Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Continued Examination Under
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 08/10/2026 has been entered.
Response to Arguments
Applicant's arguments filed 8/10/26 have been fully considered but they are not persuasive.
In regards to the 101 rejection, applicant argues:
“The Examiner’s mental-process characterization therefore does not address the amended claim as a whole. The claim is not directed to generally thinking about a simplified process or making a generic prediction. It requires a process-step sequence mining operation that analyzes combinatorial dependence structures among plural process-steps and plural queue-times, uncovers rules or conditions governing how those process-steps and queue-times influence characteristics of a testable product generated by the sequence, generates quality-related or yield related rules or conditions, making a prediction based on those generated rules or conditions.”
Examiner notes that the process-step sequence mining operation is not a part of the mental abstract idea and is simply using a computer as a tool to perform abstract ideas -Mere instructions to apply an exception – see MPEP 2106.05(f). Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
“Applicant respectfully submits that the claims require using the process-step sequence and prediction to provide instructions to controller components “to control how equipment processes product components to generate the testable product,” and further require manufacturing tools and assembly tools controlled to execute the process-step sequence… The claims reflect that practical application through the recited controller instructions for the manufacturing and assembly equipment”
Examiner notes that using these steps to provide instructions to control to generate a product does not actually require processing physical components into testable products as argued. The claim requires to control equipment… to generate the testable product… assembly tools to execute, all of these argument statements only require something with a purpose to do something, but none actually require generating a product or executing the sequence. Therefore examiner does not agree that the limitations add practical application, and has maintained the 101 rejection.
In regards to the 103 rejection applicant argues, “The mining operation must analyze those structures, uncover rules or conditions governing how the process steps and queue times influence characteristics of the generated testable product, generate quality related or yield related conditions, and making the prediction using those generated rules or conditions”
Examiner notes points to Honda ([0055] methods to pre-process data can include but are not limited to the following: (i) time series segmentation (including based on known manufacturing steps or automatic detection of sharp changes in signal measurements, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation, [0050] The FDC pipeline 300 can also be used as a diagnostic tool for process equipment by determining the critical sensor and manufacturing steps that are causing wafers to be abnormal. This determination can improve a root cause analysis, which is not limited to equipment part failure, sensor degradation, seasonality, etc., [0075] Some root causes of rare failure modes may not have happened yet, but are known during the design phase, while other root causes may not be detected during the design phase, but have happened in past usage. Thus, if a root cause can be obtained from both FMEA and historical data, it is useful to combine root cause from both sets. The segmented times and steps identifying key features in causing wafers to be abnormal including root causes which are not detected during the design phase influencing yield and performance, and the root causes are interpreted as the uncovered conditions.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claim 1 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim recites “making a prediction, taking into account the quality-related or yield-related rules or conditions, of an impact of one or more portions of the process-step sequence on characteristics of a testable product generated by the process-step sequence”, and “uncovering the previously unidentified rules or conditions to generate quality-related; and making a prediction, taking into account the quality-related or yield related rules or conditions, of an impact of one or more portions of the process-step sequence on characteristics of the testable product generated by the process-step sequence” which is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, nothing in the claim element precludes the step from practically being performed in the mind. For example, the language in the context of this claim encompasses that the user mentally could make a decision, observation, and calculation. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. In particular, the claim recites additional elements- “A computer-implemented method comprising”, “ performing, using the processor system, a process-step sequence mining operation on the process-step sequence comprising a plurality of process steps and a plurality of queue-times”, “wherein the process-step sequence mining operation applies analysis techniques to the process-step sequence to analyze combinatorial dependence structures among the plurality of process-steps and the plurality of queue-times”, “wherein performing the process-step sequence mining comprises:”, “using the process-step sequence and the prediction to provide instructions to controller components to control how equipment processes product components to generate the testable product”, “wherein the equipment comprises manufacturing tools and assembly tools, the controller components configured to control operations performed by the manufacturing tools and the assembly tools to execute the process-step sequence” which is simply using a computer as a tool to perform abstract ideas -Mere instructions to apply an exception – see MPEP 2106.05(f). Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Accordingly these additional elements do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of “A computer-implemented method comprising”, “ performing, using the processor system, a process-step sequence mining operation on the process-step sequence comprising a plurality of process steps and a plurality of queue-times”, “wherein the process-step sequence mining operation applies analysis techniques to the process-step sequence to analyze combinatorial dependence structures among the plurality of process-steps and the plurality of queue-times”, “wherein performing the process-step sequence mining comprises:”, “using the process-step sequence and the prediction to provide instructions to controller components to control how equipment processes product components to generate the testable product”, “wherein the equipment comprises manufacturing tools and assembly tools, the controller components configured to control operations performed by the manufacturing tools and the assembly tools to execute the process-step sequence” , which is simply using a computer as a tool to perform abstract ideas -Mere instructions to apply an exception – see MPEP 2106.05(f), which is considered to be well-understood, routine, conventional activity. Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Claim 2 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim inherits the mental abstract idea from claim 1. The claim recites “wherein the process-step sequence mining operation comprises: encoding the process-step sequence to generate an encoded process-step sequence having a plurality of encoded process-steps and a plurality of encoded queue-times; and applying, using the processor, a dimensionality reduction operation to the encoded process-step sequence” which is simply using a computer as a tool to perform abstract ideas -Mere instructions to apply an exception – see MPEP 2106.05(f). Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Claim 3 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim inherits the mental abstract idea from claim 1. The claim recites “wherein: applying the dimensionality reduction operation to the encoded process-step sequence generates a reduced-dimension encoded process-step sequence; and the process-step sequence mining operation further comprises applying the reduced-dimension encoded process-step sequence to a predictive model operable to perform a task” which is simply using a computer as a tool to perform abstract ideas -Mere instructions to apply an exception – see MPEP 2106.05(f). Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Claim 4 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim inherits the mental abstract idea from claim 1. The claim recites “wherein the task comprises the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence” which is simply using a computer as a tool to perform abstract ideas -Mere instructions to apply an exception – see MPEP 2106.05(f). Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Claim 5 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim recites “wherein making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product comprises: performing a first comparison of the reduced-dimension encoded process-step sequence with a first cluster associated with a first measurement range of the characteristics; performing a second comparison of the reduced-dimension encoded process-step sequence to a second cluster associated with a second measurement range of the characteristics; and associating the reduced-dimension encoded process-step sequence with the first cluster or the second cluster based on a result of the first comparison and a result of the second comparison” which is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, nothing in the claim element precludes the step from practically being performed in the mind. For example the language in the context of this claim encompasses that the user mentally could make a decision, observation, and calculation. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea. Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Claim 6 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim recites “to predict a portion of the process-step sequence having a positive impact on the characteristics of the testable product” which is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, nothing in the claim element precludes the step from practically being performed in the mind. For example the language in the context of this claim encompasses that the user mentally could make a decision, observation, and calculation. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea. The claim additionally recites “comprising using a pattern sequence extraction module of the processor system” which falls under field of use and technological environment- see MPEP 2106.05(h) Parker v. Flook ("Flook established that limiting an abstract idea to one field of use or adding token postsolution components did not make the concept patentable"). Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Claim 7 is rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim recites “the characteristics are selected from the group consisting of wafer yield, die yield, wafer quality, and die quality… and making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence comprises evaluating the plurality of symbols against a process-step sequence language domain” which is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. That is, nothing in the claim element precludes the step from practically being performed in the mind. For example the language in the context of this claim encompasses that the user mentally could make a decision, observation, and calculation. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea. The claim additionally recites “the plurality of process-steps comprises a plurality of semiconductor product fabrication operations “, and “the testable product comprises a wafer having dies and completed integrated circuitry ready for testing” which falls under field of use and technological environment- see MPEP 2106.05(h) Parker v. Flook ("Flook established that limiting an abstract idea to one field of use or adding token postsolution components did not make the concept patentable"). The claim recites “encoding the process-step sequence to generate the encoded process-step sequence comprise converting the plurality of process-steps and the plurality of queue-times to a plurality of symbols; the dimensionality reduction operation comprises an embedding operation” which is simply using a computer as a tool to perform abstract ideas -Mere instructions to apply an exception – see MPEP 2106.05(f). Therefore these do not integrate a judicial exception into a practical application or provide significantly more. The claim is not patent eligible.
Claim 8 is rejected under 35 U.S.C. 101 for similar reason to claim 1.
Claim 9 is rejected under 35 U.S.C. 101 for similar reason to claim 2.
Claim 10 is rejected under 35 U.S.C. 101 for similar reason to claim 3.
Claim 11 is rejected under 35 U.S.C. 101 for similar reason to claim 4.
Claim 12 is rejected under 35 U.S.C. 101 for similar reason to claim 5.
Claim 13 is rejected under 35 U.S.C. 101 for similar reason to claim 6.
Claim 14 is rejected under 35 U.S.C. 101 for similar reason to claim 7.
Claim 15 is rejected under 35 U.S.C. 101 for similar reason to claim 1.
Claim 16 is rejected under 35 U.S.C. 101 for similar reason to claim 2.
Claim 17 is rejected under 35 U.S.C. 101 for similar reason to claim 3 & 4.
Claim 18 is rejected under 35 U.S.C. 101 for similar reason to claim 5.
Claim 19 is rejected under 35 U.S.C. 101 for similar reason to claim 6.
Claim 20 is rejected under 35 U.S.C. 101 for similar reason to claim 7.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Honda et al (US20220327268, herein Honda), in view of Norman et al. (US20220026891, herein Norman).
Regarding claim 1, Honda teaches A computer-implemented method comprising: performing, using the processor system, a process-step sequence mining operation on the process-step sequence comprising a plurality of process steps and a plurality of …times([0026] Wafer fabrication occurs in step 102, wherein a large number of integrated circuits are formed on a single slice of semiconductor substrate, such as silicon, known as a wafer. Many steps are required in various sequences to build different integrated circuits, [0025] Fig. 1 …semiconductor manufacturing process 100… voluminous amounts of data that represent various aspects of the process can be collected at every step and sub-step of a production run and provided as input data to various forms of systematic analysis. For example, yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation); wherein the process-step sequence mining operation applies analysis techniques to the process-step sequence to analyze combinatorial dependence structures among the plurality of process-steps and the plurality of … times ([0004] One potential application of ML models is for the timely detection and identification of problematic processing events and/or equipment-related failures, [0025] Fig. 1 …semiconductor manufacturing process 100… voluminous amounts of data that represent various aspects of the process can be collected at every step and sub-step of a production run and provided as input data to various forms of systematic analysis. For example, yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation); wherein the plurality of process-steps … comprise process-steps and … times having previously unidentified rules or conditions that govern an influence of the process-steps and the… times on characteristics of a testable product generated by the process-step sequence; ([0055] methods to pre-process data can include but are not limited to the following: (i) time series segmentation (including based on known manufacturing steps or automatic detection of sharp changes in signal measurements, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation, [0050] The FDC pipeline 300 can also be used as a diagnostic tool for process equipment by determining the critical sensor and manufacturing steps that are causing wafers to be abnormal. This determination can improve a root cause analysis, which is not limited to equipment part failure, sensor degradation, seasonality, etc., [0075] Some root causes of rare failure modes may not have happened yet, but are known during the design phase, while other root causes may not be detected during the design phase, but have happened in past usage. Thus, if a root cause can be obtained from both FMEA and historical data, it is useful to combine root cause from both sets.) (i.e. segmented times and steps identifying key features in causing wafers to be abnormal including root causes which are not detected during the design phase influencing yield and performance) ); wherein performing the process-step sequence mining operation comprises: uncovering the previously unidentified rules or conditions to generate quality-related or yield-related rules or conditions, ([0075] Some root causes of rare failure modes may not have happened yet, but are known during the design phase, while other root causes may not be detected during the design phase, but have happened in past usage. Thus, if a root cause can be obtained from both FMEA and historical data, it is useful to combine root cause from both sets, [0014] FIG. 8 is a flow chart illustrating a process for determining the most probable root cause of an excursion, [0025] yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process, [0050] The FDC pipeline 300 can also be used as a diagnostic tool for process equipment by determining the critical sensor and manufacturing steps that are causing wafers to be abnormal. This determination can improve a root cause analysis, which is not limited to equipment part failure, sensor degradation, seasonality, etc.); and making a prediction, taking into account the quality-related or yield-related rules or conditions, of an impact of one or more portions of the process-step sequence on characteristics of the testable product generated by the process-step sequence, ([0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0087] a robust predictive model with rational decision making that is production-worthy is illustrated in FIG. 9A), [0099] the optimal model and relevant effective thresholds are selected based on the consideration of trade-offs from the customer survey. The results of different combinations of limits and thresholds for different parameters can be presented graphically or as data tables in the GUI in order to illustrate for the customer the impact of changes in selection criteria. For example, if returning a false positive for chip has an economic impact that is 100 times more costly than returning a true positive, then model A 1001 on FIG. 10 should be selected and thresholds set that produce 40% true positive and less than 0.5% false positives. However, if returning a true positive is considered 100 times more important than returning a false positive for any rationale, the model B 1002 on FIG. 10 should be selected and thresholds set that produce 30% false positive and greater than 99.5% true positives, [0085] This description provides a method for guiding a user in rationally selecting predictive models for semiconductor manufacturing applications, [0113] Regression techniques can also be effective for defining the relationships between dependent variables and independent variables when modeling a target feature. As an example, predicting yield for low-yielding wafers may be identified by the customer as more critical than predicting yield for high-yielding wafers, since correction of the issue causing the low yield has a more significant impact on overall yield and cost. Further, since this issue is usually a function of equipment failure or wear, a model to predict remaining useful life (RUL) for processing equipment can be very useful for the customer and effectively modeled using regression techniques); and using the process-step sequence and the prediction to provide instructions to controller components to control how equipment processes product components to generate the testable product ([0085] This description provides a method for guiding a user in rationally selecting predictive models for semiconductor manufacturing applications, [0051] The FDC pipeline 300 can be used as a prognostic/preventive maintenance tool for process equipment by combining temporal trends in wafer prediction with drift and shift data in sensor measurements and ML features); wherein the equipment comprises manufacturing tools and assembly tools, the controller components configured to control operations performed by the manufacturing tools and the assembly tools to execute the process-step sequence ([0106] For example, FIG. 16 illustrates a process 1600 for using the decision-making tool to evaluate equipment issues. After building a model with customer preferences and limits and deploying the same into production at step 1601, an equipment failure issue arises and the first question in step 1602 is whether the failure is catastrophic. If so, then in step 1604 an immediate process shutdown is called for and the problem corrected. If the equipment failure is not catastrophic, then an inspection of the equipment is undertaken in step 1606, [0069] semiconductor tools, [0110] early detection/prediction of catastrophic failure is obviously important for any manufacturing operation, [0033] Standard statistical and process control techniques were used to analyze and utilize the datasets to improve yields and manufacturing efficiencies ).
Honda does not teach and a plurality of queue-times
Norman teaches and a plurality of queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period, [0021] Based on the additional simulation output, the processing device can identify additional candidate substrates (e.g., from the substrate queue) and can initiate the set of operations to process the number of candidate substrates from the simulation output and the additional candidate substrates over the time period, [0056] a second number of candidate substrates that were successfully processed during each of the simulated set of operations to reach the end of the second time period, [0008] first number of candidate substrates over the first time period)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Honda’s teaching a predictive model for semiconductor processes with Norman’s teaching of the semiconductor processes including the queue times. The combined teaching provides an expected result of a predictive model for semiconductor processes including the queue times. Therefore, one of ordinary skill in the art would be motivated to optimize the prediction model as disclosed by Norman [0004] enable yield enhancement, improvement of process quality control, reduce manufacturing costs and amount of scrap, as well as improve equipment uptime by identifying systematic issues quickly.
Regarding claim 2, the combination of Honda and Norman teach The computer-implemented method of claim 1, wherein the process-step sequence mining operation comprises: encoding the process-step sequence to generate an encoded process-step sequence having a plurality of encoded process-steps…; and applying, using the processor, a dimensionality reduction operation to the encoded process-step sequence (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by… we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder)
Norman further teaches and a plurality of encoded queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period, [0021] Based on the additional simulation output, the processing device can identify additional candidate substrates (e.g., from the substrate queue) and can initiate the set of operations to process the number of candidate substrates from the simulation output and the additional candidate substrates over the time period, [0056] a second number of candidate substrates that were successfully processed during each of the simulated set of operations to reach the end of the second time period, [0008] first number of candidate substrates over the first time period)
Regarding claim 3, the combination of Honda and Norman teach The computer-implemented method of claim 2, wherein: applying the dimensionality reduction operation to the encoded process-step sequence generates a reduced-dimension encoded process-step sequence; and the process-step sequence mining operation further comprises applying the reduced-dimension encoded process-step sequence to a predictive model operable to perform a task (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder)
Regarding claim 4, the combination of Honda and Norman teach The computer-implemented method of claim 3, wherein the task comprises the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder)
Regarding claim 5, the combination of Honda and Norman teach The computer-implemented method of claim 4, wherein making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product comprises: performing a first comparison of the reduced-dimension encoded process-step sequence with a first cluster associated with a first measurement range of the characteristics; performing a second comparison of the reduced-dimension encoded process-step sequence to a second cluster associated with a second measurement range of the characteristics; and associating the reduced-dimension encoded process-step sequence with the first cluster or the second cluster based on a result of the first comparison and a result of the second comparison (Honda, [0088] In step 904, a predictive model is generated for predicting relevant target features. If the results from the predictive model meet the minimum performance criteria in step 905, such as False Positive Rate (FPR) of 0.1% and False Negative Rate (FNR) of 1%, the model is saved in step 906 for use in the comparative analysis described below. Other performance criteria could be used, including Area under the Curve (AUC), Area under the Precision-Recall curve, F1 Score, Skip Rate, etc. In step 907, if additional predictive models are needed for an adequate analysis, another predictive model is generated for predicting relevant target features using a different methodology in step 912, and then compared as before to the minimum performance criteria in step 905. It should be noted that different methodologies include different algorithms and mathematical functions including statistical solutions that may be implemented as machine learning models with supervised learning and reinforced learning, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder)
Regarding claim 6, the combination of Honda and Norman teach. The computer-implemented method of claim 5 further comprising using a pattern sequence extraction module of the processor system to predict a portion of the process-step sequence having a positive impact on the characteristics of the testable product (Honda, [0087] The basic steps for building a first embodiment of a robust predictive model with rational decision making that is production-worthy is illustrated in FIG. 9A. In step 902, minimum performance criteria for the target feature(s) are obtained for a predictive model (including a machine learning model, pattern recognition model, physics-based Model, hybrid model, etc.). These criteria include but are not limited to training speed, prediction speed, accuracy, false positive rate, false negative rate, recall, precision, F1 score, F2 score, F-N score, root mean square error, root mean square log error, mean absolute error, worst case error, robustness, generalizability, etc., [0076] For example, log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features. Temporal data analysis could be transformed using a time series analysis like Autoregressive Integrated Moving Average (ARIMA), Kalman Filter, Particle Filter, etc. to extract relevant features)
Regarding claim 7, the combination of Honda and Norman teach The computer-implemented method of claim 4, wherein: the plurality of process-steps comprises a plurality of semiconductor product fabrication operations (Honda, [0026] Wafer fabrication occurs in step 102, wherein a large number of integrated circuits are formed on a single slice of semiconductor substrate, such as silicon, known as a wafer); the testable product comprises a wafer having dies and completed integrated circuitry ready for testing ([0029] The wafer is diced up into separate individual circuits or dies, and each die that passes through wafer sort and test is bonded to and electrically connected to a frame to form a package. Each die/package is then encapsulated to protect the circuit); the characteristics are selected from the group consisting of wafer yield, die yield, wafer quality, and die quality ([0039] the input data is analyzed using machine learning techniques to identify key features and/or characteristics of the data in order to classify wafers and/or lots of semiconductor devices. In step 206, the wafers or devices are classified based on the analysis of step 204, [0099] the optimal model and relevant effective thresholds are selected based on the consideration of trade-offs from the customer survey. The results of different combinations of limits and thresholds for different parameters can be presented graphically or as data tables in the GUI in order to illustrate for the customer the impact of changes in selection criteria. For example, if returning a false positive for chip has an economic impact that is 100 times more costly than returning a true positive, then model A 1001 on FIG. 10 should be selected and thresholds set that produce 40% true positive and less than 0.5% false positives. However, if returning a true positive is considered 100 times more important than returning a false positive for any rationale, the model B 1002 on FIG. 10 should be selected and thresholds set that produce 30% false positive and greater than 99.5% true positives, [0102] a key question for the customer is how much yield loss is acceptable? Further, would additional visual inspection, triggered by the predictive model, detect equipment issues earlier, such as a few hours, a few days, or few weeks earlier?) ; encoding the process-step sequence to generate the encoded process-step sequence comprise converting the plurality of process-steps [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder) … to a plurality of symbols ([0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model); the dimensionality reduction operation comprises an embedding operation; and making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence comprises evaluating the plurality of symbols against a process-step sequence language domain ([0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model).
Norman further teaches and a plurality of queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period, [0021] Based on the additional simulation output, the processing device can identify additional candidate substrates (e.g., from the substrate queue) and can initiate the set of operations to process the number of candidate substrates from the simulation output and the additional candidate substrates over the time period, [0056] a second number of candidate substrates that were successfully processed during each of the simulated set of operations to reach the end of the second time period, [0008] first number of candidate substrates over the first time period)
Regarding claim 8, Honda teaches A computer system comprising a memory communicatively coupled to a processor system, wherein the processor system is configured to perform processor system operations comprising: …(Claim 22 A robust predictive model for predicting a target feature in a semiconductor process including a processor and non-transitory storage for program instruction, the program instructions configured for causing the processor to,[0025] Fig. 1 …semiconductor manufacturing process 100… voluminous amounts of data that represent various aspects of the process can be collected at every step and sub-step of a production run and provided as input data to various forms of systematic analysis. For example, yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process); performing a process-step sequence mining operation to the process-step sequence comprising a plurality of process steps and a plurality of …times ([0026] Wafer fabrication occurs in step 102, wherein a large number of integrated circuits are formed on a single slice of semiconductor substrate, such as silicon, known as a wafer. Many steps are required in various sequences to build different integrated circuits); wherein the process-step sequence mining operation applies analysis techniques to the process-step sequence to analyze combinatorial dependence structures among the plurality of process-steps and the plurality of …times ([0004] One potential application of ML models is for the timely detection and identification of problematic processing events and/or equipment-related failures, [0025] Fig. 1 …semiconductor manufacturing process 100… voluminous amounts of data that represent various aspects of the process can be collected at every step and sub-step of a production run and provided as input data to various forms of systematic analysis. For example, yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation); wherein the plurality of process-steps and the plurality of …times comprise process-steps and …times having previously unidentified rules or conditions that govern an influence of the process-steps and the …times on characteristics of a testable product generated by the process-step sequence; ([0055] methods to pre-process data can include but are not limited to the following: (i) time series segmentation (including based on known manufacturing steps or automatic detection of sharp changes in signal measurements, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation, [0050] The FDC pipeline 300 can also be used as a diagnostic tool for process equipment by determining the critical sensor and manufacturing steps that are causing wafers to be abnormal. This determination can improve a root cause analysis, which is not limited to equipment part failure, sensor degradation, seasonality, etc., [0075] Some root causes of rare failure modes may not have happened yet, but are known during the design phase, while other root causes may not be detected during the design phase, but have happened in past usage. Thus, if a root cause can be obtained from both FMEA and historical data, it is useful to combine root cause from both sets.)wherein performing the process-step sequence mining operation comprise: uncovering the previously unidentified rules or conditions to generate quality-related; ([0075] Some root causes of rare failure modes may not have happened yet, but are known during the design phase, while other root causes may not be detected during the design phase, but have happened in past usage. Thus, if a root cause can be obtained from both FMEA and historical data, it is useful to combine root cause from both sets, [0014] FIG. 8 is a flow chart illustrating a process for determining the most probable root cause of an excursion, [0025] yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process, [0050] The FDC pipeline 300 can also be used as a diagnostic tool for process equipment by determining the critical sensor and manufacturing steps that are causing wafers to be abnormal. This determination can improve a root cause analysis, which is not limited to equipment part failure, sensor degradation, seasonality, etc.); and making a prediction, taking into account the quality-related or yield-related rules or conditions, of an impact of one or more portions of the process-step sequence on a characteristics of the testable product generated by the process-step sequence (([0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0087] a robust predictive model with rational decision making that is production-worthy is illustrated in FIG. 9A), [0099] the optimal model and relevant effective thresholds are selected based on the consideration of trade-offs from the customer survey. The results of different combinations of limits and thresholds for different parameters can be presented graphically or as data tables in the GUI in order to illustrate for the customer the impact of changes in selection criteria. For example, if returning a false positive for chip has an economic impact that is 100 times more costly than returning a true positive, then model A 1001 on FIG. 10 should be selected and thresholds set that produce 40% true positive and less than 0.5% false positives. However, if returning a true positive is considered 100 times more important than returning a false positive for any rationale, the model B 1002 on FIG. 10 should be selected and thresholds set that produce 30% false positive and greater than 99.5% true positives, [0085] This description provides a method for guiding a user in rationally selecting predictive models for semiconductor manufacturing applications, [0113] Regression techniques can also be effective for defining the relationships between dependent variables and independent variables when modeling a target feature. As an example, predicting yield for low-yielding wafers may be identified by the customer as more critical than predicting yield for high-yielding wafers, since correction of the issue causing the low yield has a more significant impact on overall yield and cost. Further, since this issue is usually a function of equipment failure or wear, a model to predict remaining useful life (RUL) for processing equipment can be very useful for the customer and effectively modeled using regression techniques) and using the process-step sequence and the prediction to provide instructions to controller components to control how equipment processes product components to generate the testable product ([0085] This description provides a method for guiding a user in rationally selecting predictive models for semiconductor manufacturing applications, [0051] The FDC pipeline 300 can be used as a prognostic/preventive maintenance tool for process equipment by combining temporal trends in wafer prediction with drift and shift data in sensor measurements and ML features)); wherein the equipment comprises manufacturing tools and assembly tools, the controller components configured to control operations performed by the manufacturing tools and the assembly tools to execute the process-step sequence ([0106] For example, FIG. 16 illustrates a process 1600 for using the decision-making tool to evaluate equipment issues. After building a model with customer preferences and limits and deploying the same into production at step 1601, an equipment failure issue arises and the first question in step 1602 is whether the failure is catastrophic. If so, then in step 1604 an immediate process shutdown is called for and the problem corrected. If the equipment failure is not catastrophic, then an inspection of the equipment is undertaken in step 1606, [0069] semiconductor tools, [0110] early detection/prediction of catastrophic failure is obviously important for any manufacturing operation, [0033] Standard statistical and process control techniques were used to analyze and utilize the datasets to improve yields and manufacturing efficiencies ).
Honda does not teach and a plurality of queue-times
Norman teaches and a plurality of queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period, [0021] Based on the additional simulation output, the processing device can identify additional candidate substrates (e.g., from the substrate queue) and can initiate the set of operations to process the number of candidate substrates from the simulation output and the additional candidate substrates over the time period, [0056] a second number of candidate substrates that were successfully processed during each of the simulated set of operations to reach the end of the second time period, [0008] first number of candidate substrates over the first time period)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Honda’s teaching a predictive model for semiconductor processes with Norman’s teaching of the semiconductor processes including the queue times. The combined teaching provides an expected result of a predictive model for semiconductor processes including the queue times. Therefore, one of ordinary skill in the art would be motivated to optimize the prediction model as disclosed by Norman [0004] enable yield enhancement, improvement of process quality control, reduce manufacturing costs and amount of scrap, as well as improve equipment uptime by identifying systematic issues quickly.
Regarding claim 9, the combination of Honda and Norman teach The computer system of claim 8, wherein the process-step sequence mining operation comprises: encoding the process-step sequence to generate an encoded process-step sequence having a plurality of encoded process-steps and a plurality of encoded… ; and applying, using the processor, a dimensionality reduction operation to the encoded process-step sequence (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by… we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder).
Norman further teaches and a plurality of encoded queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period)
Regarding claim 10, the combination of Honda and Norman teach The computer system of claim 9, wherein: applying the dimensionality reduction operation to the encoded process-step sequence generates a reduced-dimension encoded process-step sequence; and the process-step sequence mining operation further comprises applying the reduced-dimension encoded process-step sequence to a predictive model operable to perform a task (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder).
Regarding claim 11, the combination of Honda and Norman teach. The computer system of claim 10, wherein the task comprises the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder).
Regarding claim 12, the combination of Honda and Norman teach. The computer system of claim 11, wherein making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product comprises: performing a first comparison of the reduced-dimension encoded process-step sequence with a first cluster associated with a first measurement range of the characteristics; performing a second comparison of the reduced-dimension encoded process-step sequence to a second cluster associated with a second measurement range of the characteristics; and associating the reduced-dimension encoded process-step sequence with the first cluster or the second cluster based on a result of the first comparison and a result of the second comparison (Honda, [0088] In step 904, a predictive model is generated for predicting relevant target features. If the results from the predictive model meet the minimum performance criteria in step 905, such as False Positive Rate (FPR) of 0.1% and False Negative Rate (FNR) of 1%, the model is saved in step 906 for use in the comparative analysis described below. Other performance criteria could be used, including Area under the Curve (AUC), Area under the Precision-Recall curve, F1 Score, Skip Rate, etc. In step 907, if additional predictive models are needed for an adequate analysis, another predictive model is generated for predicting relevant target features using a different methodology in step 912, and then compared as before to the minimum performance criteria in step 905. It should be noted that different methodologies include different algorithms and mathematical functions including statistical solutions that may be implemented as machine learning models with supervised learning and reinforced learning, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder).
Regarding claim 13, the combination of Honda and Norman teach. The computer system of claim 12, wherein the processor system operations further comprise using a pattern sequence extraction module of the processor system to predict a portion of the process-step sequence having a positive impact on the characteristics of the testable product (Honda, [0087] The basic steps for building a first embodiment of a robust predictive model with rational decision making that is production-worthy is illustrated in FIG. 9A. In step 902, minimum performance criteria for the target feature(s) are obtained for a predictive model (including a machine learning model, pattern recognition model, physics-based Model, hybrid model, etc.). These criteria include but are not limited to training speed, prediction speed, accuracy, false positive rate, false negative rate, recall, precision, F1 score, F2 score, F-N score, root mean square error, root mean square log error, mean absolute error, worst case error, robustness, generalizability, etc., [0076] For example, log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features. Temporal data analysis could be transformed using a time series analysis like Autoregressive Integrated Moving Average (ARIMA), Kalman Filter, Particle Filter, etc. to extract relevant features)
Regarding claim 14, the combination of Honda and Norman teach. The computer system of claim 11, wherein: the plurality of process-steps comprises a plurality of semiconductor product fabrication operations (Honda, [0026] Wafer fabrication occurs in step 102, wherein a large number of integrated circuits are formed on a single slice of semiconductor substrate, such as silicon, known as a wafer); the testable product comprises a wafer having dies and completed integrated circuitry ready for testing ([0029] The wafer is diced up into separate individual circuits or dies, and each die that passes through wafer sort and test is bonded to and electrically connected to a frame to form a package. Each die/package is then encapsulated to protect the circuit); the characteristics is selected from the group consisting of wafer yield, die yield, wafer quality, and die quality ([0039] the input data is analyzed using machine learning techniques to identify key features and/or characteristics of the data in order to classify wafers and/or lots of semiconductor devices. In step 206, the wafers or devices are classified based on the analysis of step 204, [0099] the optimal model and relevant effective thresholds are selected based on the consideration of trade-offs from the customer survey. The results of different combinations of limits and thresholds for different parameters can be presented graphically or as data tables in the GUI in order to illustrate for the customer the impact of changes in selection criteria. For example, if returning a false positive for chip has an economic impact that is 100 times more costly than returning a true positive, then model A 1001 on FIG. 10 should be selected and thresholds set that produce 40% true positive and less than 0.5% false positives. However, if returning a true positive is considered 100 times more important than returning a false positive for any rationale, the model B 1002 on FIG. 10 should be selected and thresholds set that produce 30% false positive and greater than 99.5% true positives, [0102] a key question for the customer is how much yield loss is acceptable? Further, would additional visual inspection, triggered by the predictive model, detect equipment issues earlier, such as a few hours, a few days, or few weeks earlier?); encoding the process-step sequence to generate the encoded process-step sequence comprise converting the plurality of process-steps [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder) … to a plurality of symbols ([0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model); the dimensionality reduction operation comprises an embedding operation; and making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence comprises evaluating the plurality of symbols against a process-step sequence language domain ([0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model).
Norman further teaches and a plurality of queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period)
Regarding claim 15, Honda teaches A computer program product analyzing a process-step sequence, the computer program product comprising a computer readable program stored on a computer readable storage medium, wherein the computer readable program, when executed on a processor system, causes the processor system to perform processor system operations comprising: … performing a process-step sequence mining operation on the process-step sequence comprising a plurality process-steps and a plurality of …times; ([0025] Fig. 1 …semiconductor manufacturing process 100… voluminous amounts of data that represent various aspects of the process can be collected at every step and sub-step of a production run and provided as input data to various forms of systematic analysis. For example, yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process); wherein the process-step sequence mining operation applies analysis techniques to the process-step sequence to analyze combinatorial dependence structures among the plurality of process-steps and the plurality of …times ([0004] One potential application of ML models is for the timely detection and identification of problematic processing events and/or equipment-related failures, [0025] Fig. 1 …semiconductor manufacturing process 100… voluminous amounts of data that represent various aspects of the process can be collected at every step and sub-step of a production run and provided as input data to various forms of systematic analysis. For example, yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation); wherein the plurality of process-steps and the plurality of …times comprise process-steps and … times having previously unidentified rules or conditions that govern an influence of the process-steps and the …times on characteristics of a testable product generated by the process-step sequence; ([0055] methods to pre-process data can include but are not limited to the following: (i) time series segmentation (including based on known manufacturing steps or automatic detection of sharp changes in signal measurements, [0055] methods to pre-process data can include but are not limited to the following: time series segmentation, [0050] The FDC pipeline 300 can also be used as a diagnostic tool for process equipment by determining the critical sensor and manufacturing steps that are causing wafers to be abnormal. This determination can improve a root cause analysis, which is not limited to equipment part failure, sensor degradation, seasonality, etc., [0075] Some root causes of rare failure modes may not have happened yet, but are known during the design phase, while other root causes may not be detected during the design phase, but have happened in past usage. Thus, if a root cause can be obtained from both FMEA and historical data, it is useful to combine root cause from both sets.) (i.e. segmented times and steps identifying key features in causing wafers to be abnormal including root causes which are not detected during the design phase influencing yield and performance); wherein performing the process-step sequence mining operation comprises: uncovering the previously unidentified rules or conditions to generate quality-related or yield-related rules or conditions; ([0075] Some root causes of rare failure modes may not have happened yet, but are known during the design phase, while other root causes may not be detected during the design phase, but have happened in past usage. Thus, if a root cause can be obtained from both FMEA and historical data, it is useful to combine root cause from both sets, [0014] FIG. 8 is a flow chart illustrating a process for determining the most probable root cause of an excursion, [0025] yield and other performance characteristics may be calculated from selected input data for each step, as well as predictions made for key processing parameters for the entirety of the process, [0050] The FDC pipeline 300 can also be used as a diagnostic tool for process equipment by determining the critical sensor and manufacturing steps that are causing wafers to be abnormal. This determination can improve a root cause analysis, which is not limited to equipment part failure, sensor degradation, seasonality, etc.); and making a prediction, taking into account the quality-related or yield-related rules or conditions, of an impact of one or more portions of the process-step sequence on characteristics of the testable product generated by the process-step sequence ([0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0087] a robust predictive model with rational decision making that is production-worthy is illustrated in FIG. 9A), [0099] the optimal model and relevant effective thresholds are selected based on the consideration of trade-offs from the customer survey. The results of different combinations of limits and thresholds for different parameters can be presented graphically or as data tables in the GUI in order to illustrate for the customer the impact of changes in selection criteria. For example, if returning a false positive for chip has an economic impact that is 100 times more costly than returning a true positive, then model A 1001 on FIG. 10 should be selected and thresholds set that produce 40% true positive and less than 0.5% false positives. However, if returning a true positive is considered 100 times more important than returning a false positive for any rationale, the model B 1002 on FIG. 10 should be selected and thresholds set that produce 30% false positive and greater than 99.5% true positives, [0085] This description provides a method for guiding a user in rationally selecting predictive models for semiconductor manufacturing applications, [0113] Regression techniques can also be effective for defining the relationships between dependent variables and independent variables when modeling a target feature. As an example, predicting yield for low-yielding wafers may be identified by the customer as more critical than predicting yield for high-yielding wafers, since correction of the issue causing the low yield has a more significant impact on overall yield and cost. Further, since this issue is usually a function of equipment failure or wear, a model to predict remaining useful life (RUL) for processing equipment can be very useful for the customer and effectively modeled using regression techniques); and using the process-step sequence and the prediction to provide instructions to controller components to control how equipment processes product components to generate the testable product ([0085] This description provides a method for guiding a user in rationally selecting predictive models for semiconductor manufacturing applications, [0051] The FDC pipeline 300 can be used as a prognostic/preventive maintenance tool for process equipment by combining temporal trends in wafer prediction with drift and shift data in sensor measurements and ML features); wherein the equipment comprises manufacturing tools and assembly tools, the controller components configured to control operations performed by the manufacturing tools and the assembly tools to execute the process-step sequence ([0106] For example, FIG. 16 illustrates a process 1600 for using the decision-making tool to evaluate equipment issues. After building a model with customer preferences and limits and deploying the same into production at step 1601, an equipment failure issue arises and the first question in step 1602 is whether the failure is catastrophic. If so, then in step 1604 an immediate process shutdown is called for and the problem corrected. If the equipment failure is not catastrophic, then an inspection of the equipment is undertaken in step 1606, [0069] semiconductor tools, [0110] early detection/prediction of catastrophic failure is obviously important for any manufacturing operation, [0033] Standard statistical and process control techniques were used to analyze and utilize the datasets to improve yields and manufacturing efficiencies ).
Honda does not teach and a plurality of queue-times
Norman teaches and a plurality of queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period, [0021] Based on the additional simulation output, the processing device can identify additional candidate substrates (e.g., from the substrate queue) and can initiate the set of operations to process the number of candidate substrates from the simulation output and the additional candidate substrates over the time period, [0056] a second number of candidate substrates that were successfully processed during each of the simulated set of operations to reach the end of the second time period, [0008] first number of candidate substrates over the first time period)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Honda’s teaching a predictive model for semiconductor processes with Norman’s teaching of the semiconductor processes including the queue times. The combined teaching provides an expected result of a predictive model for semiconductor processes including the queue times. Therefore, one of ordinary skill in the art would be motivated to optimize the prediction model as disclosed by Norman [0004] enable yield enhancement, improvement of process quality control, reduce manufacturing costs and amount of scrap, as well as improve equipment uptime by identifying systematic issues quickly.
Regarding claim 16, the combination of Honda and Norman teach. The computer program product of claim 15, wherein the process-step sequence mining operation comprises: encoding the process-step sequence to generate an encoded process-step sequence having a plurality of encoded process-steps and a plurality of encoded …; and applying, using the processor, a dimensionality reduction operation to the encoded process-step sequence (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by… we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder)
Norman further teaches and a plurality of … queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period)
Regarding claim 17, the combination of Honda and Norman teach The computer program product of claim 16, wherein: applying the dimensionality reduction operation to the encoded process-step sequence generates a reduced-dimension encoded process-step sequence; the process-step sequence mining operation further comprises applying the reduced-dimension encoded process-step sequence to a predictive model operable to perform a task (Honda, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder) ; and the task comprises the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence ([0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder).
Regarding claim 18, the combination of Honda and Norman teach The computer program product of claim 17, wherein making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product comprises: performing a first comparison of the reduced-dimension encoded process-step sequence with a first cluster associated with a first measurement range of the characteristics; performing a second comparison of the reduced-dimension encoded process-step sequence to a second cluster associated with a second measurement range of the characteristics; and associating the reduced-dimension encoded process-step sequence with the first cluster or the second cluster based on a result of the first comparison and a result of the second comparison (Honda, [0088] In step 904, a predictive model is generated for predicting relevant target features. If the results from the predictive model meet the minimum performance criteria in step 905, such as False Positive Rate (FPR) of 0.1% and False Negative Rate (FNR) of 1%, the model is saved in step 906 for use in the comparative analysis described below. Other performance criteria could be used, including Area under the Curve (AUC), Area under the Precision-Recall curve, F1 Score, Skip Rate, etc. In step 907, if additional predictive models are needed for an adequate analysis, another predictive model is generated for predicting relevant target features using a different methodology in step 912, and then compared as before to the minimum performance criteria in step 905. It should be noted that different methodologies include different algorithms and mathematical functions including statistical solutions that may be implemented as machine learning models with supervised learning and reinforced learning, [0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder)
Regarding claim 19, the combination of Honda and Norman teach The computer program product of claim 18, wherein the processor system operations further comprise using a pattern sequence extraction module of the processor system to predict a portion of the process-step sequence having a positive impact on the characteristics of the testable product (Honda, [0087] The basic steps for building a first embodiment of a robust predictive model with rational decision making that is production-worthy is illustrated in FIG. 9A. In step 902, minimum performance criteria for the target feature(s) are obtained for a predictive model (including a machine learning model, pattern recognition model, physics-based Model, hybrid model, etc.). These criteria include but are not limited to training speed, prediction speed, accuracy, false positive rate, false negative rate, recall, precision, F1 score, F2 score, F-N score, root mean square error, root mean square log error, mean absolute error, worst case error, robustness, generalizability, etc., [0076] For example, log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features. Temporal data analysis could be transformed using a time series analysis like Autoregressive Integrated Moving Average (ARIMA), Kalman Filter, Particle Filter, etc. to extract relevant features)
Regarding claim 20, the combination of Honda and Norman teach The computer program product of claim 17, wherein: the plurality of process-steps comprises a plurality of semiconductor product fabrication operations (Honda, [0026] Wafer fabrication occurs in step 102, wherein a large number of integrated circuits are formed on a single slice of semiconductor substrate, such as silicon, known as a wafer); the testable product comprises a wafer having dies and completed integrated circuitry ready for testing ([0029] The wafer is diced up into separate individual circuits or dies, and each die that passes through wafer sort and test is bonded to and electrically connected to a frame to form a package. Each die/package is then encapsulated to protect the circuit); the characteristics are selected from the group consisting of wafer yield, die yield, wafer quality, and die quality ([0039] the input data is analyzed using machine learning techniques to identify key features and/or characteristics of the data in order to classify wafers and/or lots of semiconductor devices. In step 206, the wafers or devices are classified based on the analysis of step 204, [0099] the optimal model and relevant effective thresholds are selected based on the consideration of trade-offs from the customer survey. The results of different combinations of limits and thresholds for different parameters can be presented graphically or as data tables in the GUI in order to illustrate for the customer the impact of changes in selection criteria. For example, if returning a false positive for chip has an economic impact that is 100 times more costly than returning a true positive, then model A 1001 on FIG. 10 should be selected and thresholds set that produce 40% true positive and less than 0.5% false positives. However, if returning a true positive is considered 100 times more important than returning a false positive for any rationale, the model B 1002 on FIG. 10 should be selected and thresholds set that produce 30% false positive and greater than 99.5% true positives, [0102] a key question for the customer is how much yield loss is acceptable? Further, would additional visual inspection, triggered by the predictive model, detect equipment issues earlier, such as a few hours, a few days, or few weeks earlier?); encoding the process-step sequence to generate the encoded process-step sequence comprise converting the plurality of process-steps … ([0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder) to a plurality of symbols ([0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model); the dimensionality reduction operation comprises an embedding operation; and making the prediction of the impact of the one or more portions of the process-step sequence on the characteristics of the testable product generated by the process-step sequence comprises evaluating the plurality of symbols against a process-step sequence language domain ([0155] the data set can be prepared for modeling by assigning to each chip the raw measurement fields… predictor variables … we can use linear as well as non-linear dimensionality reduction on the full dataset. Approaches can include: Auto encoder, [0076] log data could be parsed using Natural Language Processing (NLP) and Text Mining techniques to convert the log data into features…data from a “normal” operation condition should be added to the ML model training set such that a “normal” class is one of the possible predictions of the model).
Norman further teaches and a plurality of queue-times ([0055] the number of substrates of the substrate queue that can be started that the initiating operation within the time period)
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Seidel (US11640386) discloses modeling unknown influences (process parameters) for a process event using a processing tool to process workpieces such as a wafer.
Smith (US6965895) discloses analyzing manufacturing data tracking wafers progress.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to YVONNE T FOLLANSBEE whose telephone number is (571)272-0634. The examiner can normally be reached Monday - Friday 1pm - 9pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Robert Fennema can be reached at (571) 272-2748. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format.
For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/YVONNE TRANG FOLLANSBEE/Examiner, Art Unit 2117
/ALICIA M. CHOI/Primary Patent Examiner, Art Unit 2117
Prosecution Insights