DETAILED ACTION
This Office Action is responsive to the Applicant’s submission, filed on February 18, 2026, amending claims 1, 5-7, 9, 13, 15, 17 and 18. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1-6 and 15-18 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by the article entitled “Compensated-DNN: Energy Efficient Low-Precision Deep Neural Networks by Compensating Quantization Errors” by Jain et al. (“Jain”).
Regarding claim 1, Jain generally describes “compensated-DNN,” in which errors introduced due to quantization are dynamically compensated for during execution (see e.g. the Abstract). Like claimed, Jain particularly teaches:
receiving, at a neural network weight layer of an artificial neural network, an incoming excitation vector, the artificial neural network including one or more operations involving one or more scalar values to be computed across an output data vector of the artificial neural network (see e.g. section 3.1 “Background:” Jain discloses that deep neural networks are comprised of multiple layers of neurons, in which each neuron computes a dot-product of an input and its weights. Jain further discloses that a deep neural network can be quantized, in which the weights and input values are reduced in precision, but that such quantization produces quantization error and reduces the accuracy of the neural network – see e.g. section 1 “INTRODUCTION.” To compensate for the quantization error, Jain teaches adding bits to the inputs and weights to indicate the error incurred due to quantization of the respective inputs and weights, and thus produces a “Fixed-Point with Error Compensation” number format – see e.g. section 4.2 “Fixed Point with Error Compensation.” Jain further discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, the added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to the quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. The compensation sum is added to the dot-product output to produce a compensated output – see e.g. section 4.3 “Computation vs. Compensation Cost.” The inputs to the neural network layer are considered an “incoming excitation vector” like claimed. Accordingly, Jain teaches receiving, at a neural network weight layer of an artificial neural network, an incoming excitation vector, and wherein the artificial neural network includes one or more operations involving one or more scalar values, i.e. a quantization error, to be computed across an output data vector of the artificial neural network.); and
using a predicted representation of the one or more scalar values generated simultaneously with computation of the output data vector during forward inference of the artificial neural network by the incoming excitation vector to apply the one or more operations to the output data vectors (see e.g. section 4.3 “Computation vs. Compensation Cost:” as noted above, Jain discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. The compensation sum is considered a “predicted representation of the one or more scalar values” like claimed. Accordingly, Jain teaches using a predicted representation of the one or more scalar values generated simultaneously with computation of the output data vector during forward inference of the artificial neural network by the incoming excitation vector to apply the one or more operations to the output data vector, as is claimed.).
Consequently, Jain teaches a method like that of claim 1.
As per claim 2, Jain further teaches that a computation used to compute an exact representation of the one or more scalar values from the output data vector is avoided (as noted above, Jain discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. A computation used to compute an exact representation of the compensation error from the output data vector is thus avoided.). Accordingly, Jain further teaches a method like that of claim 2.
As per claim 3, Jain further teaches that the one or more operations can include a mean (e.g. “half the resolution”) and/or a standard deviation (see e.g. section 4.2 “Fixed Point with Error Compensation”). Accordingly, Jain further teaches a method like that of claim 3.
As per claim 4, Jain further teaches that the one or more operations involve access to every element of the output data vector (i.e. to compute a compensated dot-product output for every output of the output data vector) (see e.g. section 4.1 “Compensated-DNN: Insights & Approach” and section 4.3 “Computation vs. Compensation Cost.”). Accordingly, Jain further teaches a method like that of claim 4.
As per claim 5, Jain further teaches providing training input from a set of training data (i.e. a labelled dataset) to the artificial neural network, the training input providing trained weights for the neural network weight layer (see e.g. section 3.1 “Background”). Accordingly, Jain further teaches a method like that of claim 5.
As per 6, Jain further teaches (i) providing additional training weights for predicting the one or more scalar values from the incoming excitation vector used to compute the output data vector, and (ii) producing the predicted representation simultaneously with the computation of the output data vector based on the additional training weights (see e.g. section 4.2 “Fixed Point with Error Compensation:” as noted above, Jain teaches adding bits to the inputs and weights of a neural network layer to indicate the error incurred due to quantization of the respective inputs and weights, and thus produces a “Fixed-Point with Error Compensation” number format. Jain further discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, the added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to the quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. The compensation sum is added to the dot-product output to produce a compensated output – see e.g. section 4.3 “Computation vs. Compensation Cost.” As further noted above, the inputs to the neural network layer are considered an “incoming excitation vector” like claimed. The bits added to the inputs and/or weights of the neural network layer are considered additional “training weights,” when given the broadest reasonable interpretation of the term. Accordingly, Jain teaches: providing additional training weights, i.e. bits added to the inputs and/or weights of a neural network layer, for predicting the one or more scalar values, i.e. the error approximation, from the incoming excitation vector used to compute the output data vector; and producing the predicted representation simultaneously with the computation of the output data vector based on the additional training weights.). Accordingly, Jain further teaches a method like that of claim 6.
Regarding claim 15, Jain generally describes “compensated-DNN,” in which errors introduced due to quantization are dynamically compensated for during execution (see e.g. the Abstract). Like claimed, Jain particularly teaches:
receiving, at a neural network weight layer of an artificial neural network, an incoming excitation vector, the artificial neural network including one or more operations involving one or more scalar values to be computed across an output data vector of the artificial neural network (see e.g. section 3.1 “Background:” Jain discloses that deep neural networks are comprised of multiple layers of neurons, in which each neuron computes a dot-product of an input and its weights. Jain further discloses that a deep neural network can be quantized, in which the weights and input values are reduced in precision, but that such quantization produces quantization error and reduces the accuracy of the neural network – see e.g. section 1 “INTRODUCTION.” To compensate for the quantization error, Jain teaches adding bits to the inputs and weights to indicate the error incurred due to quantization of the respective inputs and weights, and thus produces a “Fixed-Point with Error Compensation” number format – see e.g. section 4.2 “Fixed Point with Error Compensation.” Jain further discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, the added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to the quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. The compensation sum is added to the dot-product output to produce a compensated output – see e.g. section 4.3 “Computation vs. Compensation Cost.” The inputs to the neural network layer are considered an “incoming excitation vector” like claimed. Accordingly, Jain teaches receiving, at a neural network weight layer of an artificial neural network, an incoming excitation vector, and wherein the artificial neural network includes one or more operations involving one or more scalar values, i.e. a quantization error, to be computed across an output data vector of the artificial neural network.); and
using a predicted representation of the one or more scalar values generated simultaneously with computation of the output data vector during forward inference of the artificial neural network by the incoming excitation vector to apply the one or more operations to the output data vectors while avoiding a computation used to compute an exact representation of the one or more scalar values from the output data vector (see e.g. section 4.3 “Computation vs. Compensation Cost:” as noted above, Jain discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. The compensation sum is considered a “predicted representation of the one or more scalar values” like claimed. Accordingly, Jain teaches using a predicted representation of the one or more scalar values generated simultaneously with computation of the output data vector during forward inference of the artificial neural network by the incoming excitation vector to apply the one or more operations to the output data vector, as is claimed. A computation used to compute an exact representation of the compensation error from the output data vector is thus avoided.).
Jain discloses that the above-described tasks can be implemented via computer readable program code, which is necessarily stored in memory for execution by a computer (see e.g. section 5 “EXPERIMENTAL METHODLOGY”). Such a memory comprising computer-readable program code to implement the above-described teachings of Jain is considered a non-transitory computer readable storage medium like that of claim 15.
As per claim 16, Jain further teaches that the one or more operations can include a mean (e.g. “half the resolution”) and/or a standard deviation (see e.g. section 4.2 “Fixed Point with Error Compensation”). Jain also teaches that the one or more operations involve access to every element of the output data vector (i.e. to compute a compensated dot-product output for every output of the output data vector) (see e.g. section 4.1 “Compensated-DNN: Insights & Approach” and section 4.3 “Computation vs. Compensation Cost.”). Accordingly, Jain further teaches a non-transitory computer readable storage medium like that of claim 16.
As per claim 17, Jain further teaches providing training input from a set of training data (i.e. a labelled dataset) to the artificial neural network, the training input providing trained weights for the neural network weight layer (see e.g. section 3.1 “Background”). Accordingly, Jain further teaches a non-transitory computer readable storage medium like that of claim 17.
As per claim 18, Jain further teaches (i) providing additional training weights for predicting the one or more scalar values from the incoming excitation vector used to compute the output data vector, and (ii) producing the predicted representation simultaneously with the computation of the output data vector based on the additional training weights (see e.g. section 4.2 “Fixed Point with Error Compensation:” as noted above, Jain teaches adding bits to the inputs and weights of a neural network layer to indicate the error incurred due to quantization of the respective inputs and weights, and thus produces a “Fixed-Point with Error Compensation” number format. Jain further discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, the added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to the quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. The compensation sum is added to the dot-product output to produce a compensated output – see e.g. section 4.3 “Computation vs. Compensation Cost.” As further noted above, the inputs to the neural network layer are considered an “incoming excitation vector” like claimed. The bits added to the inputs and/or weights of the neural network layer are considered additional “training weights,” when given the broadest reasonable interpretation of the term. Accordingly, Jain teaches: providing additional training weights, i.e. bits added to the inputs and/or weights of a neural network layer, for predicting the one or more scalar values, i.e. the error approximation, from the incoming excitation vector used to compute the output data vector; and producing the predicted representation simultaneously with the computation of the output data vector based on the additional training weights.). Accordingly, Jain further teaches a non-transitory computer readable storage medium like that of claim 18.
Claims 7, 9-13 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over the article to Jain cited above, and also over the article entitled, “New Memristor-Based Crossbar Array Architecture with 50-% Area Reduction and 48-% Power Saving for Matrix-Vector Multiplication of Analog Neuromorphic Computing” by Truong et al. (“Truong”).
Regarding claim 7, Jain teaches a method like that of claim 5, as is described above, which entails performing one or more operations involving one or more scalar values to be computed across an output data vector of an artificial neural network, and providing training input to an artificial neural network, wherein the training input provides training weights for the neural network weight layer. Jain demonstrates that additional training weights can be provided on an artificial intelligence tile comprising the neural network weight layer (see e.g. section 4.3 “Computation vs. Compensation Cost” and Figure 8: Jain demonstrates a processing element design that comprises a computation unit for computing a dot product between the neural network weights and an input. The processing element design further comprises a compensation unit for computing a compensation sum using quantization error bits added to the neural network layer weights and inputs – see “Computation vs. Compensation Cost” and Figure 8. Like also noted above in the rejection for claim 6, the quantization error bits added to the inputs and/or weights of the neural network layer are considered additional “training weights,” when given the broadest reasonable interpretation of the term. Jain thus demonstrates that additional training weights, i.e. added quantization error bits, are provided on an artificial intelligence tile, i.e. on a processing element, comprising the neural network weight layer.). Jain, however, does not explicitly disclose that additional training weights are provided on one or more columns of an analog artificial intelligence tile comprising the neural network weight layer, as is required by claim 7.
Truong nevertheless generally teaches adding one or more columns to an analog artificial intelligence tile comprising a neural network weight layer to perform particular functionality (see e.g. section II “New Memristor-based crossbar array,” and in particular FIG. 3 therein).
It would have been obvious to one of ordinary skill in the art, having the teachings of Jain and Truong before the effective filing date of the claimed invention, to implement the method taught by Jain on an analog artificial intelligence tile like taught by Truong, and such that the additional training weights are provided on one or more columns of the analog artificial intelligence tile comprising the neural network weight layer. It would have been advantageous to one of ordinary skill to utilize such a combination because it can provide better area density and energy efficiency, as is taught by Truong (see e.g. the Abstract). Accordingly, Jain and Truong are considered to teach, to one of ordinary skill in the art, a method like that of claim 7.
Regarding claim 9, Jain generally describes “compensated-DNN,” in which errors introduced due to quantization are dynamically compensated for during execution (see e.g. the Abstract). Like claimed, Jain particularly teaches:
receiving, at a neural network weight layer of an artificial neural network, an incoming excitation vector (see e.g. section 3.1 “Background:” Jain discloses that deep neural networks are comprised of multiple layers of neurons, in which each neuron computes a dot-product of an input and its weights. The inputs to the neural network layer are considered an “incoming excitation vector” like claimed. Accordingly, Jain teaches receiving, at a neural network weight layer of an artificial neural network, an incoming excitation vector.);
computing an output data vector based on trained weights in the neural network weight layer (see e.g. section 3.1 “Background:” as noted above, Jain discloses that deep neural networks are comprised of multiple layers of neurons, in which each neuron computes a dot-product of an input and its weights. The dot products output by the neural network layer are considered an output data vector, which is based on trained weights in the neural network weight layer.);
storing additional trained weights distinct from the trained weights of the neural network weight layer for providing a predicted representation of one or more scalar values to be applied to the output data vector (see e.g. section 1 “INTRODUCTION:” Jain further discloses that a deep neural network can be quantized, in which the weights and input values are reduced in precision, but that such quantization produces quantization error and reduces the accuracy of the neural network. To compensate for the quantization error, Jain teaches adding bits to the inputs and weights to indicate the error incurred due to quantization of the respective inputs and weights, and thus produces a “Fixed-Point with Error Compensation” number format – see e.g. section 4.2 “Fixed Point with Error Compensation.” Jain further discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, the added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to the quantization – see e.g. section 4.3 “Computation vs. Compensation Cost”. The compensation sum is added to the dot-product output to produce a compensated output – see e.g. section 4.3 “Computation vs. Compensation Cost.” The quantization error bits added to the inputs and/or weights of the neural network layer are considered additional “training weights,” when given the broadest reasonable interpretation of the term. Accordingly, Jain teaches storing additional training weights, i.e. bits added to the inputs and/or weights of a neural network layer, which are distinct from the trained weights of the neural network weight layer and are for providing a predicted representation of one or more scalar values, i.e. the error approximation, to be applied to the output data vector.); and
using the predicted representation of the one or more scalar values generated simultaneously with computing the output data vector during forward inference of the artificial neural network by the incoming excitation vector to apply the one or more operations to the output data vector, while avoiding a computation used to compute an exact representation of the one or more scalar values from the output data vector (see e.g. section 4.3 “Computation vs. Compensation Cost:” as noted above, Jain discloses that, simultaneously with computing the dot product of the quantized input and neural network weights, added quantization error bits are used to simultaneously compute a compensation sum, which is an approximation of the error in the dot-product output due to quantization. The compensation sum is considered a “predicted representation of the one or more scalar values” like claimed. Accordingly, Jain teaches using a predicted representation of the one or more scalar values generated simultaneously with computation of the output data vector during forward inference of the artificial neural network by the incoming excitation vector to apply the one or more operations to the output data vector, as is claimed. A computation used to compute an exact representation of the compensation error from the output data vector is thus avoided.).
Consequently, Jain teaches a method similar to that of claim 9 but does not explicitly disclose that the neural network is implemented on an analog artificial intelligence tile, wherein the additional trained weights are stored in one or more rows or columns of the artificial intelligence tile, as is required by claim 9.
Truong nevertheless generally teaches implementing a neural network on an analog artificial intelligence tile wherein one or more rows or columns are added to the artificial intelligence tile to perform particular functionality (see e.g. section II “New Memristor-based crossbar array,” and in particular FIG. 3 therein).
It would have been obvious to one of ordinary skill in the art, having the teachings of Jain and Truong before the effective filing date of the claimed invention, to implement the method taught by Jain on an analog artificial intelligence tile like taught by Truong, and such that the additional training weights are provided on one or more rows or columns of the analog artificial intelligence tile comprising the neural network weight layer. It would have been advantageous to one of ordinary skill to utilize such a combination because it can provide better area density and energy efficiency, as is taught by Truong (see e.g. the Abstract). Accordingly, Jain and Truong are considered to teach, to one of ordinary skill in the art, a computer implemented method like that of claim 9.
As per claim 10, Jain further teaches that the one or more operations can include a mean (e.g. “half the resolution”) and/or a standard deviation (see e.g. section 4.2 “Fixed Point with Error Compensation”). Accordingly, the above-described combination of Jain and Truong further teaches a method like that of claim 10.
As per claim 11, Jain further teaches that the one or more operations involve access to every element of the output data vector (i.e. to compute a compensated dot-product output for every output of the output data vector) (see e.g. section 4.1 “Compensated-DNN: Insights & Approach” and section 4.3 “Computation vs. Compensation Cost.”). Accordingly, the above-described combination of Jain and Truong further teaches a method like that of claim 11.
As per claim 12, Jain further teaches providing training input from a set of training data (i.e. a labelled dataset) to the artificial neural network, the training input providing the trained weights for the neural network weight layer (see e.g. section 3.1 “Background”). Accordingly, the above-described combination of Jain and Truong further teaches a method like that of claim 12.
As per claim 13, Jain further teaches producing the predicted representation (i.e. the compensation sum) simultaneously with the computation of the output data vector based on the additional training weights (see e.g. section 4.3 “Computation vs. Compensation Cost”). Accordingly, the above-described combination of Jain and Truong further teaches a method like that of claim 13.
Regarding claim 19, Jain teaches a non-transitory computer readable storage medium like that of claim 18, as is described above, which entails providing additional training weights for predicting one or more scalar values from the incoming excitation vector used to compute the output data vector, and producing the predicted representation simultaneously with the computation of the output data vector based on the additional training weights. Jain demonstrates that the additional training weights can be provided on an artificial intelligence tile comprising the neural network weight layer (see e.g. section 4.3 “Computation vs. Compensation Cost” and Figure 8: Jain demonstrates a processing element design that comprises a computation unit for computing a dot product between the neural network weights and an input. The processing element design further comprises a compensation unit for computing a compensation sum using quantization error bits added to the neural network layer weights and inputs – see “Computation vs. Compensation Cost” and Figure 8. Like noted above, the quantization error bits added to the inputs and/or weights of the neural network layer are considered additional “training weights,” when given the broadest reasonable interpretation of the term. Jain thus demonstrates that additional training weights, i.e. added quantization error bits, are provided on an artificial intelligence tile, i.e. on a processing element, comprising the neural network weight layer.). Jain, however, does not explicitly disclose that additional training weights are provided on one or more columns of an analog artificial intelligence tile comprising the neural network weight layer, as is required by claim 19.
Truong nevertheless generally teaches adding one or more columns to an analog artificial intelligence tile comprising a neural network weight layer to perform particular functionality (see e.g. section II “New Memristor-based crossbar array,” and in particular FIG. 3 therein).
It would have been obvious to one of ordinary skill in the art, having the teachings of Jain and Truong before the effective filing date of the claimed invention, to modify the non-transitory computer readable storage medium taught by Jain such that the artificial neural network is implemented on an analog artificial intelligence tile like taught by Truong, the additional training weights being provided on one or more columns of the analog artificial intelligence tile comprising the neural network weight layer. It would have been advantageous to one of ordinary skill to utilize such a combination because it can provide better area density and energy efficiency, as is taught by Truong (see e.g. the Abstract). Accordingly, Jain and Truong are considered to teach, to one of ordinary skill in the art, a non-transitory computer readable storage medium like that of claim 19.
Allowable Subject Matter
Claims 8, 14 and 20 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims. The following is a statement of reasons for the indication of allowable subject matter.
Regarding claim 8, training neural network weights by minimizing a loss between an output of the neural network and a given ground truth is generally well-known in the art. However, the prior art does not teach or suggest training additional training weights, together with the weights of the neural network, by minimizing a loss between a predicted representation and a calculated representation of one or more scalar values, the scalar values being computed across an output data vector of the artificial neural network, and wherein once trained, the additional training weights are used to predict the one or more scalar values from an incoming excitation vector, the predicted representation of the one or more scalar values being generated simultaneously with computation of the output data vector during forward inference of the neural network and used to apply one or more operations to the output data vector, as is required by claim 8 by virtue of its dependency from claims 1, 5 and 6.
Regarding claim 14, like noted above, training neural network weights by minimizing a loss between an output of the neural network and a given ground truth is generally well-known in the art. However, the prior art does not teach or suggest training additional training weights, by minimizing a loss between a predicted representation and a calculated representation of one or more scalar values, wherein the additional training weights are distinct from the trained weights of a neural network weight layer, wherein the additional training weights are stored in one or more rows or columns of an analog artificial intelligence tile, wherein the additional training weights are used for providing the predicted representation of the one or more scalar values, and wherein the predicted representation of the one or more scalar values is generated using the additional training weights simultaneously with the computation of an output data vector during forward inference of the neural network – the output data vector being computed based on the trained weights in the neural network weight layer – and the predicted representation of the one or more scalar values is used to apply one or more operations to the output data vector, while avoiding a computation used to compute an exact representation of the one or more scalar values from the output data vector, as is required by claim 14 by virtue of its dependency from claim 9.
Regarding claim 20, and like noted above, training neural network weights by minimizing a loss between an output of the neural network and a given ground truth is generally well-known in the art. However, the prior art does not teach or suggest training additional training weights, together with the weights of a neural network weight layer, by minimizing a loss between a predicted representation and a calculated representation of one or more scalar values, the scalar values being computed across an output data vector of the artificial neural network, and wherein once trained, the additional training weights are used to predict the one or more scalar values from an incoming excitation vector, the predicted representation of the one or more scalar values being generated simultaneously with computation of the output data vector during forward inference of the neural network and used to apply one or more operations to the output data vector, as is required by claim 20 by virtue of its dependency from claims 18, 17 and 15.
Response to Arguments
The Examiner acknowledges the Applicant’s amendments to claims 1, 5-7, 9, 13, 15, 17 and 18. In response to these amendments, the objections presented in the previous Office Action to claims 1-8 and 18-20 are respectfully withdrawn, as are the 35 U.S.C. § 101 rejections presented in the previous Office Action to claims 1-20.
The Applicant’s arguments concerning the prior art rejections presented in the previous Office Action have been considered, but are moot in view of the new grounds of rejection presented above, which are required in response to the Applicant’s amendments.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BLAINE T BASOM whose telephone number is (571)272-4044. The examiner can normally be reached Monday-Friday, 9:00 am - 5:30 pm, EST.
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, Matt Ell can be reached at (571)270-3264. 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.
/BTB/
10/3/2026
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights