DETAILED ACTION
Notice of 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 37 CFR 1.114
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 2/3/2026 has been entered.
Information Disclosure Statement
The information disclosure statement submitted on 4/28/2026 has been considered.
Response to Amendment
Applicant’s Amendment and remarks dated 2/3/2026 have been considered. Claims 3-4 and 15-16 are canceled. Claims 1, 5-13, and 17-25 are pending.
Claim Objections. The previous objections to claims 1, 13, 18, and 24-25 are withdrawn in view of Applicant’s claim amendments.
Response to Arguments
On page 7 of Applicant’s 2/3/2026 Amendment and remarks, Applicant asserts that no new matter has been added via the claim amendments, and that at least paras. 0023, 0036, and 0050 provide sufficient written description support.
The examiner agrees that the amendments to independent claims 1, 13, and 25 are supported at least by original claims 3-4, para. 0044 of the instant specification, and the portions of the specification identified by Applicant.
On pages 7-9 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 101, Applicant argues that “as a whole,” claim 1 cannot practically be performed in the human mind.
The examiner respectfully disagrees. The examiner identified several mental processes and Applicant has not rebutted any of the identified mental processes. The examiner’s limitation-by-limitation analysis is consistent with the limitation-by-limitation analyses set forth in the USPTO’s Subject Matter Eligibility examples.
Moreover, “as a whole”, the claims relate to the mental processes of characterizing materials, and the claims merely implement such mental process using a generic machine learning model.
Applicant’s citations to the specification on pages 8-9 just reinforce that the claimed invention is merely the automation of the mental processes associated with steps for materials characterization, where the only improvement is to use a generic machine learning model to automate the mental processes.
On pages 9-10 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 101, Applicant argues that the “outputting” step “continues to underscore the practical application which addresses the complex and dense challenges with a specific technological approach, and transforming available data into a useful new aspect of material characterization.
The examiner respectfully disagrees. As explained in the office action, such “outputting” is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. Moreover, the “outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation amounts to extra solution activity because it is a mere nominal or tangential addition to the claim, amounting to mere data output (see MPEP 2106.05(g)).
As explained by MPEP 2106.05(c), a “transformation” requires a physical transformation of a physical article or substance and does not apply to merely transforming mental processes, and there are no transformations being performed by the “outputting” step as alleged by Applicant.
On pages 10-11 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 101, Applicant argues that the amendments to claim 1 specifying that limitations are performed in “real-time” overcomes the rejections.
The examiner respectfully disagrees. The “real-time” limitations merely refer to the use of a generic machine learning model to perform mental processes more quickly and efficiently than a human. But the only improvement, if any, is to the mental processes themselves, and not to the machine learning technology.
On pages 11-12 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 103, Applicant argues:
PNG
media_image1.png
178
648
media_image1.png
Greyscale
PNG
media_image2.png
132
644
media_image2.png
Greyscale
The examiner respectfully disagrees. Simply put, LUDWIG teaches the concept of performing materials characterization using thin-film material libraries, and explicitly contemplates future use of such libraries using machine learning. (see LUDWIG, p. 6). TEHRANI supplies the machine learning model to search for materials having desired properties. The combination of LUDWIG and TEHRANI is straightforward as explained in the office action, particularly given the high-level nature of the claims, which merely apply a generic machine learning model to known techniques for arranging materials on a substrate for analysis.
The examiner further disagrees that the “rejection unreasonably stretches Ludwig’s mere mention of numbered crosses for ‘navigation.” The recited “position encoding” is merely noting the location of a material on the substrate, such as in a grid. (see Fig. 1 of instant specification). LUDWIG describes and depicts such a grid (see Fig. 2 of LUDWIG), and explains that the numbered crosses relate to measurement areas for different materials. Therefore, the examiner respectfully disagrees that there is even any “stretching” involved. The numbered crosses of LUDWIG squarely read on the broadest reasonable interpretation of “positional encoding” of the claims.
On page 12 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 103, Applicant argues:
PNG
media_image3.png
576
672
media_image3.png
Greyscale
The examiner respectfully disagrees. The numbered crosses are available at certain temperatures as acknowledged by Applicant. The claims do not have any limitations excluding the temperature ranges utilized by LUDWIG, or any limitations related to temperature at all.
Moreover, the “according to known physical, chemical, and or treatment attributes” is taught by LUDWIG, as attributes are known for each material fabricated on the structure, and therefore the numbered crosses are associated with known attributes. Moreover, LUDWIG explains at page 2 that experiments are held using known composition spaces and properties, such that the numbered crosses are associated with known material attributes and the materials are not randomly distributed across the substrate.
On pages 12-13 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 103, Applicant mischaracterizes the office action’s analysis with respect to the “correlate the definitional data with the operational data” limitation.
The examiner respectfully submits that LUDWIG at page 3 expressly taches “multifunctional existence diagrams, comprising correlations between composition, processing, structure, and properties”, where composition and structure correspond to recited “operational data” and measured properties correspond to measured “definitional data” as explained in the office action. Moreover, because this is expressly taught by LUDWIG, there is no reason to explain why such “multifunctional existence diagrams” would work with the secondary reference, TEHRANI, as alleged by Applicant, because TEHRANI is not relied upon for this particular limitation.
On pages 13-14 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 103, Applicant makes several arguments with respect to TEHRANI reference.
In response to Applicant's arguments against the references individually, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). Here, Applicant ignores the actual combination of LUDWIG and TEHRANI proposed by the examiner. TEHRANI teaches the high-level concept of using machine learning to predict material properties, which is precisely what is claimed.
In response to Applicant's argument that the examiner's conclusion of obviousness is based upon improper hindsight reasoning, it must be recognized that any judgment on obviousness is in a sense necessarily a reconstruction based upon hindsight reasoning. But so long as it takes into account only knowledge which was within the level of ordinary skill at the time the claimed invention was made, and does not include knowledge gleaned only from the applicant's disclosure, such a reconstruction is proper. See In re McLaughlin, 443 F.2d 1392, 170 USPQ 209 (CCPA 1971).
On page 15 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 103, Applicant argues that the newly-added “according to the positional encoding via the definitional and the operational data” overcomes the prior art.
The examiner respectfully disagrees. This limitation merely means that the positional encoding for a next location of inquiry is made in view of the definitional and the operational data, and as explained above, LUDWIG fabricates the material with the operational and definitional properties in mind, such that the numbered crosses are associated with such operational and definitional properties, and in combination with TEHRANI, the search for materials with desired mechanical properties will take such operational and definitional properties into account when selecting the next material for analysis.
On page 15 of Applicant’s 2/3/2026 Amendment and remarks, with respect to the rejection of claim 1 under 35 U.S.C. 103, Applicant argues that “Dutt’s confidence level is applied in an irrelevant manner for a different purpose, and would not lead the other cited art to achieve the claims.”
The examiner respectfully disagrees. While DUTT does not relate specifically to the materials characterization field, DUTT is relied upon for the basic concept of having a confidence interval, or value, with respect to machine learning predictions. One of ordinary skill would have been motivated to apply such confidence rate to the machine learning model of TEHRANI to evaluate the correctness of the output of such model.
On page 16 of Applicant’s 2/3/2026 Amendment and remarks, Applicant argues that the remaining claims should be allowed for the same reasons argued with respect to claim 1.
The examiner respectfully disagrees for the same reasons explained with respect to claim 1.
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, 5-13, 17-25 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Step 1 of the Alice/Mayo framework, Claims 1 and 5-12 and 25 are directed to a method (a process) and Claims 13 and 17-24 are directed to a system (a machine), which each fall within one of the four statutory categories of inventions.
Regarding Claim 1
Step 2A, prong 1 (Is the claim directed to a law of nature, a natural phenomenon or an abstract idea).
Claim 1 recites the following mental processes, that in each case under the broadest reasonable interpretation, covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components (e.g., “machine learning model”).
positionally encoding a set of material samples (under the broadest reasonable interpretation, a human can mentally determine how to positionally encode a set of material samples, such as in a grid or layout for future fabrication)
detecting definitional data from at least some of the material samples as definitional samples (under the broadest reasonable interpretation, a human can mentally analyze samples to identify some features and define samples with those features as definitional samples)
correlating the definitional data with the operational data (under the broadest reasonable interpretation, a human can mentally correlate definitional data with operational data, such as mentally putting such data in a table where known properties (operational data) are logically placed next to related measured data (definitional data))
characterizing at least some of the definitional samples as characterization training data based on correlation of the definitional and the operational data, and (under the broadest reasonable interpretation, characterizing, or selecting certain definitional samples, and designating such samples as characterization training data, is a decision that a human can make mentally)
where characterization at least some of the definitional samples as characterization training data includes determining ... a location on the at least one substrate of a next-material sample for characterization as characterization training data according to the positional encoding via the definitional and the operational data (under the broadest reasonable interpretation, a human can mentally consider the next sample on the substrate that will be characterized and included as characterization training data for the machine learning model, where such next sample is identified via the positional encoding, and identified via the definitional and operational data)
including determining a predicted output (under the broadest reasonable interpretation, a human can mentally determine a predicted output, such as predicting that a copper nanoparticle can conduct electricity)
determining an experimental output (under the broadest reasonable interpretation, a human can mentally analyze experimental data or measured data in order to determine results of the experiment or measurement)
comparing the predicted and experimental outputs to determine a predictive error value, and (under the broadest reasonable interpretation, a human can mentally compare predicted and experimental outputs to determine if there is an error or not, and can mentally quantify an error value if desired)
determining a confidence value for predictive output of each of the material samples based on the predictive error value. (under the broadest reasonable interpretation, a human can mentally determine a confidence value based on the predictive error value, such as determining a high confidence value if there is no error)
wherein determining the location on the at least one substrate of the next-material sample for characterization includes determination of the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount. (under the broadest reasonable interpretation, a human can mentally evaluate the locations on the substrate of the next-material sample and select a sample that will likely increase the confidence value by the greatest amount)
Step 2A, prong 2 (Does the claim recite additional elements that integrate the judicial exception into a practical application?).
The judicial exception is not integrated into a practical application. In particular, the claim recites the additional element (e.g., “machine learning model”) which is recited at a high-level of generality such that they amount to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Regarding the “on at least one substrate according to known physical, chemical, and/or treatment attributes as operational data” limitation, this limitation merely describes the data that will be processed (material samples on a substrate), and therefore such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use (the encoding of a particular type of data). As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Regarding the “inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. The machine learning model is recited at a high level of generality and therefore amounts to mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)). Moreover, the “outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation amounts to extra solution activity because it is a mere nominal or tangential addition to the claim, amounting to mere data output (see MPEP 2106.05(g)).
Regarding the “in real-time by the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Step 2B (Does the claim recite additional elements that amount to significantly more than the judicial exception?)
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the additional element (e.g., “machine learning model”) is recited at a high-level of generality such that they amount to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Regarding the “on at least one substrate according to known physical, chemical, and/or treatment attributes as operational data” limitation, such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use as explained above, which is not significantly more than the judicial exception. MPEP 2106.05(h).
Regarding the “inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. The recited machine learning model is recited at a high level of generality and therefore amounts to instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)). Moreover, the “outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation amounts to extra solution activity because it is a mere nominal or tangential addition to the claim, amounting to mere data output (see MPEP 2106.05(g)). The courts have similarly found limitations directed to displaying a result, recited at a high level of generality, to be well-understood, routine, and conventional. See (MPEP 2106.05(d)(II), "presenting offers and gathering statistics.", “determining an estimated outcome and setting a price”)
Regarding the “in real-time by the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Accordingly, at Step 2B after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Regarding Claim 5
Step 2A, Prong 1
wherein characterizing at least some of the definitional samples as characterization training data is determined to be complete upon reaching a predetermined threshold confidence value (under the broadest reasonable interpretation, a human can mentally determine that once a predetermined threshold confidence value is reached, that the characterization of training data is complete)
Regarding Step 2A, Prong 2, the claim does not include any additional elements that integrate the judicial exception into a practical application and regarding Step 2B, there are no additional elements recited that amount to significantly more than the judicial exception.
Regarding Claim 6
Step 2A, Prong 1
outputting a next-location for detection and a predicted output of the corresponding sample (under the broadest reasonable interpretation, a human can mentally perform these outputting steps after mentally reviewing the output from the machine learning model, because entering the training data into the machine learning model to output these “outputs” amounts to adding the words “apply it” to the judicial exception)
comparing the detected definitional data with the predicted output (under the broadest reasonable interpretation, a human can mentally compare detected data with predicted output, for example, to see if there is a match)
Step 2A, Prong 2
Regarding the “entering the training data into the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)).
Regarding the “detecting definitional data of the next-material sample” limitation, such additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data for use in the claimed process (see MPEP 2106.05(g))).
Step 2B
Regarding the “entering the training data into the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding the “detecting definitional data of the next-material sample” limitation, as discussed above, the additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
Regarding Claim 7
Step 2A, Prong 2
Regarding the “wherein physical, chemical, and/or treatment attributes as operational data includes one or more of: precursor gradient among material samples across the at least one substrate, chemical constituent gradient among material samples across the at least one substrate, and treatment gradient by exposure to irradiation with different wavelengths among material samples across the at least one substrate” limitation, this limitation merely describes the operational data that is known, and therefore such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B
Regarding the “wherein physical, chemical, and/or treatment attributes as operational data includes one or more of: precursor gradient among material samples across the at least one substrate, chemical constituent gradient among material samples across the at least one substrate, and treatment gradient by exposure to irradiation with different wavelengths among material samples across the at least one substrate” limitation, such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use as explained above, which is not significantly more than the judicial exception. MPEP 2106.05(h).
Regarding Claim 8
Step 2A, Prong 1
Regarding the “wherein detecting definitional data includes determining one or more of catalytic activity, electrochemical activity, chemical product distribution resultant from reaction, elemental distribution and/or geometry, mechanical-physical properties, thermal properties, optical properties, catalytic and/or corrosion evolution, and/or fluorescence intensity” limitation, this limitation merely describes the definitional data that is measured, and therefore such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B
Regarding the “wherein detecting definitional data includes determining one or more of catalytic activity, electrochemical activity, chemical product distribution resultant from reaction, elemental distribution and/or geometry, mechanical-physical properties, thermal properties, optical properties, catalytic and/or corrosion evolution, and/or fluorescence intensity” limitation, such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use as explained above, which is not significantly more than the judicial exception. MPEP 2106.05(h).
Regarding Claim 9
Step 2A, Prong 1
determining one or more physical, chemical, and/or treatment attributes of a next material collection for further characterization (under the broadest reasonable interpretation, a human can mentally determine attributes of a next material for further characterization, such as determining that certain metals have magnetic properties)
Regarding Step 2A, Prong 2, the claim does not include any additional elements that integrate the judicial exception into a practical application and regarding Step 2B, there are no additional elements recited that amount to significantly more than the judicial exception.
Regarding Claim 10
Step 2A, Prong 2
Regarding the “wherein detecting definitional data further comprises obtaining definitional data concerning material samples of another known material collection as at least some of the definitional samples” limitation, such additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data for use in the claimed process (see MPEP 2106.05(g))).
Step 2B
Regarding the “wherein detecting definitional data further comprises obtaining definitional data concerning material samples of another known material collection as at least some of the definitional samples” limitation, as discussed above, the additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
Regarding Claim 11
Step 2A, Prong 2
Regarding the “wherein the material samples are each defined on the nano- or micro-scale” limitation, this limitation merely describes attributes of the data being processed (nano- or micro-scale), and therefore such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use. As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B
Regarding the “wherein the material samples are each defined on the nano- or micro-scale” limitation, such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use as explained above, which is not significantly more than the judicial exception. MPEP 2106.05(h).
Regarding Claim 12
Step 2A, Prong 2
Regarding the “wherein detecting definitional data from at least some of the material samples includes moving between material samples at the nano- or micro-scale” limitation, such additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data for use in the claimed process (see MPEP 2106.05(g))).
Step 2B
Regarding the “wherein detecting definitional data from at least some of the material samples includes moving between material samples at the nano- or micro-scale” limitation, as discussed above, the additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
Regarding Claim 13
Step 2A, prong 1 (Is the claim directed to a law of nature, a natural phenomenon or an abstract idea).
Claim 13 recites the following mental processes, that in each case under the broadest reasonable interpretation, covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components (e.g., “machine learning model”, “sensor”, “processor”, “memory”, and “instructions).
wherein the material samples are positionally encoded (under the broadest reasonable interpretation, a human can mentally determine how to positionally encode a set of material samples, such as in a grid or layout for future fabrication)
detect definitional data from at least some of the material samples as definitional samples (under the broadest reasonable interpretation, a human can mentally analyze samples to identify some features and define samples with those features as definitional samples)
to correlate the definitional data with the operational data, and (under the broadest reasonable interpretation, a human can mentally correlate definitional data with operational data, such as mentally putting such data in a table where known properties (operational data) are logically placed next to related measured data (definitional data))
wherein the configuration to characterize at least some of the definitional samples as characterization training data includes configuration to determine, … a location on the at least one substrate of a next-material sample for characterization as characterization training data according to the positional encoding via the definitional and the operational data, including configuration (under the broadest reasonable interpretation, a human can mentally consider the next sample on the substrate that will be characterized and included as characterization training data for the machine learning model, where such next sample is identified via the positional encoding, and identified via the definitional and operational data)
to determine a predicted output (under the broadest reasonable interpretation, a human can mentally determine a predicted output, such as predicting that a copper nanoparticle can conduct electricity)
to determine an experimental output (under the broadest reasonable interpretation, a human can mentally analyze experimental data or measured data in order to determine results of the experiment or measurement)
to compare the predicted and experimental outputs to determine a predictive error value, and (under the broadest reasonable interpretation, a human can mentally compare predicted and experimental outputs to determine if there is an error or not)
to determine a confidence value for predictive output of each of the material samples based on the predictive error value (under the broadest reasonable interpretation, a human can mentally determine a confidence value based on the predictive error value, such as determining a high confidence value if there is no error)
wherein the configuration to determine the location on the at least one substrate of the next-material sample for characterization includes configuration to determine the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount. (under the broadest reasonable interpretation, a human can mentally evaluate the locations on the substrate of the next-material sample and select a sample that will likely increase the confidence value by the greatest amount)
Step 2A, Prong 2
Regarding the “a data collection system comprising at least one sensor configured to detect definitional data from at least some material samples of a set of material samples as definitional samples” limitation, such additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data (such as by using a sensor) for use in the claimed process (see MPEP 2106.05(g))).
Regarding the “on at least one substrate according to known physical, chemical, and/or treatment attributes as operational data” limitation, this limitation merely describes the data that will be processed (material samples on a substrate), and therefore such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use (the encoding of a particular type of data). As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Regarding the “a characterization control system comprising at least one processor configured to execute instructions stored on memory to conduct characterization of the set of material samples on the at least one substrate, the characterization control system configured to operate the data collection system to” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)).
Regarding the “wherein the characterization control system includes a machine learning model configured to receive the characterization training data as input, and to output characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. . The machine learning model is recited at a high level of generality and therefore amounts to instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)). Moreover, the “output characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation amounts to extra solution activity because it is a mere nominal or tangential addition to the claim, amounting to mere data output (see MPEP 2106.05(g)).
Regarding the “in real-time by the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Step 2B
Regarding the “a data collection system comprising at least one sensor configured to detect definitional data from at least some material samples of a set of material samples as definitional samples” limitation, as discussed above, the additional element of a data gathering step is recited at a high level of generality and amounts to extra-solution activity of receiving data, i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
Regarding the “on at least one substrate according to known physical, chemical, and/or treatment attributes as operational data” limitation, such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use as explained above, which is not significantly more than the judicial exception. MPEP 2106.05(h).
Regarding the “a characterization control system comprising at least one processor configured to execute instructions stored on memory to conduct characterization of the set of material samples on the at least one substrate, the characterization control system configured to operate the data collection system to” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Regarding the “wherein the characterization control system includes a machine learning model configured to receive the characterization training data as input, and to output characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. The machine learning model is recited at a high level of generality and therefore amounts to instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)). Moreover, the “output characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation amounts to extra solution activity because it is a mere nominal or tangential addition to the claim, amounting to mere data output (see MPEP 2106.05(g)). The courts have similarly found limitations directed to displaying a result, recited at a high level of generality, to be well-understood, routine, and conventional. See (MPEP 2106.05(d)(II), "presenting offers and gathering statistics.", “determining an estimated outcome and setting a price”)
Regarding the “in real-time by the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Accordingly, at Step 2B after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Regarding Claims 17-24
Claims 17-24 depend from claim 13, and each claim a system that correlates to dependent claims 5-12, and therefore claims 17-24 are rejected for the same reasons explained above with respect to claim 13 and claims 5-12, respectively.
Regarding Claim 25
Step 2A, prong 1 (Is the claim directed to a law of nature, a natural phenomenon or an abstract idea).
Claim 25 recites the following mental processes, that in each case under the broadest reasonable interpretation, covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components (e.g., “machine learning model”).
detecting definitional data from material samples as definitional samples (under the broadest reasonable interpretation, a human can mentally analyze samples to identify some features and define samples with those features as definitional samples)
the material samples positionally encoded (under the broadest reasonable interpretation, a human can mentally determine how to positionally encode a set of material samples, such as in a grid or layout for future fabrication)
correlating the definitional data with the operational data, including correlating based on definitional data obtained from another material chip having materials samples positionally encoded on a substrate according to known physical, chemical, and/or treatment attributes as operational data; (under the broadest reasonable interpretation, a human can mentally correlate definitional data with operational data, such as mentally putting such data in a table where known properties (operational data) are logically placed next to related measured data (definitional data))
wherein characterizing at least some of the definitional samples as characterization training data includes determining, … a location on the at least one substrate of a next-material sample for characterization as characterization training data according to the positional encoding via the definitional and the operational data, including (under the broadest reasonable interpretation, a human can mentally consider the next sample on the substrate that will be characterized and included as characterization training data for the machine learning model, where such next sample is identified via the positional encoding, and identified via the definitional and operational data)
determining a predicted output (under the broadest reasonable interpretation, a human can mentally determine a predicted output, such as predicting that a copper nanoparticle can conduct electricity)
determining an experimental output (under the broadest reasonable interpretation, a human can mentally analyze experimental data or measured data in order to determine results of the experiment or measurement)
comparing the predicted and experimental outputs to determine a predictive error value, and (under the broadest reasonable interpretation, a human can mentally compare predicted and experimental outputs to determine if there is an error or not)
determining a confidence value for predictive output of each of the material samples based on the predictive error value (under the broadest reasonable interpretation, a human can mentally determine a confidence value based on the predictive error value, such as determining a high confidence value if there is no error)
wherein determining the location on the at least one substrate of the next-material sample for characterization includes determination of the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount. (under the broadest reasonable interpretation, a human can mentally evaluate the locations on the substrate of the next-material sample and select a sample that will likely increase the confidence value by the greatest amount)
Step 2A, prong 2 (Does the claim recite additional elements that integrate the judicial exception into a practical application?).
The judicial exception is not integrated into a practical application. In particular, the claim recites the additional element (e.g., “machine learning model”) which is recited at a high-level of generality such that they amount to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Regarding the “on a substrate according to known physical, chemical, and/or treatment attributes as operational data” limitation, this limitation merely describes the data that will be processed (material samples on a substrate), and therefore such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use (the encoding of a particular type of data). As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible "simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use." Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981). Thus, limitations that amount to merely indicating a field of use or technological environment in which to apply a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Regarding the “inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. The machine learning model is recited at a high level of generality and therefore amounts to instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)). Moreover, the “outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation amounts to extra solution activity because it is a mere nominal or tangential addition to the claim, amounting to mere data output (see MPEP 2106.05(g)).
Regarding the “in real-time by the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Step 2B (Does the claim recite additional elements that amount to significantly more than the judicial exception?)
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception. As discussed above, the additional element (e.g., “machine learning model”) is recited at a high-level of generality such that they amount to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Regarding the “on a substrate according to known physical, chemical, and/or treatment attributes as operational data” limitation, such limitation amounts to no more than generally linking the use of a judicial exception to a particular technological environment or field of use as explained above, which is not significantly more than the judicial exception. MPEP 2106.05(h).
Regarding the “inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception. The machine learning model is recited at a high level of generality and therefore amounts to instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)). Moreover, the “outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data” limitation amounts to extra solution activity because it is a mere nominal or tangential addition to the claim, amounting to mere data output (see MPEP 2106.05(g)). The courts have similarly found limitations directed to displaying a result, recited at a high level of generality, to be well-understood, routine, and conventional. See (MPEP 2106.05(d)(II), "presenting offers and gathering statistics.", “determining an estimated outcome and setting a price”)
Regarding the “in real-time by the machine learning model” limitation, such limitation is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, because the limitation merely provides instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not add significantly more than the judicial exception. (See MPEP 2106.05(f)).
Accordingly, at Step 2B after considering all claim elements individually and as an ordered combination, it is determined that the claims do not integrate the judicial exception into a practical application.
Claim Rejections - 35 USC § 103
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 5-13, and 17-25 rejected under 35 U.S.C. 103 as being unpatentable over Ludwig, Alfred. "Discovery of new materials using combinatorial synthesis and high-throughput characterization of thin-film materials libraries combined with computational methods." npj Computational Materials 5.1 (2019) (submitted in Applicant’s 2/21/2025 IDS), hereinafter referenced as LUDWIG, in view of Mansouri Tehrani, Aria, et al. "Machine learning directed search for ultra incompressible, superhard materials." Journal of the American Chemical Society 140.31 (2018): pp. 9844-9853, hereinafter referenced as TEHRANI, and further in view of US 20210406744 A1, hereinafter referenced as DUTT.
Regarding Claim 1
LUDWIG teaches:
A method of real-time characterizing a material collection, the method comprising: (LUDWIG, p. 2, left column: “This perspective discusses the possibilities of combinatorial synthesis and high-throughput characterization in combination with computational methods in the endeavor to efficiently identify new materials in multi-dimensional search spaces”)
positionally encoding a set of material samples on at least one substrate (LUDWIG, p. 3, right column: “Whereas in some cases completely continuous MLs are used, it is in most cases better to use a ML which is structured into measurement areas (MAs), however, without losing to much of the composition spread. In the following examples, the MLs comprise 342 MAs over a 100mm diameter substrate (typically thermally oxidized Si wafer) see Fig. 2. The crosses, defining Mas, are fabricated by a photolithographic lift-off process to achieve a pre-patterned substrate. The numbered crosses, where no thin film is applied, are used for thickness measurements (e.g., with profilometry) and are useful for navigation on the ML when performing high-throughput characterization.”;
Examiner’s Note (EN): the measurement areas are arranged on the substrate such that the “numbered crosses” (corresponding to “positionally encoding”) can be used to navigate over the substrate; the examiner notes that the broadest reasonable interpretation of “positionally encoding” includes identifying the location of a material sample, which can be in a grid as shown by Fig. 1 of the instant specification, and LUDWIG discloses (see Fig. 2) a pre-patterned substrate (such as a grid), where each measurement area is numbered (corresponding to recited “positional encoding”))
according to known physical, chemical, and/or treatment attributes as operational data; (LUDWIG, p. 2, left column: “Thus, discoveries can be expected in (I) (compositionally) unexplored search spaces, by fabricating and characterizing parts of the unexplored composition space and (II) by performing experiments in (known) composition spaces, by testing for special, but not yet investigated functionalities.”
LUDWIG, p. 2, right column: “A ML is a well-defined set of materials—suitable for high-throughput characterization—produced in one experiment under identical conditions. Combinatorial and high-throughput methods for materials discovery and accelerated development have been developed in the last decades.... They involve, first, identification of the chemical materials compositions and their crystallographic structure. Second, structural and functional properties can be identified and be further optimized by combinatorial processing.”
Examiner’s Note (EN): LUDWIG teaches that chemical and structure (corresponding to “physical”) properties of the materials are known when fabricated on the substrate, corresponding to recited “known physical, chemical, and/or treatment attributes as operational data”)
detecting definitional data from at least some of the material samples as definitional samples; (LUDWIG, p. 4, left column and Fig. 2: “Further high-throughput characterization methods correspond to the specific material development goals, i.e., functional properties. Adequate screening parameters (e.g., electrical resistivity, Seebeck coefficient, Young’s modulus, hardness, etc.) need to be defined or descriptors of the property of interest, if it cannot be directly measured. In many cases measuring one screening parameter is not enough, and a different ML design might be necessary to enable measuring all necessary parameters. High-throughput electrical resistivity measurements serve well for delivering descriptor values for properties related to it”; Examiner’s Note (EN): As shown in Fig. 2, the characterization of materials includes characterization with respect to electrical properties, magnetic properties, photoelectric-chemical properties, optical properties, microstructure, mechanical properties, composition, and crystal structures and phase constitution; the broadest reasonable interpretation of “definitional data” includes data measured from the material samples to characterized as explained in para. 0041 of the instant specification, and LUDWIG discloses measuring the samples on the substrate as depicted in Fig. 2)
correlating the definitional data with the operational data; (LUDWIG, p. 3, right column: “The acquired large and consistent datasets of intrinsic and extrinsic properties enable materials discoveries and efficient optimization of identified materials. Furthermore, they are the basis for multifunctional existence diagrams, comprising correlations between composition, processing, structure and properties.; Examiner’s Note (EN): chemical composition and physical structure are known, operational data as disclosed by p. 2, right column (see explanation above), and are correlated with measured properties (corresponding to measured “definitional data” as recited in the claim) (see explanation above), which are correlated at least by the “multifunctional existence diagrams” that provide for correlation of operational data and definitional data)
determining an experimental output (LUDWIG, p. 4, left column: “Further high-throughput characterization methods correspond to the specific material development goals, i.e., functional properties. Adequate screening parameters (e.g., electrical resistivity, Seebeck coefficient, Young’s modulus, hardness, etc.) need to be defined or descriptors of the property of interest, if it cannot be directly measured. In many cases measuring one screening parameter is not enough, and a different ML design might be necessary to enable measuring all necessary parameters. High-throughput electrical resistivity measurements serve well for delivering descriptor values for properties related to it”)
However, LUDWIG fails to explicitly teach:
characterizing at least some of the definitional samples as characterization training data based on correlation of the definitional and the operational data, and
inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data
wherein characterizing at least some of the definitional samples as characterization training data includes determining, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as characterization training data according to the positional encoding via the definitional and the operational data
determining a predicted output,
comparing the predicted and experimental outputs to determine a predictive error value, and
determining a confidence value for predictive output of each of the material samples based on the predictive error value;
wherein determining the location on the at least one substrate of the next-material sample for characterization includes determination of the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount.
However, in a related field of endeavor (synthesis and characterization of materials, see p.9849, “Synthesis and Characterization of Two High-Hardness Materials”), TEHRANI teaches:
characterizing at least some of the definitional samples as characterization training data based on correlation of the definitional and the operational data, and (TEHRANI, p. 9846, “Results and Discussion” section: “Machine learning of elastic moduli begins with the construction of a robust training set. In this case, the Materials Project DFT calculated B and G are ideal to train the SVR machine-learning model”;
TEHRANI, p. 9846, “Mechanical Property Measurements” section: “samples analyzed using laboratory X-ray powder diffraction were examined under compression to measure the equation of state (EoS) to compare the experimental and machine learning-predicted bulk modulus”
Examiner’s Note (EN): TEHRANI explicitly teaches creating a training set, corresponding to recited “characterizing at least some of the definitional samples as characterization training data”; in combination with LUDWIG, a training set is constructed as in TEHRANI, using the measured samples (as in both LUDWIG and TEHRANI), including the correlation of the measured data (corresponding to “definitional data”) with known data (corresponding to “operational data”) as disclosed by LUDWIG)
inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data, (TEHRANI, p. 9845, “Introduction” section: “Here, we develop a method based on machine learning to vastly expand the number of materials with their elastic moduli predicted. This approach employs a combination of compositional and structural descriptors to build a ML model using the Materials Project data as a training set.”
TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”
Examiner’s Note (EN): TEHRANI discloses that the trained machine learning model is used to “direct the search for materials with desired mechanical properties,” corresponding to the recited “other than the definitional samples,”; in combination with LUDWIG, the machine learning model of TEHRANI is now trained using the data collected and correlated by LUDWIG, to investigate materials on the substate of LUDWIG).
wherein characterizing at least some of the definitional samples as training data includes determining, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as training data according to the positional encoding via the definitional and the operational data (TEHRANI, p. 9846, “Results and Discussion” section: “Machine learning of elastic moduli begins with the construction of a robust training set. In this case, the Materials Project DFT calculated B and G are ideal to train the SVR machine-learning model”;
TEHRANI, p. 9848: “Nevertheless, most of the ML approaches developed thus far, irrespective of ML method or choice of descriptors, are able to predict elastic moduli with impressive accuracy in a fraction of the time that it currently takes ab initio calculations.”
TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.” Examiner’s Note (EN): TEHRANI explicitly teaches creating a training set, corresponding to recited “characterizing at least some of the definitional samples as characterization training data”; in combination with LUDWIG (which teaches “numbered crosses” that define “measurement areas (MAs)” on the patterned substrate, se p. 3, right column), the machine learning model of TEHRANI directs “the search for materials with desired mechanical properties” on the patterned substrate of LUDWIG, corresponding to recited “determining, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as training data” as recited in this claim because the search in TEHRANI relates to suggesting similar materials with desired properties for further analysis and can suggest similar materials based on the location of the “numbered crosses” of LUDWIG and taking into account the definitional and operational data of LUDWIG)
determining a predicted output, (TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”)
comparing the predicted and experimental outputs to determine a predictive error value, and (TEHRANI, p. 9846, “Machine-Learning Bulk...” section: “As shown in Figure 1a, remarkable agreement is obtained between the DFT-calculated bulk modulus (BDFT) and the ML-predicted values (BSVR) with the cross-validated root-mean-square error (RMSECV)”l Examiner’s Note (EN): in combination with LUDWIG, the machine learning model of TEHRANI now measures samples (corresponding to “experimental outputs”, which also include outputs from measurements of LUDWIG), and the cross-validating of predicted vs. measured values corresponds to the “comparing the predicted and experimental outputs to determine a predictive error value” limitation because the cross-validating output is a root mean square error)
… includes determination of the location on the at least one substrate of the next-material sample (TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI as explained above. As disclosed by TEHRANI, one of ordinary skill would have been motivated to do so in order to “direct the search for materials with desired mechanical properties.” (TEHRANI, p. 9848, “Screening Crystal Structure ...” section). As further disclosed by TEHRANI, “One area where ML may also be useful is in the search for materials with exceptional mechanical properties, such as high incompressibility or extreme hardness” because machine learning provides advantages over the previous trial-and-error approach. (TEHRANI, p. 9845, “Introduction” section). One of ordinary skill would further understand that using machine learning models to make materials science predictions can be more efficient than using brute force and measuring every possibly material or sample.
The examiner further notes that LUDWIG expressly contemplates using machine learning to “identify rapidly areas of interest in a ML”, which is precisely what TEHRANI teaches. (LUDWIG, p. 6, left column).
However, LUDWIG and TEHRANI fail to explicitly teach:
determining a confidence value for predictive output of each of the material samples based on the predictive error value
wherein determining the location on the at least one substrate of the next-material sample for characterization … to increase the confidence value by the greatest amount.
However, in a related field of endeavor (machine learning algorithms, see para. 0040), DUTT teaches and makes obvious:
determining a confidence value for predictive output of each of the material samples based on the predictive error value. (DUTT, para. 0049: “In determining whether a model satisfies a target accuracy, the training size calculator may compute a confidence interval for a percentage of example predictions that are below an error threshold and determine whether the target percentile is equal to or below a lower bound of the confidence interval. If the target percentile is equal to or below the lower bound, the unknown actual percentage of example predictions below an error threshold is, with a probability of a defined confidence level, larger than the target percentile. Therefore, the training size calculator may stop adding additional training examples to the training data and return the model.”; (EN): in combination with LUDWIG and TEHRANI, the machine learning model of TEHRANI uses the confidence levels of DUTT, which are tired to an error rate, to determine how confident the model is in the predicted output)
wherein determining the location on the at least one substrate of the next-material sample for characterization includes determination of the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount. (DUTT, para. 0049: “In determining whether a model satisfies a target accuracy, the training size calculator may compute a confidence interval for a percentage of example predictions that are below an error threshold and determine whether the target percentile is equal to or below a lower bound of the confidence interval. If the target percentile is equal to or below the lower bound, the unknown actual percentage of example predictions below an error threshold is, with a probability of a defined confidence level, larger than the target percentile. Therefore, the training size calculator may stop adding additional training examples to the training data and return the model.”; (EN): in combination with LUDWIG and TEHRANI, the machine learning model of TEHRANI uses the confidence levels of DUTT when determining the next material samples to characterize)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI and DUTT as explained above. As disclosed by DUTT, one of ordinary skill would have been motivated to do so because “the confidence interval may be an estimate of an unknown error of the machine learning model on test data.” (para. 0046). One of ordinary skill would understand that using confidence levels, which are tied to error rate (model accuracy), would enable one of ordinary skill to objectively evaluate the correctness of the output of a predictive model and to use such confidence as an indication of the next materials to characterize.
Regarding Claim 5
LUDWIG, TEHRANI, and DUTT disclose the method of claim 1 as explained above. However, LUDWIG and TEHRANI fail to explicitly teach:
wherein characterizing at least some of the definitional samples as characterization training data is determined to be complete upon reaching a predetermined threshold confidence value.
However, in a related field of endeavor (machine learning algorithms, see para. 0040), DUTT teaches:
wherein characterizing at least some of the definitional samples as characterization training data is determined to be complete upon reaching a predetermined threshold confidence value. (DUTT, para. 0049: “In determining whether a model satisfies a target accuracy, the training size calculator may compute a confidence interval for a percentage of example predictions that are below an error threshold and determine whether the target percentile is equal to or below a lower bound of the confidence interval. If the target percentile is equal to or below the lower bound, the unknown actual percentage of example predictions below an error threshold is, with a probability of a defined confidence level, larger than the target percentile. Therefore, the training size calculator may stop adding additional training examples to the training data and return the model.”;
(EN): in combination with LUDWIG and TEHRANI, the machine learning model of TEHRANI uses the confidence levels of DUTT during training to determine when the model is sufficiently accurate to stop training)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI and DUTT as explained above. As disclosed by DUTT, one of ordinary skill would have been motivated to do so because “the confidence interval may be an estimate of an unknown error of the machine learning model on test data.” (para. 0046). One of ordinary skill would understand that using confidence levels, which are tied to error rate (model accuracy), would enable one of ordinary skill to end machine learning training and conserve resources when the machine learning model has reached acceptable accuracy.
Regarding Claim 6
LUDWIG, TEHRANI, and DUTT disclose the method of claim 1 as explained above. However, LUDWIG fails to explicitly teach:
entering the training data into the machine learning model and
outputting a next-location for detection and a predicted output of the corresponding sample, and
detecting definitional data of the next-material sample and comparing the detected definitional data with the predicted output.
However, in a related field of endeavor (synthesis and characterization of materials, see p.9849, “Synthesis and Characterization of Two High-Hardness Materials”), TEHRANI teaches and makes obvious:
entering the training data into the machine learning model and (TEHRANI, p. 9845, “Introduction” section: “This approach employs a combination of compositional and structural descriptors to build a ML model using the Materials Project data as a training set.”
TEHRANI, p. 9845, “Experimental Methods” section: “The machine-learning model was next created based on a Support Vector Machine Regression (SVR) algorithm ... The SVR employed a radial basis function (RBF) as the kernel function and was trained with a 10-fold cross-validation scheme.”
Examiner’s Note (EN): in combination with LUDWIG and DUTT, the machine learning model of TEHRANI is now trained to additionally use the definitional data measured by LUDWIG in addition to the Materials Project dataset of TEHRANI)
outputting a next-location for detection and a predicted output of the corresponding sample, and (TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”;
Examiner’s Note (EN): in combination with LUDWIG and DUTT, the machine learning model of TEHRANI directs “the search for materials with desired mechanical properties” corresponding to “outputting a next-location for detection and a predicted output of the corresponding sample” as recited in this claim because the search in TEHRANI relates to suggesting similar materials with desired properties for further analysis)
detecting definitional data of the next-material sample and comparing the detected definitional data with the predicted output. (TEHRANI, p. 9846, “Machine-Learning Bulk...” section: “As shown in Figure 1a, remarkable agreement is obtained between the DFT-calculated bulk modulus (BDFT) and the ML-predicted values (BSVR) with the cross-validated root-mean-square error (RMSECV)”;
TEHRANI, p. 9847, Fig. 1 explanation: “Cross-validated values of B using the SVR model against the training set”;
TEHRANI, p. 9847, Fig. 2 explanation: “Cross-validated values of G using the SVR model against the training set”;
Examiner’s Note (EN): in combination with LUDWIG and DUTT, the machine learning model of TEHRANI now measures samples (corresponding to “detecting definitional data” which LUDWIG also teaches), and the cross-validating of predicted vs. measured values corresponds to the “comparing the detected definitional data with the predicted output” limitation)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI as explained above. As disclosed by TEHRANI, one of ordinary skill would have been motivated to do so in order to “direct the search for materials with desired mechanical properties.” (TEHRANI, p. 9848, “Screening Crystal Structure ...” section). As further disclosed by TEHRANI, “One area where ML may also be useful is in the search for materials with exceptional mechanical properties, such as high incompressibility or extreme hardness” because machine learning provides advantages over the previous trial-and-error approach. (TEHRANI, p. 9845, “Introduction” section). One of ordinary skill would further understand that using machine learning models to make materials science predictions can be more efficient than using brute force and measuring every possibly material or sample.
The examiner further notes that LUDWIG expressly contemplates using machine learning to “identify rapidly areas of interest in a ML”, which is precisely what TEHRANI teaches. (LUDWIG, p. 6, left column).
Regarding Claim 7
LUDWIG, TEHRANI, and DUTT disclose the method of claim 1 as explained above. LUDWIG further teaches:
wherein physical, chemical, and/or treatment attributes as operational data includes one or more of: precursor gradient among material samples across the at least one substrate, chemical constituent gradient among material samples across the at least one substrate, and treatment gradient by exposure to irradiation with different wavelengths among material samples across the at least one substrate. (LUDWIG, p. 3, left column: “is also possible to synthesize “focused” compositional gradient MLs around, e.g., a predicted composition, where the composition range is tailored in a limited range.”
LUDWIG, p. 5, left column: “These contain e.g., composition spreads around predicted compositions, synthesized simultaneously at different temperatures. This can be realized by gradient and step heater methods.”
LUDWIG, p. 5, right column: “A database of experimental datasets, acquired in a consistent manner by high-throughput experimentation, would be advantageous: as MLs are fabricated in single experiments, all materials in the ML are very well comparable and due to the continuous composition gradients, the quality and consistency of the data is assured.”
Examiner’s Note (EN): LUDWIG teaches a compositional gradient, corresponding to the recited “chemical constituent gradient among material samples across the at least one substrate” limitation)
Regarding Claim 8
LUDWIG, TEHRANI, and DUTT disclose the method of claim 1 as explained above. LUDWIG further teaches:
wherein detecting definitional data includes determining one or more of catalytic activity, electrochemical activity, chemical product distribution resultant from reaction, elemental distribution and/or geometry, mechanical-physical properties, thermal properties, optical properties, catalytic and/or corrosion evolution, and/or fluorescence intensity. (LUDWIG, p. 4, left column: “Further high-throughput characterization methods correspond to the specific material development goals, i.e., functional properties. Adequate screening parameters (e.g., electrical resistivity, Seebeck coefficient, Young’s modulus, hardness, etc.) need to be defined or descriptors of the property of interest, if it cannot be directly measured. In many cases measuring one screening parameter is not enough, and a different ML design might be necessary to enable measuring all necessary parameters. High-throughput electrical resistivity measurements serve well for delivering descriptor values for properties related to it”; Examiner’s Note (EN): As shown in Fig. 2, the characterization of materials includes characterization with respect to at least photoelectric-chemical properties, optical properties, mechanical properties, which correspond to some of the types of definitional data recited in this claim)
Regarding Claim 9
LUDWIG, TEHRANI, and DUTT disclose the method of claim 1 as explained above. However, LUDWIG fails to explicitly teach:
determining one or more physical, chemical, and/or treatment attributes of a next material collection for further characterization.
However, in a related field of endeavor (synthesis and characterization of materials, see p.9849, “Synthesis and Characterization of Two High-Hardness Materials”), TEHRANI teaches and makes obvious:
determining one or more physical, chemical, and/or treatment attributes of a next material collection for further characterization. (TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”; Examiner’s Note (EN): elastic moduli pertains to a physical attribute, and in combination with LUDWIG and DUTT, the machine learning model of TEHRANI is now trained using the data collected and correlated by LUDWIG, to investigate next materials on the substate of LUDWIG; the examiner notes that the broadest reasonable interpretation of “and/or” means that only one alternative limitation needs to be shown in the prior art, and the “elastic moduli” corresponds to a physical attribute of a material (e.g., the elasticity)).
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI and DUTT as explained above. As disclosed by TEHRANI, one of ordinary skill would have been motivated to do so in order to “direct the search for materials with desired mechanical properties.” (TEHRANI, p. 9848, “Screening Crystal Structure ...” section). As further disclosed by TEHRANI, “One area where ML may also be useful is in the search for materials with exceptional mechanical properties, such as high incompressibility or extreme hardness” because machine learning provides advantages over the previous trial-and-error approach. (TEHRANI, p. 9845, “Introduction” section). One of ordinary skill would further understand that using machine learning models to make materials science predictions can be more efficient than using brute force and measuring every possibly material or sample.
The examiner further notes that LUDWIG expressly contemplates using machine learning to “identify rapidly areas of interest in a ML”, which is precisely what TEHRANI teaches. (LUDWIG, p. 6, left column).
Regarding Claim 10
LUDWIG, TEHRANI, and DUTT disclose the method of claim 1 as explained above. LUDWIG further teaches:
wherein detecting definitional data further comprises obtaining definitional data concerning material samples of another known material collection as at least some of the definitional samples. (LUDWIG, p. 5, left column: “The microstructure of thin films is determining their extrinsic properties. Therefore, not only synthesizability and stability of new materials should be addressed by combinatorial materials science, but also the microstructural diversity, which is closely linked to processing. This means for each composition of interest, the search space is extended by the different microstructures a material can be created in. To address this challenge “combinatorial processing” can be applied to create “processing libraries”.; Examiner’s Note: Extension of the search space, by investigating a different library, corresponds to “determining definitional data concerning material samples of another known material collection as at least some of the definitional samples” because using additional substrates with different synthesized materials correspond to “another known material collection”)
Regarding Claim 11
LUDWIG, TEHRANI, and DUTT disclose the method of claim 1 as explained above. LUDWIG further teaches:
wherein the material samples are each defined on the nano- or micro-scale. (LUDWIG, p. 5, right column: “These MLs comprise films of the same composition in different microstructures: amorphous, nano- or microcrystalline, single- or multi-phase constitution, etc.”)
Regarding Claim 12
LUDWIG, TEHRANI, and DUTT disclose the method of claim 11 as explained above. LUDWIG further teaches:
wherein detecting definitional data from at least some of the material samples includes moving between material samples at the nano- or micro-scale. (LUDWIG, p. 4, left column: “Photoelectrochemical properties can be measured with an automated scanning droplet cell, however then the ML needs to be deposited on a homogeneous conductive layer such as Pt. For many materials it is also important to determine their microstructure in high-throughput”;
LUDWIG, p. 5, right column: “These MLs comprise films of the same composition in different microstructures: amorphous, nano- or microcrystalline, single- or multi-phase constitution, etc.”)
Examiner’s Note: An “automated scanning droplet cell” scans, or moves, between samples and LUDWIG describes that samples can be either nano- or micro-crystalline)
Regarding Claim 13
LUDWIG teaches:
A real-time material collection characterization system, the system comprising: (LUDWIG, p. 2, left column: “This perspective discusses the possibilities of combinatorial synthesis and high-throughput characterization in combination with computational methods in the endeavor to efficiently identify new materials in multi-dimensional search spaces”)
a data collection system comprising at least one sensor configured to detect definitional data from at least some material samples of a set of material samples as definitional samples, (LUDWIG, p. 4, left column, “Photoelectrochemical properties can be measured with an automated scanning droplet cell”)
wherein the material samples are positionally encoded on at least one substrate (LUDWIG, p. 3, right column: “Whereas in some cases completely continuous MLs are used, it is in most cases better to use a ML which is structured into measurement areas (MAs), however, without losing to much of the composition spread. In the following examples, the MLs comprise 342 MAs over a 100mm diameter substrate (typically thermally oxidized Si wafer) .... The numbered crosses, where no thin film is applied, are used for thickness measurements (e.g., with profilometry) and are useful for navigation on the ML when performing high-throughput characterization.”;
Examiner’s Note (EN): the measurement areas are arranged on the substrate such that the “numbered crosses” (corresponding to “positionally encoding”) can be used to navigate over the substrate; the examiner notes that the broadest reasonable interpretation of “positionally encoded” includes identifying the location of a material sample, which can be in a grid as shown by Fig. 1 of the instant specification, and LUDWIG discloses (see Fig. 2) a pre-patterned substrate (such as a grid), where each measurement area is numbered (corresponding to recited “positionally encoded”))
according to known physical, chemical, and/or treatment attributes as operational data; and (LUDWIG, p. 2, left column: “Thus, discoveries can be expected in (I) (compositionally) unexplored search spaces, by fabricating and characterizing parts of the unexplored composition space and (II) by performing experiments in (known) composition spaces, by testing for special, but not yet investigated functionalities.”
LUDWIG, p. 2, right column: “A ML is a well-defined set of materials—suitable for high-throughput characterization—produced in one experiment under identical conditions. Combinatorial and high-throughput methods for materials discovery and accelerated development have been developed in the last decades.... They involve, first, identification of the chemical materials compositions and their crystallographic structure. Second, structural and functional properties can be identified and be further optimized by combinatorial processing.”
Examiner’s Note (EN): LUDWIG teaches that chemical and structure (corresponding to “physical”) properties of the materials are known when fabricated on the substrate, corresponding to recited “known physical, chemical, and/or treatment attributes as operational data”)
a characterization control system comprising at least one processor configured to execute instructions stored on memory to conduct characterization of the set of material samples on the at least one substrate, the characterization control system configured to (LUDWIG, p. 5, right column: “Data management should include the FAIR guiding principles: data needs to be findable, accessible, interoperable, and reusable, both for humans and computers”; (EN): a computer has a processor to access data and a memory to store data regarding the characterization of materials)
operate the data collection system to detect definitional data from at least some of the material samples as definitional samples, (LUDWIG, p. 4, left column and Fig. 2: “Further high-throughput characterization methods correspond to the specific material development goals, i.e., functional properties. Adequate screening parameters (e.g., electrical resistivity, Seebeck coefficient, Young’s modulus, hardness, etc.) need to be defined or descriptors of the property of interest, if it cannot be directly measured. In many cases measuring one screening parameter is not enough, and a different ML design might be necessary to enable measuring all necessary parameters. High-throughput electrical resistivity measurements serve well for delivering descriptor values for properties related to it”; Examiner’s Note (EN): As shown in Fig. 2, the characterization of materials includes characterization with respect to electrical properties, magnetic properties, photoelectric-chemical properties, optical properties, microstructure, mechanical properties, composition, and crystal structures and phase constitution; the broadest reasonable interpretation of “definitional data” includes data measured from the material samples to characterized as explained in para. 0041 of the instant specification, and LUDWIG discloses measuring the samples on the substrate as depicted in Fig. 2)
to correlate the definitional data with the operational data, (LUDWIG, p. 3, right column: “The acquired large and consistent datasets of intrinsic and extrinsic properties enable materials discoveries and efficient optimization of identified materials. Furthermore, they are the basis for multifunctional existence diagrams, comprising correlations between composition, processing, structure and properties.; Examiner’s Note (EN): chemical composition and physical structure are known, operational data as disclosed by p. 2, right column (see explanation above), and are correlated with measured properties (corresponding to measured “definitional data” as recited in the claim) (see explanation above), which are correlated at least by the “multifunctional existence diagrams” that provide for correlation of operational data and definitional data))
to determine an experimental output (LUDWIG, p. 4, left column: “Further high-throughput characterization methods correspond to the specific material development goals, i.e., functional properties. Adequate screening parameters (e.g., electrical resistivity, Seebeck coefficient, Young’s modulus, hardness, etc.) need to be defined or descriptors of the property of interest, if it cannot be directly measured. In many cases measuring one screening parameter is not enough, and a different ML design might be necessary to enable measuring all necessary parameters. High-throughput electrical resistivity measurements serve well for delivering descriptor values for properties related to it”)
However, LUDWIG fails to explicitly teach:
to characterize at least some of the definitional samples as characterization training data based on the correlation of extracted definitional and operational data,
wherein the characterization control system includes a machine learning model configured to receive the characterization training data as input, and to output characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data,
wherein the configuration to characterize at least some of the definitional samples as training data includes configuration to determine, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as characterization training data according to the positional encoding via the definitional and the operational data, including configuration to
determine a predicted output,
compare the predicted and experimental outputs for determining a predictive error value, and
determine a confidence value for predictive output of each of the material samples based on the predictive error value;
wherein the configuration to determine the location on the at least one substrate of the next-material sample for characterization includes configuration to determine the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount.
However, in a related field of endeavor (synthesis and characterization of materials, see p.9849, “Synthesis and Characterization of Two High-Hardness Materials”), TEHRANI teaches and makes obvious:
to characterize at least some of the definitional samples as characterization training data based on the correlation of extracted definitional and operational data, (TEHRANI, p. 9846, “Results and Discussion” section: “Machine learning of elastic moduli begins with the construction of a robust training set. In this case, the Materials Project DFT calculated B and G are ideal to train the SVR machine-learning model”;
TEHRANI, p. 9846, “Mechanical Property Measurements” section: “samples analyzed using laboratory X-ray powder diffraction were examined under compression to measure the equation of state (EoS) to compare the experimental and machine learning-predicted bulk modulus”
Examiner’s Note (EN): TEHRANI explicitly teaches creating a training set, corresponding to recited “characterizing at least some of the definitional samples as training data”; in combination with LUDWIG, a training set is constructed as in TEHRANI, using the measured samples (as in both LUDWIG and TEHRANI), including the correlation of the measured data (corresponding to “definitional data”) with known data (corresponding to “operational data”) as disclosed by LUDWIG)
wherein the characterization control system includes a machine learning model configured to receive the characterization training data as input, and to output characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data. (TEHRANI, p. 9845, “Introduction” section: “Here, we develop a method based on machine learning to vastly expand the number of materials with their elastic moduli predicted. This approach employs a combination of compositional and structural descriptors to build a ML model using the Materials Project data as a training set.”
TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”
Examiner’s Note (EN): TEHRANI discloses that the trained machine learning model is used to “direct the search for materials with desired mechanical properties,” corresponding to the recited “other than the definitional samples,”; in combination with LUDWIG, the machine learning model of TEHRANI is now trained using the data collected and correlated by LUDWIG, to investigate materials on the substate of LUDWIG).
wherein the configuration to characterize at least some of the definitional samples as training data includes configuration to determine, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as training data. (TEHRANI, p. 9846, “Results and Discussion” section: “Machine learning of elastic moduli begins with the construction of a robust training set. In this case, the Materials Project DFT calculated B and G are ideal to train the SVR machine-learning model”;
TEHRANI, p. 9848: “Nevertheless, most of the ML approaches developed thus far, irrespective of ML method or choice of descriptors, are able to predict elastic moduli with impressive accuracy in a fraction of the time that it currently takes ab initio calculations.”
TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.” Examiner’s Note (EN): TEHRANI explicitly teaches creating a training set, corresponding to recited “characterizing at least some of the definitional samples as characterization training data”; in combination with LUDWIG (which teaches “numbered crosses” that define “measurement areas (MAs)” on the patterned substrate, se p. 3, right column), the machine learning model of TEHRANI directs “the search for materials with desired mechanical properties” on the patterned substrate of LUDWIG, corresponding to recited “determining, by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as training data” as recited in this claim because the search in TEHRANI relates to suggesting similar materials with desired properties for further analysis and can suggest similar materials based on the location of the “numbered crosses” of LUDWIG and taking into account the definitional and operational data of LUDWIG)
… determine the location on the at least one substrate of the next-material sample (TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI as explained above. As disclosed by TEHRANI, one of ordinary skill would have been motivated to do so in order to “direct the search for materials with desired mechanical properties.” (TEHRANI, p. 9848, “Screening Crystal Structure ...” section). As further disclosed by TEHRANI, “One area where ML may also be useful is in the search for materials with exceptional mechanical properties, such as high incompressibility or extreme hardness” because machine learning provides advantages over the previous trial-and-error approach. (TEHRANI, p. 9845, “Introduction” section). One of ordinary skill would further understand that using machine learning models to make materials science predictions can be more efficient than using brute force and measuring every possibly material or sample.
The examiner further notes that LUDWIG expressly contemplates using machine learning to “identify rapidly areas of interest in a ML”, which is precisely what TEHRANI teaches. (LUDWIG, p. 6, left column).
However, LUDWIG and TEHRANI fail to explicitly teach:
determine a confidence value for predictive output of each of the material samples based on the predictive error value
wherein the configuration to determine the location on the at least one substrate of the next-material sample for characterization … to increase the confidence value by the greatest amount.
However, in a related field of endeavor (machine learning algorithms, see para. 0040), DUTT teaches and makes obvious:
determine a confidence value for predictive output of each of the material samples based on the predictive error value. (DUTT, para. 0049: “In determining whether a model satisfies a target accuracy, the training size calculator may compute a confidence interval for a percentage of example predictions that are below an error threshold and determine whether the target percentile is equal to or below a lower bound of the confidence interval. If the target percentile is equal to or below the lower bound, the unknown actual percentage of example predictions below an error threshold is, with a probability of a defined confidence level, larger than the target percentile. Therefore, the training size calculator may stop adding additional training examples to the training data and return the model.”; (EN): in combination with LUDWIG and TEHRANI, the machine learning model of TEHRANI uses the confidence levels of DUTT, which are tired to an error rate, to determine how confident the model is in the predicted output)
wherein the configuration to determine the location on the at least one substrate of the next-material sample for characterization includes configuration to determine the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount. (DUTT, para. 0049: “In determining whether a model satisfies a target accuracy, the training size calculator may compute a confidence interval for a percentage of example predictions that are below an error threshold and determine whether the target percentile is equal to or below a lower bound of the confidence interval. If the target percentile is equal to or below the lower bound, the unknown actual percentage of example predictions below an error threshold is, with a probability of a defined confidence level, larger than the target percentile. Therefore, the training size calculator may stop adding additional training examples to the training data and return the model.”; (EN): in combination with LUDWIG and TEHRANI, the machine learning model of TEHRANI uses the confidence levels of DUTT when determining the next material samples to characterize)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI and DUTT as explained above. As disclosed by DUTT, one of ordinary skill would have been motivated to do so because “the confidence interval may be an estimate of an unknown error of the machine learning model on test data.” (para. 0046). One of ordinary skill would understand that using confidence levels, which are tied to error rate (model accuracy), would enable one of ordinary skill to objectively evaluate the correctness of the output of a predictive model and to use such confidence as an indication of the next materials to characterize.
Claim 17 depends from claim 13 and claims a system that corresponds to the method of claim 5 and is therefore rejected for the same reasons explained above with respect to claims 5 and 13.
Claim 18 depends from claim 13 and claims a system that corresponds to the method of claim 6 and is therefore rejected for the same reasons explained above with respect to claims 6 and 13.
Claim 19 depends from claim 13 and claims a system that corresponds to the method of claim 7 and is therefore rejected for the same reasons explained above with respect to claims 7and 13.
Claim 20 depends from claim 13 and claims a system that corresponds to the method of claim 8 and is therefore rejected for the same reasons explained above with respect to claims 8 and 13.
Claim 21 depends from claim 13 and claims a system that corresponds to the method of claim 9 and is therefore rejected for the same reasons explained above with respect to claims 9 and 13.
Claim 22 depends from claim 13 and claims a system that corresponds to the method of claim 10 and is therefore rejected for the same reasons explained above with respect to claims 10 and 13.
Claim 23 depends from claim 13 and claims a system that corresponds to the method of claim 11 and is therefore rejected for the same reasons explained above with respect to claims 11 and 13.
Claim 24 depends from claim 23 and claims a system that corresponds to the method of claim 12 and is therefore rejected for the same reasons explained above with respect to claims 12 and 23.
Regarding Claim 25
LUDWIG traches:
A method of characterizing a material chip in real-time, the method comprising: (LUDWIG, p. 2, left column: “This perspective discusses the possibilities of combinatorial synthesis and high-throughput characterization in combination with computational methods in the endeavor to efficiently identify new materials in multi-dimensional search spaces”;
LUDWIG, p. 3, right column: “the MLs comprise 342 MAs over a 100mm diameter substrate (typically thermally oxidized Si wafer)”)
detecting definitional data from material samples as definitional samples, (LUDWIG, p. 4, left column: “Further high-throughput characterization methods correspond to the specific material development goals, i.e., functional properties. Adequate screening parameters (e.g., electrical resistivity, Seebeck coefficient, Young’s modulus, hardness, etc.) need to be defined or descriptors of the property of interest, if it cannot be directly measured. In many cases measuring one screening parameter is not enough, and a different ML design might be necessary to enable measuring all necessary parameters. High-throughput electrical resistivity measurements serve well for delivering descriptor values for properties related to it”; Examiner’s Note (EN): As shown in Fig. 2, the characterization of materials includes characterization with respect to electrical properties, magnetic properties, photoelectric-chemical properties, optical properties, microstructure, mechanical properties, composition, and crystal structures and phase constitution; the broadest reasonable interpretation of “definitional data” includes data measured from the material samples to characterized as explained in para. 0041 of the instant specification, and LUDWIG discloses measuring the samples on the substrate as depicted in Fig. 2)
the material samples positionally encoded on a substrate (LUDWIG, p. 3, right column: “Whereas in some cases completely continuous MLs are used, it is in most cases better to use a ML which is structured into measurement areas (MAs), however, without losing to much of the composition spread. In the following examples, the MLs comprise 342 MAs over a 100mm diameter substrate (typically thermally oxidized Si wafer) see Fig. 2. The crosses, defining Mas, are fabricated by a photolithographic lift-off process to achieve a pre-patterned substrate. The numbered crosses, where no thin film is applied, are used for thickness measurements (e.g., with profilometry) and are useful for navigation on the ML when performing high-throughput characterization.”;
Examiner’s Note (EN): the measurement areas are arranged on the substrate such that the “numbered crosses” (corresponding to “positionally encoding”) can be used to navigate over the substrate; the examiner notes that the broadest reasonable interpretation of “positionally encoding” includes identifying the location of a material sample, which can be in a grid as shown by Fig. 1 of the instant specification, and LUDWIG discloses (see Fig. 2) a pre-patterned substrate (such as a grid), where each measurement area is numbered (corresponding to recited “positional encoding”))
according to known physical, chemical, and/or treatment attributes as operational data; (LUDWIG, p. 2, left column: “Thus, discoveries can be expected in (I) (compositionally) unexplored search spaces, by fabricating and characterizing parts of the unexplored composition space and (II) by performing experiments in (known) composition spaces, by testing for special, but not yet investigated functionalities.”
LUDWIG, p. 2, right column: “A ML is a well-defined set of materials—suitable for high-throughput characterization—produced in one experiment under identical conditions. Combinatorial and high-throughput methods for materials discovery and accelerated development have been developed in the last decades.... They involve, first, identification of the chemical materials compositions and their crystallographic structure. Second, structural and functional properties can be identified and be further optimized by combinatorial processing.”
Examiner’s Note (EN): LUDWIG teaches that chemical and structure (corresponding to “physical”) properties of the materials are known when fabricated on the substrate, corresponding to recited “known physical, chemical, and/or treatment attributes as operational data”)
correlating the definitional data with the operational data, (LUDWIG, p. 3, right column: “The acquired large and consistent datasets of intrinsic and extrinsic properties enable materials discoveries and efficient optimization of identified materials. Furthermore, they are the basis for multifunctional existence diagrams, comprising correlations between composition, processing, structure and properties.; Examiner’s Note (EN): chemical composition and physical structure are known, operational data as disclosed by p. 2, right column (see explanation above), and are correlated with measured properties (corresponding to measured “definitional data” as recited in the claim)
including correlating based on definitional data obtained from another material chip having materials samples positionally encoded on a substrate according to known physical, chemical, and/or treatment attributes as operational data; (LUDWIG, p. 5, left column: “The microstructure of thin films is determining their extrinsic properties. Therefore, not only synthesizability and stability of new materials should be addressed by combinatorial materials science, but also the microstructural diversity, which is closely linked to processing. This means for each composition of interest, the search space is extended by the different microstructures a material can be created in. To address this challenge “combinatorial processing” can be applied to create “processing libraries”.; Examiner’s Note: Extension of the search space, by investigating a different library, corresponds to “determining definitional data concerning material samples of another known material collection as at least some of the definitional samples” because using additional substrates with different synthesized materials correspond to “another known material collection” where such substrates have MAs too as explained above with respect to this claim)
determining an experimental output (LUDWIG, p. 4, left column: “Further high-throughput characterization methods correspond to the specific material development goals, i.e., functional properties. Adequate screening parameters (e.g., electrical resistivity, Seebeck coefficient, Young’s modulus, hardness, etc.) need to be defined or descriptors of the property of interest, if it cannot be directly measured. In many cases measuring one screening parameter is not enough, and a different ML design might be necessary to enable measuring all necessary parameters. High-throughput electrical resistivity measurements serve well for delivering descriptor values for properties related to it”)
However, LUDWIG fails to explicitly teach:
characterizing at least some of the definitional samples as characterization training data based on correlation of the definitional and the operational data; and
inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data.
wherein characterizing at least some of the definitional samples as characterization training data includes determining, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as characterization training data according to the positional encoding via the definitional and the operational data
determining a predicted output,
comparing the predicted and experimental outputs to determine a predictive error value, and
determining a confidence value for predictive output of each of the material samples based on the predictive error value;
wherein determining the location on the at least one substrate of the next-material sample for characterization includes determination of the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount.
However, in a related field of endeavor (synthesis and characterization of materials, see p.9849, “Synthesis and Characterization of Two High-Hardness Materials”), TEHRANI teaches and makes obvious:
characterizing at least some of the definitional samples as training data based on correlation of the definitional and the operational data; and (TEHRANI, p. 9846, “Results and Discussion” section: “Machine learning of elastic moduli begins with the construction of a robust training set. In this case, the Materials Project DFT calculated B and G are ideal to train the SVR machine-learning model”;
TEHRANI, p. 9846, “Mechanical Property Measurements” section: “samples analyzed using laboratory X-ray powder diffraction were examined under compression to measure the equation of state (EoS) to compare the experimental and machine learning-predicted bulk modulus”
Examiner’s Note (EN): in combination with LUDWIG, a training set is constructed as in LUDWIG, using the measured samples (as in both LUDWIG and TEHRANI), including the correlation of the measured with known data as disclosed by LUDWIG)
inputting the characterization training data to a machine learning model and outputting characterization of at least a portion of the set of material samples other than the definitional samples, based on the characterization training data. (TEHRANI, p. 9845, “Introduction” section: “Here, we develop a method based on machine learning to vastly expand the number of materials with their elastic moduli predicted. This approach employs a combination of compositional and structural descriptors to build a ML model using the Materials Project data as a training set.”
TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”
Examiner’s Note (EN): TEHRANI discloses that the trained machine learning model is used to “direct the search for materials with desired mechanical properties,” corresponding to the recited “other than the definitional samples,”; in combination with LUDWIG, the machine learning model of TEHRANI is now trained using the data collected and correlated by LUDWIG, to investigate materials on the substate of LUDWIG).
wherein characterizing at least some of the definitional samples as characterization training data includes determining, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as characterization training data according to the positional encoding via the definitional and the operational data (TEHRANI, p. 9846, “Results and Discussion” section: “Machine learning of elastic moduli begins with the construction of a robust training set. In this case, the Materials Project DFT calculated B and G are ideal to train the SVR machine-learning model”;
TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”;
Examiner’s Note (EN): TEHRANI explicitly teaches creating a training set, corresponding to recited “characterizing at least some of the definitional samples as characterization training data”; in combination with LUDWIG (which teaches “numbered crosses” that define “measurement areas (MAs)” on the patterned substrate, se p. 3, right column), the machine learning model of TEHRANI directs “the search for materials with desired mechanical properties” on the patterned substrate of LUDWIG, corresponding to recited “determining, in real-time by the machine learning model, a location on the at least one substrate of a next-material sample for characterization as training data” as recited in this claim because the search in TEHRANI relates to suggesting similar materials with desired properties for further analysis and can suggest similar materials based on the location of the “numbered crosses” of LUDWIG and taking into account the definitional and operational data of LUDWIG)
determining a predicted output, (TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”)
comparing the predicted and experimental outputs to determine a predictive error value, and (TEHRANI, p. 9846, “Machine-Learning Bulk...” section: “As shown in Figure 1a, remarkable agreement is obtained between the DFT-calculated bulk modulus (BDFT) and the ML-predicted values (BSVR) with the cross-validated root-mean-square error (RMSECV)”l Examiner’s Note (EN): in combination with LUDWIG, the machine learning model of TEHRANI now measures samples (corresponding to “experimental outputs”, which also include outputs from measurements of LUDWIG), and the cross-validating of predicted vs. measured values corresponds to the “comparing the predicted and experimental outputs to determine a predictive error value” limitation because the cross-validating output is a root mean square error)
… includes determination of the location on the at least one substrate of the next-material sample (TEHRANI, p. 9848, “Screening Crystal Structure ...” section: “The ultimate goal of ML is to employ a training set that is capable of predicting the elastic moduli for entire crystallographic databases, which can then act as a proxy to direct the search for materials with desired mechanical properties.”)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI as explained above. As disclosed by TEHRANI, one of ordinary skill would have been motivated to do so in order to “direct the search for materials with desired mechanical properties.” (TEHRANI, p. 9848, “Screening Crystal Structure ...” section). As further disclosed by TEHRANI, “One area where ML may also be useful is in the search for materials with exceptional mechanical properties, such as high incompressibility or extreme hardness” because machine learning provides advantages over the previous trial-and-error approach. (TEHRANI, p. 9845, “Introduction” section). One of ordinary skill would further understand that using machine learning models to make materials science predictions can be more efficient than using brute force and measuring every possibly material or sample.
The examiner further notes that LUDWIG expressly contemplates using machine learning to “identify rapidly areas of interest in a ML”, which is precisely what TEHRANI teaches. (LUDWIG, p. 6, left column).
However, LUDWIG and TEHRANI fail to explicitly teach:
determining a confidence value for predictive output of each of the material samples based on the predictive error value
wherein determining the location on the at least one substrate of the next-material sample for characterization … to increase the confidence value by the greatest amount.
However, in a related field of endeavor (machine learning algorithms, see para. 0040), DUTT teaches and makes obvious:
determining a confidence value for predictive output of each of the material samples based on the predictive error value. (DUTT, para. 0049: “In determining whether a model satisfies a target accuracy, the training size calculator may compute a confidence interval for a percentage of example predictions that are below an error threshold and determine whether the target percentile is equal to or below a lower bound of the confidence interval. If the target percentile is equal to or below the lower bound, the unknown actual percentage of example predictions below an error threshold is, with a probability of a defined confidence level, larger than the target percentile. Therefore, the training size calculator may stop adding additional training examples to the training data and return the model.”; (EN): in combination with LUDWIG and TEHRANI, the machine learning model of TEHRANI uses the confidence levels of DUTT, which are tired to an error rate, to determine how confident the model is in the predicted output)
wherein determining the location on the at least one substrate of the next-material sample for characterization includes determination of the location on the at least one substrate of the next-material sample to increase the confidence value by the greatest amount. (DUTT, para. 0049: “In determining whether a model satisfies a target accuracy, the training size calculator may compute a confidence interval for a percentage of example predictions that are below an error threshold and determine whether the target percentile is equal to or below a lower bound of the confidence interval. If the target percentile is equal to or below the lower bound, the unknown actual percentage of example predictions below an error threshold is, with a probability of a defined confidence level, larger than the target percentile. Therefore, the training size calculator may stop adding additional training examples to the training data and return the model.”; (EN): in combination with LUDWIG and TEHRANI, the machine learning model of TEHRANI uses the confidence levels of DUTT when determining the next material samples to characterize)
Before the effective filing date of the present application, it would have been obvious to one of ordinary skill in the art to combine the teachings of LUDWIG with TEHRANI and DUTT as explained above. As disclosed by DUTT, one of ordinary skill would have been motivated to do so because “the confidence interval may be an estimate of an unknown error of the machine learning model on test data.” (para. 0046). One of ordinary skill would understand that using confidence levels, which are tied to error rate (model accuracy), would enable one of ordinary skill to objectively evaluate the correctness of the output of a predictive model and to use such confidence as an indication of the next materials to characterize.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 20220380709 A1 (Nelson). “Addition of the curing agent shortly before printing does not affect compartment generation but permanent solidification and trapping of the droplets allows for easier handling and imaging of the material during characterization as there is no danger of yielding the bath. The array of embedded compartments 1230 several days after printing (as depicted in FIG. 12) shows varying coloured compartments, which is evidence of differently sized nanoparticles having grown within each compartment.” (para. 0146). See. Fig. 6.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHAEL C LEE whose telephone number is (571)272-4933. The examiner can normally be reached M-F 12:00 pm - 8:00 pm ET.
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, Omar Fernandez Rivas can be reached at 571-272-2589. 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.
/MICHAEL C. LEE/Examiner, Art Unit 2128
Prosecution Insights