: Calculate the window Euclidean distance for all historical qualified batches, specifically as follows: ; In the formula, Let be the Euclidean distance between the k-th adjacent sliding windows. This is the summation operator, which means summing all terms from i=1 to i=m-1. Let CD be the CD offset of the i-th batch in the k-th sliding window. This represents the CD offset for the i-th batch within the (k-1)-th sliding window. It is the square of the difference in CD offset between the batches at the i-th position in two adjacent time windows; Sort all historical qualified batches by window Euclidean distance in ascending order, and take the 95th percentile as the model activation threshold; Step S4-5: Calculate the Euclidean distance between the offset vectors of two adjacent time windows (CD). Step S4-6: When the Euclidean distance between the CD offset vectors of two adjacent time windows is less than or equal to the model activation threshold, it is determined that the offset trend is stable and the monitoring of subsequent batches continues. When the Euclidean distance between two adjacent time windows is greater than or equal to the model activation threshold, it is determined that the offset trend changes abruptly and the offset prediction model is activated to predict the CD offset of the next batch.

[0011] By monitoring trend abrupt changes through Euclidean distance over a time window and combining this with a predictive model to anticipate the CD offset of the next batch, we can shift from "passive monitoring" to "proactive early warning" and reduce the generation of defective batches.

[0012] Step S5: Construct the quantitative mapping relationship between offset and process parameters, solve for the process parameter adjustment amount, and then perform parameter adjustment based on the solved process parameter adjustment amount.

[0013] Step S5-1: Extract the key process parameters that affect CD, calculate the correlation coefficient of the key process parameters, screen the key process parameters by Pearson correlation coefficient, and include the key process parameters with |r|≥m into the mapping model. The |r| is a quantitative index that measures the linear correlation strength between the key CD and a certain process parameter, and m is set by professionals according to the business scenario. Step S5-2: Standardize the selected key process parameters using Z-score. Step S5-3: Based on the standardized q key process parameters, establish a mapping model, specifically as follows: ; In the formula, b0 is a constant term, b1,...,b q The sensitivity coefficients of the critical process parameters are P1,...Pq, where P1,...Pq is the critical process parameter sensitivity coefficient. q CD represents the CD value for the 1st, ..., qth key process parameter selected. Step S5-4: Convert the mapping model into a matrix, specifically: ; In the formula, Let P be a q×1 order CD value matrix, and let P be a q×1 order key process parameter matrix. This is a 1×q order critical process parameter sensitivity coefficient matrix. It is a q×1 order constant term matrix; The sensitivity coefficient matrix of key process parameters is solved using the least squares method. ; The predicted offset is input into the CD value matrix to obtain the process parameter adjustment matrix, and the key process parameters are adjusted according to the process parameter adjustment matrix.

[0014] By using Pearson correlation coefficient to screen key process parameters, a quantitative mapping model is established to accurately solve for the adjustment amount of process parameters, avoiding blind adjustments and improving the efficiency and accuracy of process optimization.

[0015] The system includes a qualified CD dataset construction module, a benchmark statistics and threshold setting module, a CD offset determination module, an offset trend prediction module, and a process parameter adjustment module. The qualified CD dataset construction module is used to collect historical qualified chip data. It collects CD values ​​and batch numbers by sampling wafers from each batch, removes outliers in CD values ​​using the 3σ principle, and integrates them to form a historical qualified CD dataset. The benchmark statistics and threshold setting module is used to calculate the average value of each batch of CDs based on the historical qualified CD dataset and construct a batch average value sequence to obtain a global benchmark statistic. Then, the threshold range is calculated and determined based on this statistic, and the rationality of the threshold range is verified. The CD offset determination module is used to first set two types of CD offsets, then combine the mean of the batch mean sequence of historical qualified CD datasets to calculate the CD offset of each batch, and finally determine the CD offset type of each batch based on the absolute value of the offset. The offset trend prediction module is used to fit the CD offset trend through a polynomial and build an offset prediction model. After setting a time window, it calculates the Euclidean distance between the CD offset vectors of adjacent windows to quantify the trend change. At the same time, it sets a model start threshold and determines whether to start the model based on the comparison between the Euclidean distance and the threshold to predict the CD offset of the next batch. The process parameter adjustment module is used to screen key process parameters affecting CD using Pearson correlation coefficient, construct a quantitative mapping model between CD and key process parameters, solve for the sensitivity coefficient of key process parameters, calculate the process parameter adjustment amount by combining the predicted offset, and finally perform process parameter adjustment based on the adjustment amount.

[0016] The qualified CD dataset construction module includes a data sampling and acquisition unit and a data preprocessing and integration unit. The data sampling and acquisition unit is used to collect the CD values ​​and batch numbers of each batch of historical qualified chips. The data preprocessing and integration unit is used to remove outliers in CD values ​​and integrate them to form a historical qualified CD dataset. The benchmark statistics and threshold setting module includes a global benchmark statistics unit and a threshold setting verification unit; The global benchmark statistics unit is used to calculate the batch mean sequence and global benchmark statistics of the historical qualified CD dataset; The threshold setting verification unit is used to calculate and determine the threshold range and verify its compliance.

[0017] The CD offset determination module includes an offset type setting unit and an offset calculation determination unit; The offset type setting unit is used to set two types of CD offset: small offset and significant offset; The offset calculation and determination unit is used to calculate the offset of each batch of CDs and determine their offset type. The offset trend prediction module includes a trend quantification analysis unit and a prediction model activation unit; The trend quantification analysis unit is used to fit the offset trend and quantify the trend change using Euclidean distance. The prediction model activation unit is used to set the model activation threshold and activate prediction based on trend mutations.

[0018] The process parameter adjustment module includes a mapping model construction unit and an adjustment amount solution execution unit; The mapping model construction unit is used to screen key process parameters and construct a CD-process parameter mapping model. The adjustment amount calculation and execution unit is used to calculate the process parameter adjustment amount and execute the parameter adjustment.

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention collects chip CD values ​​by sampling and removes outliers using the 3σ principle to build a high-quality dataset. Then, it calculates global baseline statistics and sets verification thresholds to provide accurate basis for CD offset determination and avoid misjudgment.

[0020] 2. This invention distinguishes CD offset types, combines time window Euclidean distance to monitor trend changes and activates a prediction model to achieve proactive early warning and reduce defective chip batches.

[0021] 3. This invention uses the Pearson coefficient to screen key process parameters, constructs a mapping model to calculate adjustment amounts, avoids blind adjustments, and improves the accuracy of process optimization. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a chip fabrication monitoring method based on integrated circuit optimization according to the present invention. Figure 2 This is a schematic diagram of the structure of a chip processing monitoring system based on integrated circuit optimization according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1: As Figures 1-2 As shown, the present invention provides a technical solution, a chip processing monitoring method based on integrated circuit optimization, the method comprising the following steps: Step S1: Collect historical chips. Sampling is performed on each batch of wafers to collect key data. The key data is preprocessed to construct a historical qualified CD dataset. Step S1-1: Collect qualified chips from N historical batches. For each batch of chips, randomly select x sampling points on the chip wafer in a uniform distribution manner, with the sampling points covering the center and edge positions of the wafer. Collect the CD value and batch number of each sampling point. The key data are CD value and batch number. Step S1-2: Use the 3σ principle to remove outliers from the collected CD values, specifically as follows: ; In the formula, x is the CD value. The CD mean of the original historical data. The CD standard deviation of the original historical data; The CD values ​​processed by the 3σ principle are integrated to construct a historical qualified CD dataset.

[0025] By sampling to cover key locations on the wafer and eliminating outliers using the 3σ principle, the integrity and accuracy of historical qualified CD datasets are ensured, providing reliable data support for subsequent benchmarking and model building.

[0026] Step S2: Based on the preprocessed data, calculate the global baseline statistic, set the judgment threshold, and verify the reasonableness of the threshold. Step S2-1: Calculate the mean CD value for each batch, specifically as follows: ; In the formula, Let m be the mean CD value of batch t, ​​where t is the batch number and m is the number of CD sampling points in batch t. The CD measurement value of the i-th sampling point in the t-th batch; Construct a batch mean sequence of historical qualified CD datasets ; Step S2-2: Calculate the mean and standard deviation of the batch mean sequence of the historical qualified CD dataset, specifically as follows: ; In the formula, Let N be the global baseline mean of historically qualified CDs, N be the total number of historically qualified batches, and t be the sequence number of the historically qualified batch. Let CD be the average value of batch t, ​​and N be the total number of qualified batches in history; ; In the formula, The standard deviation of the historical average of qualified CD batches. The summation symbol indicates that the corresponding data from the 1st to the Nth historical qualified batches are summed. The mean and standard deviation of the batch mean sequence of the historical qualified CD dataset are set as global benchmark statistics; Step S2-3: Calculate the standard error based on the mean and standard deviation of the batch mean sequence of the historical qualified CD dataset. Then, calculate the fluctuation range based on the standard error and quantiles. Specifically: ; In the formula, SE is the standard error. is the standard deviation of the historical average of qualified CD batches, and N is the total number of historical qualified batches; The fluctuation range is: In the formula, SE is the standard error, and z 0.995 It is the 99.5th percentile of the normal distribution; Step S2-4: The decision threshold is calculated based on the mean of the batch mean sequence of the historical qualified CD dataset, combined with the standard error. Specifically: Upper control limit: ; In the formula, UCL is the upper control limit. Let SE be the global baseline mean of historical qualified CDs, and z be the standard error. 0.995 It is the 99.5th percentile of the normal distribution; Lower control limit: ; In the formula, LCL is the lower control limit. Let SE be the global baseline mean of historical qualified CDs, and z be the standard error. 0.995 It is the 99.5th percentile of the normal distribution; Determine the threshold range: ; Step S2-5: Compare the batch mean of the batch mean sequence of the historical qualified CD dataset with the judgment threshold interval. When the batch mean of the batch mean sequence of the historical qualified CD dataset is above J% and falls within the judgment threshold interval, the judgment threshold interval is qualified. J% is set by professionals according to the business scenario.

[0027] Establish global benchmark statistics and reasonable judgment thresholds, clarify the normal fluctuation range of CD values, avoid misjudgments caused by ambiguous judgment criteria, and improve the scientific nature of subsequent offset judgments.

[0028] Step S3: Set the CD offset type. Calculate the CD offset for each batch based on the average CD value of each batch and the average of the batch average sequence of the historical qualified CD dataset. Determine the type of the CD offset. Step S3-1: Set the CD offset type to small offset and significant offset. Small offset is caused by random fluctuations, and significant drift is caused by process drift. Step S3-2: Calculate the CD offset for each batch based on the average CD value of each batch and the average of the batch average sequence of the historical qualified CD dataset. Specifically: ; In the formula, This is the CD offset. The CD mean for each batch, where μ is the mean of the batch mean sequence of the historical qualified CD dataset; Step S3-3, when When >0, the CD value is too large and offset. When <0, the CD value is too small and offset; when | When the price is within the fluctuation range, it is considered a minor offset, and no trend analysis is performed. If the value is not within the threshold range, it is considered a significant shift and should be included in trend analysis.

[0029] Distinguish between minor and significant offsets to accurately pinpoint the problem type, avoid over-processing random fluctuations that have no impact, and focus on process drift issues that require attention.

[0030] Step S4: Set a time window, calculate the Euclidean distance between the offset vectors of two adjacent time windows, quantify the offset trend change, and construct an offset prediction model based on the quantified offset trend change. Step S4-1: Plot an offset trend chart based on the batch number and CD offset, and fit the offset trend, specifically as follows: ; In the formula, Let a0, a1, a2, ..., a be the CD offset for batch t, ​​where t is the batch number.p For polynomial coefficients, This is the random error term; Step S4-2: Solve for the polynomial coefficients using the least squares method, and then substitute the polynomial coefficients into the offset trend fitting formula to construct the offset prediction model. Step S4-3: Set the time window size to m, and construct the CD offset vector using the sliding window method, specifically as follows: ; In the formula, Let be the CD offset vector for the k-th sliding time window, where k is the index of the sliding window. This represents the CD offset for the kth batch. Let m be the CD offset for the (k+1)th batch, and m be the size of the sliding time window. This is the CD offset for the last batch within this time window; Step S4-4: Calculate the window Euclidean distance for all historical qualified batches, specifically as follows: ; In the formula, Let be the Euclidean distance between the k-th adjacent sliding windows. This is the summation operator, which means summing all terms from i=1 to i=m-1. Let CD be the CD offset of the i-th batch in the k-th sliding window. This represents the CD offset for the i-th batch within the (k-1)-th sliding window. It is the square of the difference in CD offset between the batches at the i-th position in two adjacent time windows; Sort all historical qualified batches by window Euclidean distance in ascending order, and take the 95th percentile as the model activation threshold; Step S4-5: Calculate the Euclidean distance between the offset vectors of two adjacent time windows (CD). Step S4-6: When the Euclidean distance between the CD offset vectors of two adjacent time windows is less than or equal to the model activation threshold, it is determined that the offset trend is stable and the monitoring of subsequent batches continues. When the Euclidean distance between two adjacent time windows is greater than or equal to the model activation threshold, it is determined that the offset trend changes abruptly and the offset prediction model is activated to predict the CD offset of the next batch.

[0031] By monitoring trend abrupt changes through Euclidean distance over a time window and combining this with a predictive model to anticipate the CD offset of the next batch, we can shift from "passive monitoring" to "proactive early warning" and reduce the generation of defective batches.

[0032] Step S5: Construct the quantitative mapping relationship between offset and process parameters, solve for the process parameter adjustment amount, and then perform parameter adjustment based on the solved process parameter adjustment amount.

[0033] Step S5-1: Extract the key process parameters that affect CD, calculate the correlation coefficient of the key process parameters, screen the key process parameters by Pearson correlation coefficient, and include the key process parameters with |r|≥m into the mapping model. The |r| is a quantitative index that measures the linear correlation strength between the key CD and a certain process parameter, and m is set by professionals according to the business scenario. Step S5-2: Standardize the selected key process parameters using Z-score. Step S5-3: Based on the standardized q key process parameters, establish a mapping model, specifically as follows: ; In the formula, b0 is a constant term, b1,...,b q The sensitivity coefficients of the critical process parameters are P1,...Pq, where P1,...Pq is the critical process parameter sensitivity coefficient. q CD represents the CD value for the 1st, ..., qth key process parameter selected. Step S5-4: Convert the mapping model into a matrix, specifically: ; In the formula, Let P be a q×1 order CD value matrix, and let P be a q×1 order key process parameter matrix. This is a 1×q order critical process parameter sensitivity coefficient matrix. It is a q×1 order constant term matrix; The sensitivity coefficient matrix of key process parameters is solved using the least squares method. ; The predicted offset is input into the CD value matrix to obtain the process parameter adjustment matrix, and the key process parameters are adjusted according to the process parameter adjustment matrix.

[0034] By using Pearson correlation coefficient to screen key process parameters, a quantitative mapping model is established to accurately solve for the adjustment amount of process parameters, avoiding blind adjustments and improving the efficiency and accuracy of process optimization.

[0035] The system includes a qualified CD dataset construction module, a benchmark statistics and threshold setting module, a CD offset determination module, an offset trend prediction module, and a process parameter adjustment module. The qualified CD dataset construction module is used to collect historical qualified chip data. It collects CD values ​​and batch numbers by sampling wafers from each batch, removes outliers in CD values ​​using the 3σ principle, and integrates them to form a historical qualified CD dataset. The benchmark statistics and threshold setting module is used to calculate the average value of each batch of CDs based on the historical qualified CD dataset and construct a batch average value sequence to obtain a global benchmark statistic. Then, the threshold range is calculated and determined based on this statistic, and the rationality of the threshold range is verified. The CD offset determination module is used to first set two types of CD offsets, then combine the mean of the batch mean sequence of historical qualified CD datasets to calculate the CD offset of each batch, and finally determine the CD offset type of each batch based on the absolute value of the offset. The offset trend prediction module is used to fit the CD offset trend through a polynomial and build an offset prediction model. After setting a time window, it calculates the Euclidean distance between the CD offset vectors of adjacent windows to quantify the trend change. At the same time, it sets a model start threshold and determines whether to start the model based on the comparison between the Euclidean distance and the threshold to predict the CD offset of the next batch. The process parameter adjustment module is used to screen key process parameters affecting CD using Pearson correlation coefficient, construct a quantitative mapping model between CD and key process parameters, solve for the sensitivity coefficient of key process parameters, calculate the process parameter adjustment amount by combining the predicted offset, and finally perform process parameter adjustment based on the adjustment amount.

[0036] The qualified CD dataset construction module includes a data sampling and acquisition unit and a data preprocessing and integration unit. The data sampling and acquisition unit is used to collect the CD values ​​and batch numbers of each batch of historical qualified chips. The data preprocessing and integration unit is used to remove outliers in CD values ​​and integrate them to form a historical qualified CD dataset. The benchmark statistics and threshold setting module includes a global benchmark statistics unit and a threshold setting verification unit; The global benchmark statistics unit is used to calculate the batch mean sequence and global benchmark statistics of the historical qualified CD dataset; The threshold setting verification unit is used to calculate and determine the threshold range and verify its compliance.

[0037] The CD offset determination module includes an offset type setting unit and an offset calculation determination unit; The offset type setting unit is used to set two types of CD offset: small offset and significant offset; The offset calculation and determination unit is used to calculate the offset of each batch of CDs and determine their offset type. The offset trend prediction module includes a trend quantification analysis unit and a prediction model activation unit; The trend quantification analysis unit is used to fit the offset trend and quantify the trend change using Euclidean distance. The prediction model activation unit is used to set the model activation threshold and activate prediction based on trend mutations.

[0038] The process parameter adjustment module includes a mapping model construction unit and an adjustment amount solution execution unit; The mapping model construction unit is used to screen key process parameters and construct a CD-process parameter mapping model. The adjustment amount calculation and execution unit is used to calculate the process parameter adjustment amount and execute the parameter adjustment.

[0039] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for monitoring chip processing based on integrated circuit optimization, characterized by: The intelligent management method comprises the following steps: Step S1, collect historical chips, sample each batch of wafer of the chips, collect key data, pretreat the key data, and construct a historical qualified CD data set; Step S2, based on the pretreated data, calculate the global reference statistics, set the judgment threshold, and verify the rationality of the threshold; Step S3, set the CD offset type, calculate the CD offset of each batch according to the CD mean value of each batch combined with the mean value of the batch mean value sequence of the historical qualified CD data set, and determine the type of the CD offset; Step S4, set a time window, calculate the Euclidean distance of the CD offset vectors of two adjacent time windows, quantify the offset trend change, and construct an offset prediction model according to the quantified offset trend change; Step S5, construct the quantitative mapping relationship between the offset and the process parameters, solve the process parameter adjustment amount, and then execute parameter adjustment according to the solved process parameter adjustment amount.

2. The method of claim 1, wherein: The specific steps of step S1 are as follows: Step S1-1, collect N batches of qualified chips, for each batch of chips, randomly select x sampling points on the wafer in a uniform distribution manner and a manner of covering the wafer center level edge position, collect the CD value and batch number of each sampling point, and the key data is the CD value and batch number; Step S1-2, remove the outliers of the collected CD values using the 3σ principle, integrate the CD values processed by the 3σ principle, and construct a historical qualified CD data set.

3. The method of claim 2, wherein: The specific steps of step S2 are as follows: Step S2-1, calculate the CD mean value of each batch in batches, and construct a batch mean value sequence of the historical qualified CD data set; Step S2-2, calculate the mean value and standard deviation of the batch mean value sequence of the historical qualified CD data set, and set the mean value and standard deviation of the batch mean value sequence of the historical qualified CD data set as the global reference statistics; Step S2-3, calculate the standard error according to the mean value and standard deviation of the batch mean value sequence of the historical qualified CD data set, and calculate the fluctuation range combined with the quantile according to the standard error; Step S2-4, set the mean value of the batch mean value sequence of the historical qualified CD data set as the judgment threshold, and calculate the judgment threshold interval combined with the standard error; Step S2-5, compare the batch mean value of the batch mean value sequence of the historical qualified CD data set with the judgment threshold interval, and when the batch mean value of the batch mean value sequence of the historical qualified CD data set is J% above the judgment threshold interval, the judgment threshold interval is qualified, and J% is set by a professional according to the business scenario.

4. The method of claim 3, wherein: The specific steps of step S3 are as follows: Step S3-1, set the CD offset type as micro offset and significant offset, the micro offset is caused by random fluctuation, and the significant offset is caused by process drift; Step S3-2, calculate the CD offset of each batch according to the CD mean value of each batch combined with the mean value of the batch mean value sequence of the historical qualified CD data set, specifically: ； wherein is the CD offset, is the CD mean for each lot, and μ is the mean of the sequence of lot means for the historical qualified CD dataset. Step S3-3, when >0, the CD value is large offset, when <0, the CD value is small offset, when | is in the fluctuation range, it is determined as a small offset, no trend analysis is performed, when | is not in the determination threshold interval, it is determined as a significant offset, which needs to be included in the trend analysis.

5. The method of claim 4, wherein: The specific steps of step S4 are as follows: Step S4-1, draw an offset trend graph according to the batch number and the CD offset, and fit the offset trend, specifically: ； wherein is the CD offset for the tth batch, t is the batch number, a0, a1, a2,..., a p is a polynomial coefficient, is a random error term; Step S4-2, solve the polynomial coefficients by the least square method, and then substitute the polynomial coefficients into the offset trend fitting formula to construct an offset prediction model; Step S4-3, set the size of the time window as m, and use the sliding window method to construct the CD offset vector; Step S4-4, count the window Euclidean distances of all historical qualified batches, sort the window Euclidean distances of all historical qualified batches in ascending order, and take the 95% quantile as the model starting threshold; Step S4-5, calculate the Euclidean distance of the CD offset vectors of two adjacent time windows; Step S4-6, when the Euclidean distance of the CD offset vectors of two adjacent time windows is less than or equal to the model starting threshold, it is determined that the offset trend is stable, and the subsequent batches are continuously monitored; when the Euclidean distance of two adjacent time windows is greater than or equal to the model starting threshold, it is determined that the offset trend is mutated, and the offset prediction model is started to predict the CD offset of the next batch.

6. The method of claim 5, wherein: The specific steps of the step S5 are as follows: Step S5-1, extract the key process parameters affecting the CD, calculate the correlation coefficients of the key process parameters, and select the key process parameters by the Pearson correlation coefficient, and the key process parameters with |r|≥m are included in the mapping model, wherein |r| is a quantitative index for measuring the linear correlation strength between the key CD and a process parameter, and m is set by a professional according to the business scenario; Step S5-2, standardize the selected key process parameters using Z-score; Step S5-3, establish a mapping model according to the standardized q key process parameters, specifically: ； where b0is a constant term, b1,...,b q are the critical process parameter sensitivity coefficients for the 1st,...,qth critical process parameters, P1,...P q are the selected 1st,...,qth critical process parameters, and CD is the CD value. Step S5-4, convert the mapping model into a matrix, specifically: ； In the formula, is a q × 1 order CD value matrix, P is a q × 1 order key process parameter matrix, is a 1 × q order key process parameter sensitivity coefficient matrix, is a q × 1 order constant term matrix; Solving a sensitivity coefficient matrix of key process parameters using a least square method ; Input the predicted offset into the CD value matrix to obtain a process parameter adjustment matrix, and adjust the key process parameters according to the process parameter adjustment matrix.

7. A chip fabrication monitoring system based on integrated circuit optimization, characterized in that: The system includes a qualified CD dataset construction module, a reference statistics and threshold setting module, a CD offset judgment module, an offset trend prediction module, and a process parameter adjustment module; The qualified CD dataset construction module is used to collect historical qualified chip data, sample and collect CD values and batch numbers on each batch wafer, remove CD value outliers by using the 3σ principle, and integrate to form a historical qualified CD dataset; The reference statistics and threshold setting module is used to calculate the mean of each batch CD based on the historical qualified CD dataset, construct a batch mean sequence, further obtain a global reference statistic, and then calculate a judgment threshold interval, and verify the rationality of the threshold interval; The CD offset judgment module is used to first set two types of CD offsets, then calculate the CD offset of each batch in combination with the mean of the batch mean sequence of the historical qualified CD dataset, and finally determine the CD offset type of each batch according to the absolute value of the offset; The offset trend prediction module is used to fit the CD offset trend by a polynomial and construct an offset prediction model, calculate the Euclidean distance of the CD offset vectors of adjacent windows after setting the time window to quantify the trend change, set a model starting threshold, judge whether to start the model according to the comparison between the Euclidean distance and the threshold, and predict the CD offset of the next batch; The process parameter adjustment module is configured to screen key process parameters affecting CD through a Pearson correlation coefficient, construct a quantitative mapping model of CD and the key process parameters, solve a key process parameter sensitivity coefficient, calculate a process parameter adjustment amount in combination with a predicted offset, and finally execute process parameter adjustment according to the adjustment amount.

8. The system for monitoring the processing of a chip based on the optimization of an integrated circuit according to claim 7, characterized in that: The qualified CD data set construction module comprises a data sampling collection unit and a data preprocessing integration unit. The data sampling collection unit is configured to collect CD values of sampling points and batch numbers of each batch of historical qualified chips. The data preprocessing integration unit is configured to eliminate abnormal values of the CD values and integrate to form a historical qualified CD data set. The reference statistics and threshold setting module comprises a global reference statistics unit and a threshold setting verification unit. The global reference statistics unit is configured to calculate a batch mean value sequence and a global reference statistic of the historical qualified CD data set. The threshold setting verification unit is configured to calculate a judgment threshold interval and verify its eligibility.

9. The system for monitoring the processing of a chip based on the optimization of an integrated circuit according to claim 7, characterized in that: The CD offset determination module comprises an offset type setting unit and an offset amount calculation and determination unit. The offset type setting unit is configured to set two types of CD micro-offset and significant offset. The offset amount calculation and determination unit is configured to calculate CD offset amounts of each batch and determine the offset types. The offset trend prediction module comprises a trend quantitative analysis unit and a prediction model starting unit. The trend quantitative analysis unit is configured to fit an offset trend and quantize a trend change through a Euclidean distance. The prediction model starting unit is configured to set a model starting threshold and start prediction according to a trend mutation.

10. The system for monitoring the processing of a chip based on the optimization of an integrated circuit according to claim 7, characterized in that: The process parameter adjustment module comprises a mapping model construction unit and an adjustment amount solving and executing unit. The mapping model construction unit is configured to screen key process parameters and construct a CD and process parameter mapping model. The adjustment amount solving and executing unit is configured to solve a process parameter adjustment amount and execute parameter adjustment.

Images