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 .
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 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.
Claims 1 – 3, 10 – 15, 18 - 20 and 26 are rejected under 35 U.S.C. 103 as being unpatentable over US20030022399A1 (Chiou) in view of US20200020098A1 (Odry).
In regards to claim 1 (Chiou) shows:
A method of determining matching performance between tools used in semiconductor manufacturing, the method comprising: Chiou [0014] teaches a manufacturing process comprising four steps with multiple available tools for each step, creating 48 possible paths, specifically for semiconductor manufacturing processes.
obtaining a plurality of data sets related to a plurality of tools; Chiou [0016] teaches a wafer acceptance tester that carries out tests of lots of wafers processed through the 48 paths, obtaining a group of test results for each lot of wafers.
determining a matching metric and/or matching correction based on characterizing the reduced data sets in the reduced space; Chiou [0019-0021] teaches calculating an estimate P for each path and allowing engineers to select the path with the smallest estimate as the prior choice for accomplishment of the target value.
Chiou differs from the claimed invention in that it does not explicitly disclose obtaining a representation of the data sets in a reduced space having a reduced dimensionality to obtain reduced data sets, the obtaining the representation comprising; performing one or more nonlinear dimensionality reduction techniques on the data sets; using an encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation.
Odry teaches obtaining a representation of the data sets in a reduced space having a reduced dimensionality to obtain reduced data sets, the obtaining the representation comprising: Odry [0048] teaches an encoder network that transforms input data into a manifold representing a reduced dimensionality variable set where the encoder learns the parameters of distribution of a set of latent variables.
performing one or more nonlinear dimensionality reduction techniques on the data sets; Odry [0032] teaches manifold learning of data that projects original high-dimensional data into a lower-dimensional nonlinear space, a submanifold, where data separability is improved.
using an encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation; Odry [0033-0034] teaches a variational autoencoder with an encoder network and a decoder network where the encoder maps input data into continuous latent variables and the decoder network reconstructs the input from these latent variables.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to apply advanced dimensionality reduction and neural network techniques to semiconductor manufacturing data, allowing for more efficient pattern recognition and tool matching that would improve manufacturing efficiency by selecting optimal tool paths.
In regards to claim 2 (Chiou) shows the method as claimed in claim 1:
wherein each data set is related to a different respective tool; Chiou [0014] teaches different available tools for each manufacturing step, with specific designations, where each tool has its own data.
In regards to claim 3 (Chiou) shows the method as claimed in claim 1:
wherein the data sets relate to a variation of one or more tool and/or manufacturing parameters over time; Chiou [0017] teaches calculating a mean value W and a variation σ of the test results for each lot of wafers, showing variation of manufacturing parameters over time.
In regards to claim 10 (Chiou) does not show the method as claimed in claim 1, comprising using the encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation.
Odry teaches comprising using the encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation; Odry [0048-0050] teaches using an encoder network to generate a latent representation of observations and a decoder network to reconstruct the observations from the latent space representation.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to leverage encoder-decoder architectures to create compressed representations of tool manufacturing data that can capture essential variations while removing noise, leading to more accurate tool matching.
In regards to claim 11 (Chiou) does not show the method as claimed in claim 1, comprising performing one or more nonlinear dimensionality reduction techniques on the data sets, wherein the one or more nonlinear dimensionality reduction techniques comprises performing clustering and manifold learning on the datasets to group the data sets into data groups and determining matched tools as those belonging to a common data group.
Odry teaches comprising performing one or more nonlinear dimensionality reduction techniques on the data sets; Odry [0035] teaches latent variables are tuned to synthesize data similar to the input data, capturing shape variability within the data, which represents nonlinear dimensionality reduction.
Odry teaches wherein the one or more nonlinear dimensionality reduction techniques comprises performing clustering and manifold learning on the datasets to group the data sets into data groups; Odry [0063-0064] teaches that data in the latent space manifold are sparse and fall into clusters, with a clustering step performed to identify which data points are closer to each other.
Odry teaches determining matched tools as those belonging to a common data group; Odry [0064] teaches the latent variable values corresponding to an image with no tissue of a target tissue type will fit within one or more of the clusters, allowing for grouping similar items together.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to utilize clustering and manifold learning techniques to automatically group similar manufacturing tools based on their performance characteristics, enabling more efficient tool matching decisions.
In regards to claim 12 (Chiou) shows:
obtain a plurality of data sets related to a plurality of tools used in a semiconductor manufacturing process; Chiou [0016] teaches obtaining a group of test results for each lot of wafers processed through different tool paths.
determine a matching metric and/or matching correction based on characterizing the reduced data sets in the reduced space; Chiou [0019-0021] teaches calculating an estimate P for each path and selecting the path with the smallest estimate for accomplishment of the target value.
Chiou differs from the claimed invention in that it does not explicitly disclose A non-transitory computer-readable medium having instructions therein, the instructions, when executed by a computer system, configured to cause the computer system to at least; obtain a representation of the data sets in a reduced space having a reduced dimensionality to obtain reduced data sets; wherein the obtaining of the representation comprises: performance of one or more nonlinear dimensionality reduction techniques on the data sets; use of an encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation.
Odry teaches A non-transitory computer-readable medium having instructions therein, the instructions, when executed by a computer system, configured to cause the computer system to at least: Odry [0174] teaches embodying methods in tangible, non-transitory machine-readable storage media encoded with computer program code.
Odry teaches obtain a representation of the data sets in a reduced space having a reduced dimensionality to obtain reduced data sets; Odry [0048] teaches transforming input data into a manifold representing a reduced dimensionality variable set.
Odry teaches wherein the obtaining of the representation comprises: performance of one or more nonlinear dimensionality reduction techniques on the data sets; Odry [0032] teaches manifold learning of data that projects original high-dimensional data into a lower-dimensional nonlinear space.
Odry teaches use of an encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation; Odry [0033-0034] teaches using a variational autoencoder with an encoder network and a decoder network for encoding data into and decoding data from a latent space representation
The motivation to combine Chiou and Odry at the effective filing date of the invention is to apply advanced dimensionality reduction and neural network techniques to semiconductor manufacturing data, allowing for more efficient pattern recognition and tool matching that would improve manufacturing efficiency by selecting optimal tool paths.
In regards to claim 13 (Chiou) does not show the medium of claim 12: wherein the reduced space comprises a plurality of latent spaces, with individual latent spaces of the plurality of latent spaces corresponding to different regimes of a model used in defining the reduced space;
Odry teaches wherein the reduced space comprises a plurality of latent spaces, with individual latent spaces of the plurality of latent spaces corresponding to different regimes of a model used in defining the reduced space; Odry [0050] teaches transforming different types of input images (T1w and T2w) into different manifolds (M1 and M2) representing reduced dimensionality variable sets for each type.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to leverage multiple specialized latent spaces that can capture different aspects of tool performance, enabling more comprehensive analysis of different manufacturing regimes within a unified modeling framework.
In regards to claim 14 (Chiou) does not show the medium of claim 13: wherein the different regimes of the model further comprise a matching metric determination regime and/or a tool correction determination regime.
Odry teaches wherein the different regimes of the model further comprise a matching metric determination regime and/or a tool correction determination regime; Odry [0088-0089] teaches using a first network for classification of normal versus abnormal cases and a second network configured to learn from which domain the input comes from, showing different regimes for different determinations.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to implement distinct processing regimes within the model architecture that can specifically focus on either matching metric determination or tool correction, leading to more specialized and effective analysis of manufacturing tool data.
In regards to claim 15 (Chiou) does not show the medium of claim 14: wherein the one or more latent spaces comprise at least two latent spaces associated with different independent parameters comprised within the plurality of data sets;
Odry teaches wherein the one or more latent spaces comprise at least two latent spaces associated with different independent parameters comprised within the plurality of data sets; Odry [0052] teaches that at the bottleneck of the network, the latent space has two manifolds corresponding to two latent variables representing different sets of parameters of the data.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to represent different independent parameters of manufacturing tool performance in separate latent spaces, allowing for more nuanced analysis of how these parameters interact and affect overall tool matching decisions.
In regards to claim 18 (Chiou) does not show the medium of claim 12: wherein the instructions are configured to cause the computer system to use the encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation;
Odry teaches wherein the instructions are configured to cause the computer system to use the encoder-decoder network model to encode the data sets into and decode the data sets back from, the reduced space representation; Odry [0033-0034] teaches a variational autoencoder with an encoder network and a decoder network for encoding into and decoding from a latent space representation.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to implement an encoder-decoder architecture that can both compress manufacturing data into an efficient latent representation and reconstruct it accurately, ensuring that essential tool performance characteristics are preserved in the dimensionality reduction process.
In regards to claim 19 (Chiou) shows the medium of claim 12:
wherein each data set is related to a different respective tool; Chiou [0014] teaches different available tools for each step of the manufacturing process with unique designations.
In regards to claim 20 (Chiou) shows the medium of claim 12:
wherein the data sets relate to a variation of one or more tool and/or manufacturing parameters over time; Chiou [0017] teaches providing weights obtained by Exponential Weighting Moving Average based on the lots, showing variation over time.
In regards to claim 26 (Chiou) does not show the method of claim 11, further comprising: performing a first clustering and manifold learning step to obtain first groups; removing common and/or dominant data patterns per the first groups to obtain processed data sets; and performing a second clustering and manifold learning step on the processed data sets to obtain the data groups.
Odry teaches performing a first clustering and manifold learning step to obtain first groups; Odry [0072] teaches performing clustering on the latent variable values corresponding to the training image data, creating initial data groups based on manifold learning.
Odry teaches removing common and/or dominant data patterns per the first groups to obtain processed data sets; Odry [0085] teaches comparing clusters according to different parameters and determining the salience of input parameters with respect to latent variables, which effectively identifies and handles dominant patterns.
Odry teaches performing a second clustering and manifold learning step on the processed data sets to obtain the data groups; Odry [0068] teaches clustering the training image data into a set of second clusters in the latent variables based on values of an input parameter and determining probability of a subject fitting within these second clusters.
The motivation to combine Chiou and Odry at the effective filing date of the invention is to implement a multi-stage clustering approach that can identify both dominant patterns and subtle variations in tool performance data, allowing for more nuanced tool matching than traditional statistical methods.
Claims 16, 17, and 21 - 25 are rejected under 35 U.S.C. 103 as being unpatentable over US20200020098A1 (Odry) in view of US20030022399A1 (Chiou) and further in view of US20200104679A1 (Oord).
In regards to claim 16 (Odry) in view of (Chiou) does not show the medium of claim 12, wherein the representation is a latent space comprising a vector representation and the matching metric is based on a vector comparison.
Oord teaches wherein the representation is a latent space comprising a vector representation and the matching metric is based on a vector comparison; Oord [0035] teaches that latent representation of an observation can be represented as an ordered collection of numerical values such as a vector or matrix with lower dimensionality than the observation itself.
The motivation to combine Odry, Chiou, and Oord at the effective filing date of the invention is to utilize vector representations in latent space for semiconductor tool data, allowing for mathematical operations on these vectors to precisely quantify similarities between tools and optimize tool matching.
In regards to claim 17 (Odry) in view of (Chiou) shows the medium of claim 16:
decode this vector displacement into a correction for one or more of the plurality of tools, each correction making its respective tool perform more similarly to the reference; Odry [0111-0112] teaches synthesizing output from latent variables and using back-propagation to improve performance, with the decoder learning to produce better results through iterative correction.
Chiou teaches wherein the instructions are further configured to cause the computer system to: choose a reference within the latent space; Chiou [0018] teaches providing a target value T, which is the expected value of the test result of the processed wafer, serving as a reference.
Odry modified by Chiou does not teach determine the vector displacement of one or more of the plurality of tools to this reference.
Oord teaches choose a reference within the latent space; determine the vector displacement of one or more of the plurality of tools to this reference; Oord [0040] teaches processing a context latent representation to predict representations by applying linear transformations defined by parameter matrices, which mathematically calculates vector displacements between representations.
The motivation to combine Odry, Chiou, and Oord at the effective filing date of the invention is to enable vector-based correction calculations in latent space that can identify how to adjust tool parameters to achieve performance more similar to reference tools, improving overall manufacturing consistency.
In regards to claim 21 (Odry) in view of (Chiou) does not show the method of claim 1, wherein the reduced space representation is a latent space comprising a vector representation and the matching metric is based on a vector comparison.
Oord teaches wherein the reduced space representation is a latent space comprising a vector representation and the matching metric is based on a vector comparison; Oord [0035] teaches that latent representation can be represented as an ordered collection of numerical values such as a vector with lower dimensionality than the original observation.
The motivation to combine Odry, Chiou, and Oord at the effective filing date of the invention is to represent tool performance data as vectors in latent space, allowing for mathematical comparison metrics that can precisely quantify how similar different tools are performing.
In regards to claim 22 (Odry) in view of (Chiou) shows the method of claim 21, further comprising:
decoding this vector displacement into a correction for one or more of the plurality of tools, each correction making its respective tool perform more similarly to the reference; Odry [0111-0112] teaches synthesizing output from latent variables and using back-propagation to improve performance, with the decoder learning to produce better results through iterative correction.
Chiou teaches choosing a reference within the latent space; Chiou [0018] teaches providing a target value T as the expected value of the test result, serving as a reference point.
Odry modified by Chiou does not teach determining the vector displacement of one or more of the plurality of tools to this reference.
Oord teaches determining the vector displacement of one or more of the plurality of tools to this reference; Oord [0040] teaches processing a context latent representation to predict representations by applying linear transformations defined by parameter matrices, which mathematically calculates vector displacements between representations.
The motivation to combine Odry, Chiou, and Oord at the effective filing date of the invention is to leverage vector mathematics in latent space to calculate exact displacements between different tool performances and determine necessary corrections to align underperforming tools with reference standards.
In regards to claim 23 (Odry) modified by (Chiou) shows ranking the tools according to their proximity in the latent space to a tool of interest or other reference.
Chiou teaches ranking the tools according to their proximity in the latent space to a tool of interest or other reference; Chiou [0021] teaches displaying the estimates P of the paths L and selecting one path with the smallest estimate as the prior choice, effectively ranking the tools.
The motivation to combine Odry, Chiou, and Oord at the effective filing date of the invention is to enable ranking of tools based on their vector similarity in latent space, creating a more precise prioritization system for tool selection than traditional statistical methods.
In regards to claim 24 (Odry) in view of (Chiou) does not show the method of claim 21, further comprising subtracting reference data relating to a first type of tool and adding reference data relating to a second type of tool within the latent space, to match a tool of the first type with a tool of the second type.
Oord teaches subtracting reference data relating to a first type of tool and adding reference data relating to a second type of tool within the latent space, to match a tool of the first type with a tool of the second type; Oord [0040-0041] teaches applying transformations to latent representations to generate predictions for subsequent observations, which involves mathematical operations similar to adding and subtracting vector data to transform between different states.
The motivation to combine Odry, Chiou, and Oord at the effective filing date of the invention is to utilize vector arithmetic in latent space to transform representations between different tool types, allowing for more effective comparison and matching between fundamentally different types of manufacturing tools.
In regards to claim 25 (Odry) in view of (Chiou) shows the method of claim 21, further comprising:
training the model on historic scanner data sets for multiple tools and types of tools; Odry [0070-0071] teaches training a neural network using a set of training image data to determine patterns, which could be applied to historic tool data.
Response to Argument
Applicant's arguments filed on July 08, 2025 have been fully considered but are not persuasive.
With respect to claims 1-3, 10-15, 18-20 and 26, Applicant argues that Odry is non-analogous art because it is from a different field of endeavor (medical imaging versus semiconductor manufacturing) and is not reasonably pertinent to the problem of determining matching performance between semiconductor tools. This argument is not persuasive. The relevant field of endeavor is the underlying technical field of machine learning and data analysis for high-dimensional pattern recognition, not the specific application domain. Both the claimed invention and Odry apply encoder-decoder neural networks, nonlinear dimensionality reduction, latent space representations, and clustering techniques to analyze complex datasets. The field of endeavor is determined by "the nature of the problem or the disclosure's teaching" not merely the end-use application. Wyers v. Master Lock Co., 616 F.3d 1231, 1237 (Fed. Cir. 2010). Both disclosures address extracting meaningful patterns from high-dimensional data through dimensionality reduction and latent space analysis. Machine learning techniques constitute general-purpose computational tools that transcend specific application domains. A person of ordinary skill in semiconductor manufacturing seeking advanced data analysis techniques would naturally consult the broader machine learning literature, which includes Odry. The specific subject matter analyzed (tool data versus medical images) does not change the fundamental computational challenge. In re Clay, 966 F.2d 656, 659 (Fed. Cir. 1992). Furthermore, Odry is reasonably pertinent because both disclosures address how to analyze high-dimensional datasets to identify patterns, extract features, reduce dimensionality, and enable comparisons through latent space representations—precisely the problem faced by the inventors. In re Oetiker, 977 F.2d 1443, 1447 (Fed. Cir. 1992). Therefore, Odry qualifies as analogous art under both prongs of the test.
Applicant further argues that Odry teaches finding "outliers" while the claims require finding "matching" tools, characterizing these as opposite operations. This argument misunderstands both Odry's teachings and clustering analysis. Odry's clustering techniques inherently teach both identifying similar items (those that cluster together) and dissimilar items (outliers). These are complementary aspects of the same analytical framework. One cannot implement clustering without determining which items are similar and belong together. Odry [0035] explicitly teaches that "images that look similar to each other... are closer to each other within the latent space, thereby forming 'clusters'"—directly teaching similarity and matching determination. Odry [0063-0064] teaches clustering to "identify which [items] are closer to each other" and grouping "similar items together"—explicitly teaching matching operations. Items within a cluster are, by definition, matched to each other based on proximity in latent space. The claim language "determining a matching metric and/or matching correction based on characterizing the reduced data sets in the reduced space" encompasses both determining which tools match (similar in latent space) and which don't match (dissimilar/outliers). Odry's clustering approach directly reads on this limitation.
Applicant also argues that the motivation to combine is not supported because Odry does not explicitly mention "tool matching," "pattern recognition," or "manufacturing efficiency." This argument misunderstands KSR. The motivation does not require that the secondary reference explicitly mention the primary reference's specific application. KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 418 (2007). Chiou teaches basic statistical tool path selection using means, variations, and weighted averages [0017-0021]. A person of ordinary skill, aware of machine learning advances, would have been motivated to improve Chiou's simple approach by applying Odry's techniques to: (1) capture nonlinear relationships rather than just linear statistics, (2) extract features automatically rather than using only predefined statistics, (3) handle higher-dimensional data through dimensionality reduction, (4) enable nuanced comparisons through latent space representations, and (5) identify patterns through clustering. These are predictable benefits of applying known machine learning techniques to tool selection. While Odry may not use the phrase "pattern recognition," the disclosure teaches pattern recognition techniques: encoder networks learn patterns to create latent representations [0034], clustering identifies similarity patterns [0063-0064], and manifold learning discovers patterns in high-dimensional data [0032].
With respect to claims 16, 17, and 21-25, Applicant argues that Oord fails to overcome deficiencies. There are no deficiencies in the Chiou/Odry combination as explained above. Additionally, Applicant provides only a conclusory statement without identifying any specific deficiency in the Office action's application of Oord. Oord [0035] teaches latent representations as "an ordered collection of numerical values such as a vector" with lower dimensionality—teaching vector representation. Oord [0040] teaches "applying linear transformations defined by parameter matrices"—teaching vector operations including displacements and comparisons. Combined with Chiou [0018] teaching reference target values and Odry [0111-0112] teaching decoded outputs, Oord teaches the claimed vector displacement and correction limitations. A bare assertion without supporting argument is insufficient to traverse a rejection. MPEP 714.02.
The remaining arguments with respect to dependent claims have been considered but are not persuasive for the reasons stated above. The dependent claims incorporate the limitations of their respective independent claims and are properly rejected over the same combination of references. The rejections are maintained.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANWER AHMED ALAWDI whose telephone number is (703)756-1018. The examiner can normally be reached Monday - Friday 8:00 am - 5:30 pm.
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, Jack Chiang can be reached on (571)-272-7483. 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.
/ANWER AHMED ALAWDI/Examiner, Art Unit 2851
/JACK CHIANG/Supervisory Patent Examiner, Art Unit 2851