DETAILED ACTION
Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
2. Claims 1-20 are presented for examination.
Claim Rejections - 35 USC § 101
3. 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.
3.1 Claim 20 is rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because the claim is directed to a computer readable medium storing a program; however, the specification suggested the medium could include any medium including signal bearing medium and could further be interpreted as mere program stored in the medium, thus is not statutory. It is suggested by the Examiner that the medium be amended to recite “A non-transitory computer readable recording medium” executed by a computer.
3.2 Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 2A- Prong One
The claim(s) recite(s) a methods (claims 1, 11) and computer readable recording medium (claim 20) implemented on a computer for modeling a serializer/deserializer (SerDes) model, comprising: The step of: “generating a plurality of data sets comprising noise simulation data of the SerDes model and output measurement data of an actual SerDes”; “training a machine learning model based on the plurality of data sets”; and “applying the trained machine learning model and an estimation model to a model included in the SerDes model, the estimation model being configured to provide the noise simulation data as an input to the trained machine learning model”, “modeling a SerDes model comprising a transmission model, a channel model, and a reception model on a computer” in claim 11, under the broadest reasonable interpretation fall under a mental process. Therefore, the claims are directed to an abstract idea, by use of generic computer components and thus are clearly directed to an abstract idea, as constructed.
Step 2A Prong Two
This judicial exception is not integrated into a practical application because the additional limitation such as: “a computer readable recording medium”, “a program”, “a computer”, either alone or in combination, all serve to gather and process data and do not add anything more significantly to the judicial exception, but are mere instructions to apply the exception using a generic computer component that are well known, routine, and conventional activities (see fig.13) which can be of any type, including general-purpose computer previously known in the industries. Merely adding a programmable computer to perform generic computer functions does not automatically overcome an eligibility rejection. Alice, 573 U.S. at 223-24. Furthermore, the use of a general-purpose computer to apply an otherwise ineligible algorithm does not qualify as a particular machine. See Ultramerciallnc. v. Hulu, LLC, 772F.3d 709, 716-17 (Fed. Cir. 20l4); In re TLI Commc 'ns LLC v. AV Automotive, LLC, 823 F.3d 607, 613 (Fed. Cir. 2016) (mere recitation of concrete or tangible components is not an inventive concept); Eon Corp. IP Holdings LLC v. AT&T Mobility LLC, 785; and the step of “and “manufacturing a SerDes chip corresponding to the SerDes model”(in claim 11) could amount to post-solution activity and serve to gather and process data that are well-known, routine and conventional activities and are not sufficient to amount to significantly more than the judicial exception (See further MPEP 2106.05(d)(i-iv)-f); thus are not patent eligible under 35 USC 101.
Step 2B
The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because, as previously discussed above with reference to the integration of abstract idea into a practical application, the additional elements of: “a computer readable recording medium”, “a program”, “a computer”, either alone or in combination, all serve to gather and process data and do not add anything more significantly to the judicial exception, but are mere instructions to apply the exception using a generic computer component that are well known, routine, and conventional activities (see fig.13) which can be of any type, including general-purpose computer previously known in the industries. Merely adding a programmable computer to perform generic computer functions does not automatically overcome an eligibility rejection. Alice, 573 U.S. at 223-24. Furthermore, the use of a general-purpose computer to apply an otherwise ineligible algorithm does not qualify as a particular machine. See Ultramerciallnc. v. Hulu, LLC, 772F.3d 709, 716-17 (Fed. Cir. 20l4); In re TLI Commc 'ns LLC v. AV Automotive, LLC, 823 F.3d 607, 613 (Fed. Cir. 2016) (mere recitation of concrete or tangible components is not an inventive concept); Eon Corp. IP Holdings LLC v. AT&T Mobility LLC, 785; and the step of “and “manufacturing a SerDes chip corresponding to the SerDes model” in claim 11 could amount to post-solution activity and serve to gather and process data that are well-known, routine and conventional activities and are not sufficient to amount to significantly more than the judicial exception (See further MPEP 2106.05(d)(i-iv)-f); thus are not patent eligible under 35 USC 101. Therefore, using computer components amount to no more than mere instructions to perform the abstract, and thus are not sufficient to amount to significantly more than the recited abstract, as constructed.
3.3 Dependent claims 2-10, 12-19 merely include limitations pertaining to: (claim 2), “wherein the SerDes model comprises a transmission model, a channel model, and a reception model, and the model included in the SerDes model is one of the transmission model, the channel model, and the reception model” (mental process). (claim 3); “wherein the model included in the SerDes model is the reception model” (mental process); (claims 4, 14); “wherein the actual SerDes comprises a SerDes chip, and wherein the generating and storing comprises: simulating the estimation model to obtain the noise simulation data according to characteristics of the SerDes model while changing the characteristics of the SerDes model; obtaining an output value measured from the SerDes chip as the output measurement data, the SerDes chip having characteristics that are the same as the characteristics of the SerDes model; and generating the plurality of data sets by clustering the noise simulation data and the output measurement data according to characteristics” (mental process); (claims 5, 15); “wherein the simulating of the estimation model comprises obtaining the noise simulation data by outputting a single bit response (SBR) and residual noise from an input signal through a preset algorithm, the input signal being input to the estimation model” (mental process or otherwise post-solution activity); (claims 6, 16) “wherein the preset algorithm comprises a linear pulse fitting algorithm” (mathematical concept or otherwise a mental process); (claim 7); “wherein the output measurement data comprises at least one of an eye opening size value and a bit error rate (BER)” (mental process or otherwise post-solution activity); (claims 8, 17) “wherein the training of the machine learning model comprises: processing a first data set group from among the plurality of data sets into training data; training a neural network (NN) by using the training data; evaluating an accuracy of the NN based on a second data set group from among the plurality of data sets, the second data set group excluding the first data set group; and training the NN or completing the training, according to an evaluation result of the accuracy” (mental process); (claims 9, 18) “wherein noise simulation data included in a data set of the first data set group comprises a single bit response (SBR) and a residual noise value, and the processing comprises: imaging the SBR into an image including a plurality of pixels; and generating the training data by substituting the residual noise value into each of the plurality of pixels” (mental process); (claims 10, 19) “wherein the evaluating comprises: providing noise simulation data included in the second data set group to an input layer of the NN; comparing a first value output from an output layer of the NN with a second value of output measurement data included in the second data set group; and outputting a comparison result as the evaluation result of the accuracy” (mental process and/or otherwise pre/post solution activity); (claims 12-13) “wherein the actual SerDes is an experimental SerDes. (13) “wherein the trained machine learning model and the estimation model are applied to the reception model” (mental process); all of which further amount to further mathematical concept and/or mental process similar to that already recited by the independent claims and already addressed above and thus are further not patent eligible under 35 USC 101.
Claim Rejections - 35 USC § 112
4. The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
4.1 Claims 4-6, 14-16 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention. Claims4 and 14 recites the limitation “wherein the generating and storing” in line 2 of the claim. There is insufficient antecedent basis for this limitation in the claim. The claims depend therefrom inherit the same defect.
Claims 4 and 14 further recite … according to characteristics’ however, it is unclear whether applicant meant to refer to the Serdes chip characteristic, the Serdes model characteristics or some other characteristics. Further clarification is respectfully requested in return to this office actiom.
Claim Rejections - 35 USC § 103
5. 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.
5.0 Claim(s) 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Jiao et al. (U.S. Patent No. 11,621,808), in view of Dai et al. (USPG_PUB No. 2010/0329319).
5.1 In considering claims 1, 20, Jiao et al. teaches a method implemented on a computer for modeling a serializer/deserializer (SerDes) model, the method comprising:
generating a plurality of data sets (see fig.2a (testing waveform patterns Un(t) and (Ub(t), at the output of the transmitter 202, col.6 lines 18-26, FIG. 2A depicts a communication system with an exemplary machine learning system implemented to predict transient waveforms at receiver's output. In this depicted example, a serial communication system 200 includes a transmitter 202 coupled to a receiver 204 over one or more transmission channels 206a-206n. The transmitter 202 may be part of a serializer-deserializer (SerDes) 208 and the SerDes 208 is disposed in an integrated circuit (IC) 212. The receiver 204 may also be part of a SerDes 210 and the SerDes 210 is disposed in an IC 214.) comprising output measurement data of an actual SerDes (see col.3 lines 45-50, In some embodiments, the first signal and the second signal may be extracted from on-die measurements. In some embodiments, the fourth signal may be extracted from on-die measurements. Col.8 lines 25-28, The test data pattern 255 may also include transient waveforms extracted from on-die measurements or simulations at the receiver's input.); training a machine learning model based on the plurality of data sets (see fig.2A, col.6 lines 27-32, The IC 214 also includes a machine learning system 230 configured to receive training waveforms and provide a trained machine learning model to predict waveforms at the receiver's output. Also see Col.7 lines 47-56, (29) The machine learning engine 235a is configured to perform machine learning (e.g., neural network training) using the training data patterns 245 to generate a trained machine learning model 250b. The trained machine learning model 250b may be able to predict millions of bits and generate high-correlation eye diagrams. In some embodiments, the trained machine learning model's complexity may be automatically adaptive according to the training data patterns 245.); and applying the trained machine learning model and an estimation model to a model included in the SerDes model (see col.8 lines 20-39, The machine learning engine 235a is then configured to retrieve the test data pattern 255 and apply the test data pattern 255 to the trained machine learning model 250b to generate predicted data pattern 260. The predicted data pattern 260 may include transient waveforms at the receiver's output. One or more software may be used to extract eye diagrams, and/or bit error rate (BER) bathtub characterizations from the predicted data pattern 260. col.7 lines 50-56, The trained machine learning model 250b may be able to predict millions of bits and generate high-correlation eye diagrams. In some embodiments, the trained learning model’s complexity may be automatically adaptive according to the training data patterns 245. Col.9 lines 11-12, The trained machine learning model is used to predict waveforms at the receiver's output.), the estimation model being configured to provide data as an input to the trained machine learning model (see col.8 lines 31-34, The machine learning engine 235a is then configured to retrieve the test data pattern 255 and apply the test data pattern 255 to the trained machine learning model 250b to generate predicted data pattern 260.). However, he does not specifically provide for the noise simulation data.
Dai et al. provide for a simulation module 248 for simulating and quantifying noise (see fig.2, 6, para [0083] All models of passive and active components of the receiver may be extracted from a circuit simulation tool such as SPICE (Simulation Program with Integrated Circuit Emphasis) under worst case PVT. From the SPICE simulation, a frequency behavior model is extracted using AC (alternating current) analysis and is captured in MATLAB. [0100], [0100] Take clock noise source for example. The reference clock is often generated by an off-chip crystal oscillator whose frequency can vary from 10 MHz to over 200 MHz. The SerDes high speed clock (multi-GHz) is then generated with a phase locked loop (PLL) circuit with respect to the reference clock. The PLL employs either voltage controlled ring oscillator (VCO) or LC oscillator (LCO). All forms of oscillators have phase noises which contribute to timing jitter. Phase noise is expressed as the ratio of signal power to noise power measured in 1 Hz bandwidth at a given offset from the desired signal. FIG. 13 shows a typical phase noise profile. A detailed study can be found in "Oscillator phase noise: a tutorial" by T. Lee and A. Hijimiri, IEEE Journal of Solid-State Circuits, Vol. 35, No. 3, pp. 326-336, March 2000, the disclosure of which is incorporated by reference herein. Once the phase noise is obtained in the frequency domain, it can be converted into timing jitter that will be used later to construct the internal eye mask. A mathematical relationship between phase noise and timing jitter (the conversion) is described in Maxim application notes 3359 "Clock (CLK) jitter and phase noise conversion", Sep. 23, 2004, the disclosure of which is incorporated by reference herein. [0106], For example, the DFE model includes imperfection with quantization noise in tap weights.).
Jiao et al. and Dai et al. are analogous art because they are from the same field of endeavor and that the model analyzes by Dai et al. is similar to that of Jiao et al. Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.2 With regards to claim 2, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the SerDes model comprises a transmission model (see Dai et al. fig2 (transmitter 210)), a channel model (see Dai et al. fig2 (link channel 220)), and a reception model (see Dai et al. fig2 (receiver 230)), and the model included in the SerDes model 208 is one of the transmission model (see Jiao et al. fig.2A (202), col.6 lines 22-24, The transmitter 202 may be part of a serializer-deserializer (SerDes) 208 ), the channel model (206a-n), and the reception model (204)). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.3 Regarding claim 3, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the model included in the SerDes model is the reception model (see Jiao et al. receiver 204 included in SerDes model 210, col.6 lines 25-26, The receiver 204 may also be part of a SerDes 210). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.4 As per claims 4, 14, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the actual SerDes comprises a SerDes chip (see Jiao et al. fig.2A, SerDes disposed within IC chip 214), and wherein the generating and storing comprises: simulating the estimation model to obtain the noise simulation data according to characteristics of the SerDes model while changing the characteristics of the SerDes model (see Dai et al. fig.2, 6, para [0083] All models of passive and active components of the receiver may be extracted from a circuit simulation tool such as SPICE (Simulation Program with Integrated Circuit Emphasis) under worst case PVT. From the SPICE simulation, a frequency behavior model is extracted using AC (alternating current) analysis and is captured in MATLAB. [0100], Take clock noise source for example. The reference clock is often generated by an off-chip crystal oscillator whose frequency can vary from 10 MHz to over 200 MHz. The SerDes high speed clock (multi-GHz) is then generated with a phase locked loop (PLL) circuit with respect to the reference clock. Phase noise is expressed as the ratio of signal power to noise power measured in 1 Hz bandwidth at a given offset from the desired signal. FIG. 13 shows a typical phase noise profile. Once the phase noise is obtained in the frequency domain, it can be converted into timing jitter that will be used later to construct the internal eye mask. A mathematical relationship between phase noise and timing jitter (the conversion) is described in Maxim application notes 3359 "Clock (CLK) jitter and phase noise conversion", [0106], For example, the DFE model includes imperfection with quantization noise in tap weights.); obtaining an output value measured from the SerDes chip as the output measurement data, the SerDes chip having characteristics that are the same as the characteristics of the SerDes model (see Dai et al. fig.1-3, 5-6, Jiao et al. fig.2A, col.6 lines 18-32, Fig.2A depicts a communication system with an exemplary machine learning system implemented to predict transient waveforms at receiver's output. In this depicted example, a serial communication system 200 includes a transmitter 202 coupled to a receiver 204 over one or more transmission channels 206a-206n. The transmitter 202 may be part of a serializer-deserializer (SerDes) 208 and the SerDes 208 is disposed in an integrated circuit (IC) 212. The receiver 204 may also be part of a SerDes 210 and the SerDes 210 is disposed in an IC 214. The IC 214 also includes a machine learning system 230 configured to receive training waveforms and provide a trained machine learning model to predict waveforms at the receiver's output. Bathtub characterizations and eye diagrams may be then extracted from the predicted output waveforms.); and generating the plurality of data sets by clustering the noise simulation data and the output measurement data according to characteristics (see Jiao et al. 1-3, 5-6, para [0083] All models of passive and active components of the receiver may be extracted from a circuit simulation tool such as SPICE (Simulation Program with Integrated Circuit Emphasis) under worst case PVT. From the SPICE simulation, a frequency behavior model is extracted using AC (alternating current) analysis and is captured in MATLAB. [0100], Take clock noise source for example. The reference clock is often generated by an off-chip crystal oscillator whose frequency can vary from 10 MHz to over 200 MHz. The SerDes high speed clock (multi-GHz) is then generated with a phase locked loop (PLL) circuit with respect to the reference clock. Phase noise is expressed as the ratio of signal power to noise power measured in 1 Hz bandwidth at a given offset from the desired signal. FIG. 13 shows a typical phase noise profile. Once the phase noise is obtained in the frequency domain, it can be converted into timing jitter that will be used later to construct the internal eye mask. A mathematical relationship between phase noise and timing jitter (the conversion) is described in Maxim application notes 3359 "Clock (CLK) jitter and phase noise conversion", Sep. 23, 2004, the disclosure of which is incorporated by reference herein. [0106], For example, the DFE model includes imperfection with quantization noise in tap weights. Further Dai et al. fig.2A). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.5 Regarding claims 5, 15, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the simulating of the estimation model comprises obtaining the noise simulation data by outputting a single bit response (SBR) (see Dai et al. output response showing by 608)) and residual noise from an input signal through a preset algorithm, the input signal being input to the estimation model (see Dai et al. fig. 6, para [0094] The quality of serial data communication is often measured with a bit error ratio (BER) test. Bit error ratio (BER) is defined as the percentage of the received bits in error compared to the total number of bits received over a given period of time. Jiao et al. col.9 lines 55-67, FIG. 2B. structure is given by: ŷ(t)=f(y(t−1),y(t−2), .y(t−n),u(t−d),u(t−d−m),ε(t−1), . . . ,ε(t−k)) where ŷ(t) is the predicted output at time t, u is the input, and ε is the residual between the observed value y(t) and the predicted output ŷ(t). d is the delay between the input and output, n is the order of the output, m is the order of the input, and k is the order of the residual. Col.7 lines 39-46, The training input waveform u.sub.a(t) and the generated waveform y.sub.a(t) may be transient waveforms extracted from on-die measurements or simulations at the receiver's input and output, respectively. The difference between the input waveform u.sub.a(t) and the generated waveform y.sub.a(t) contains all the receiver's analog front end and equalization information. The training data patterns 245 may be stored in memory 240.). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.6 As per claims 6, 16, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the preset algorithm comprises a linear pulse fitting algorithm (see Dai et al. fig.6, 13-14, para [0052] In particular, FIG. 2 shows a visualization of a SerDes device model 200 including a transmitter portion (serializer) 210 with data source and FIR (finite impulse response) control sections, a link channel portion 220 with load parameters, a receiver portion (deserializer) 230 with VGA (variable gain amplifier), LE1 and LE2 (linear equalizers), DFE (decision feedback equalizer), and eye latch array (e.g., a "bang-bang" type utilizing a n-UI sampling arrangement, where UI denotes unit interval and n may be 2, 3, 4, 5, . . . etc.). Also shown are a rate control portion 240, a crosstalk control portion 242, an evaluation control portion 244, a TX adaptation portion 246, and a simulation portion 248.). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.7 With regards to claim 7, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the output measurement data comprises at least one of an eye opening size value (see Dai et al. para [0041] For signal integrity (SI) analysis of high speed differential serial links such as, for example, links that exist between transmitters and receivers of nodes 102 and 104 in FIG. 1, an eye diagram is often employed as a preferred/intuitive system optimization tool. As is well known with respect to eye diagrams, "eye height" is the eye opening in the amplitude domain and "eye width" relates to the signal timing and jitter. [0095], Qualitative assessment of the system performance can be obtained, because much parametric information about the channel links can be extracted out from signal rise/fall time, overshoot or ringing, width/height of eye opening, etc.) and a bit error rate (BER) (see Dai et al. para [0094] The quality of serial data communication is often measured with a bit error ratio (BER) test. Bit error ratio (BER) is defined as the percentage of the received bits in error compared to the total number of bits received over a given period of time. It is sometimes called bit error rate.). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.8 As per claims 8, 17, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the training of the machine learning model comprises: processing a first data set group from among the plurality of data sets into training data (see Jiao et al. col.6 lines 9-36, The adaptive-ordered machine learning model may be then trained (by the training data patterns 245) to be a specific machine learning model 250b (e.g., with known model variables, nonlinear degree, etc.). The trained machine learning model 250b may then be stored in the memory 240 and be used by the machine learning engine 235a in a second phase. An exemplary simulated PCC score is described in further detail with reference to FIG. 3. An exemplary method to generate the trained machine learning model 250b is described in detail with reference to FIG. 4A. When the trained machine learning model 250b is ready, in a second phase, the machine learning engine 235a may be then configured to perform operations to implement the trained machine learning model 250b to predict data patterns (e.g., transient waveform ŷ.sub.b(t)) at the receiver's output via different channels and with different test data patterns. The test data pattern 255 includes one or more test waveforms (e.g., waveform Ub(t)˜waveform u.sub.n(t)). The test data pattern 255 may also include transient waveforms extracted from on-die measurements or simulations at the receiver's input. In this depicted example, the test data pattern 255 is also stored in memory 240. The machine learning engine 235a is then configured to retrieve the test data pattern 255 and apply the test data pattern 255 to the trained machine learning model 250b to generate predicted data pattern 260. The predicted data pattern 260 may include transient waveforms at the receiver's output.); training a neural network (NN) by using the training data (see fig.2, 4A(415), training of the neural network based in the training data sets, col.7 lines 47-56, the operations includes (a) retrieving, by the processing engine, a set of training data patterns from the data store, (b) generating, by the processing engine, a model order of the neural network model in response to a Pearson Correlation Coefficient simulation result, (c) applying the model order to an initial neural network model and training the model in response to the set of training data patterns to obtain parameters used in the neural network model, the trained neural network model is used to predict a data pattern at the receiver's output in response to a test data pattern to be received at the receiver's input, (d) retrieving, by the processing engine, a test data pattern from the data store, and, (e) processing, by the trained neural network model, the retrieved test data pattern to generate a corresponding predicted data pattern.); evaluating an accuracy of the NN based on a second data set group from among the plurality of data sets, the second data set group excluding the first data set group (see Jiao et al. fig.5, col.11 line 34-col.12 line 7, The adaptive-ordered NNARMAX model shows high capabilities to track nonlinear behavior of the CTLE and high-precision accuracies over testing channels. All the test cases are from different channels and different data patterns. The transient waveforms prediction accuracies are above 99%. In the tests. the adaptive-ordered NNARMAX model also provides high-correlation predictions for all the cases, as shown in FIG. 5. All the test cases are from different channels and different data patterns. The eye height and eye width prediction accuracies are also above 96%. The adaptive-ordered NNARMAX model also provides a high-correlation eye shape prediction. For the simulation speed, the adaptive-ordered NNARMAX model is faster than other machine learning models.); and training the NN or completing the training, according to an evaluation result of the accuracy (see Jiao et al. fig.2, 4A(415), training of the neural network based in the training data sets, col.7 lines 47-56, The machine learning engine 235a is configured to perform machine learning (e.g., neural network training) using the training data patterns 245 to generate a trained machine learning model 250b. The trained machine learning model 250b may be able to predict millions of bits and generate high-correlation eye diagrams. In some embodiments, the trained machine learning model’s complexity may be automatically adaptive according to the training data patterns 245. the machine learning model may be trained only once, predict transient waveforms, eye diagram, and bathtub curve of different data patterns over different channels may be rapidly precited while preserving substantial accuracy). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.9 Regarding claims 9, 18, the combined teachings of Jiao et al. and Dai et al. teaches that wherein noise simulation data included in a data set of the first data set group comprises a single bit response (SBR) and a residual noise value (see Dai et al. fig. 6, output response showing by 608), para [0094] The quality of serial data communication is often measured with a bit error ratio (BER) test. Bit error ratio (BER) is defined as the percentage of the received bits in error compared to the total number of bits received over a given period of time. Jiao et al. col.9 lines 55-67, FIG. 2B. structure is given by: ŷ(t)=f(y(t−1),y(t−2), .y(t−n),u(t−d),u(t−d−m),ε(t−1), . . . ,ε(t−k)) where ŷ(t) is the predicted output at time t, u is the input, and ε is the residual between the observed value y(t) and the predicted output ŷ(t). d is the delay between the input and output, n is the order of the output, m is the order of the input, and k is the order of the residual. Col.7 lines 39-46, The training input waveform u.sub.a(t) and the generated waveform y.sub.a(t) may be transient waveforms extracted from on-die measurements or simulations at the receiver's input and output, respectively. The difference between the input waveform u.sub.a(t) and the generated waveform y.sub.a(t) contains all the receiver's analog front end and equalization information. The training data patterns 245 may be stored in memory 240.), and the processing comprises: imaging the SBR into an image including a plurality of pixels (see Dai fig.6, para [0075] The comparison with IEM yields a pass/fail score for a given channel and for a given transmitter setting. For each given channel, the results from all test points generate a passing zone. Passing zones from all channels of the same group are then compared. The settings that satisfy all channels, defines optimal transmitter settings for this channel group. The passing zone is the common space for all selected channels. FIG. 7 shows an illustration on a two-dimensional design space. For multi-dimensional design space, visualization can be a difficult task and the methodology can obtain the passing zone with the aid of a computer program, such as the following code in the Python programming language:Jiao et al. fig.2, 4-6); and generating the training data by substituting the residual noise value into each of the plurality of pixels (see Jiao et al. fig.2, 4-6, col.7 line 47-col.8 line 19, The machine learning engine 235a is configured to perform machine learning (e.g., neural network training) using the training data patterns 245 to generate a trained machine learning model 250b. The trained machine learning model 250b may be able to predict millions of bits and generate high-correlation eye diagrams. In some embodiments, the trained machine learning machine learning model’s complexity may be automatically adaptive according to the training data patterns 245. Due to channel loss, the current received bit may be impacted by the ISI from the previous bits.
(30) In various embodiments, the machine learning engine 235a may be configured to perform operations to implement the Pearson Correlation Coefficient (PCC) algorithm to analyze the relationship among the current output, previous inputs, and previous outputs, and provide model order suggestions. A user may set a first PCC score to select how many previous inputs to be used in the neural network model. A user may also select a second PCC score to select how may previous outputs to be used in the neural network model. In some embodiments, the first PCC score may equal the second PCC score. In various embodiments, the machine learning engine 235a may be then configured to perform operations to implement an initial machine learning model 250a. The initial machine learning model 250a may then use the model orders generated by PCC analysis to analyze the previous inputs and previous outputs to identify the different effects of previous inputs and then self-select how many previous inputs and previous outputs may be used to predict the current output according to PCC scores. The adaptive-ordered machine learning model may be then trained (by the training data patterns 245) to be a specific machine learning model 250b (e.g., with known model variables, nonlinear degree, etc.). The trained machine learning model 250b may then be stored in the memory 240 and be used by the machine learning engine 235a in a second phase. An exemplary simulated PCC score is described in further detail with reference to FIG. 3. An exemplary method to generate the trained machine learning model 250b is described in detail with reference to FIG. 4A.
(32) When the trained machine learning model 250b is ready, in a second phase, the machine learning engine 235a may be then configured to perform operations to implement the trained machine learning model 250b to predict data patterns (e.g., transient waveform ŷ.sub.b(t)) at the receiver's output via different channels and with different test data patterns. The test data pattern 255 includes one or more test waveforms (e.g., waveform Ub(t)˜waveform u.sub.n(t)). The test data pattern 255 may also include transient waveforms extracted from on-die measurements or simulations at the receiver's input. In this depicted example, the test data pattern 255 is also stored in memory 240. The machine learning engine 235a is then configured to retrieve the test data pattern 255 and apply the test data pattern 255 to the trained machine learning model 250b to generate predicted data pattern 260). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.10 As per claims 10, 19, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the evaluating comprises: providing noise simulation data included in the second data set group to an input layer of the NN (see Dai et al. fig.13-16, para [0099-0100); further see Jiao et al. fig.2, 4-6); comparing a first value output from an output layer of the NN with a second value of output measurement data included in the second data set group (see Dai et al. para [0089] When the cost function is minimized, either at the end of the adaptation state machine, or by sweeping the receiver design space, the resulting signal waveform is saved and the corresponding post-processing eye diagram is compared against a pre-defined internal eye mask. The internal eye mask ensures that the expected or accepted performance level will be met for a specific application. The generation of IEM is detailed below in section III (Generation of Internal Eye Mask). Jiao et al. col.11 lines 42-54, FIG. 6 shows side by side comparison of eye diagrams from the actual waveforms and the predicted waveforms. In this case, the channel loss is higher than CTLE relative gain, and the predicted eye diagram shows under-equation); and outputting a comparison result as the evaluation result of the accuracy (see Jiao et al. fig.5-7, 20output of the comparison of fig.7, fig.5 further shows accuracy output display). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.11 With regards to claim 11, Jiao et al. teaches a method for manufacturing a serializer/deserializer (SerDes), the method comprising:
modeling a SerDes model comprising a transmission model (see fig.2A (202)), a channel model (see fig.2A(206a-n)), and a reception model (see fig.2A (204)) on a computer; and manufacturing a SerDes chip corresponding to the SerDes model (see fig.2,4-5, col.5-6 lines 18-32, FIG. 2A depicts a communication system with an exemplary machine learning system implemented to predict transient waveforms at receiver's output. In this depicted example, a serial communication system 200 includes a transmitter 202 coupled to a receiver 204 over one or more transmission channels 206a-206n. The transmitter 202 may be part of a serializer-deserializer (SerDes) 208 and the SerDes 208 is disposed in an integrated circuit (IC) 212. The receiver 204 may also be part of a SerDes 210 and the SerDes 210 is disposed in an IC 214. The IC 214 also includes a machine learning system 230 configured to receive training waveforms and provide a trained machine learning model to predict waveforms at the receiver's output. Bathtub characterizations and eye diagrams may be then extracted from the predicted output waveforms. The actual I/O bond pads connected, for example, to the I/O logic element 115, may be manufactured using metal layered above the various illustrated logic blocks), wherein the modeling of the SerDes model comprises: generating a plurality of data sets (see fig.2A-B (testing waveform patterns Un(t) and (Ub(t), at the output of the transmitter 202, col.6 lines 18-26, FIG. 2A depicts a communication system with an exemplary machine learning system implemented to predict transient waveforms at receiver's output. In this depicted example, a serial communication system 200 includes a transmitter 202 coupled to a receiver 204 over one or more transmission channels 206a-206n. The transmitter 202 may be part of a serializer-deserializer (SerDes) 208 and the SerDes 208 is disposed in an integrated circuit (IC) 212. The receiver 204 may also be part of a SerDes 210 and the SerDes 210 is disposed in an IC 214.) comprising output measurement data of an actual SerDes (see col.3 lines 45-50, In some embodiments, the first signal and the second signal may be extracted from on-die measurements. In some embodiments, the fourth signal may be extracted from on-die measurements. Col.8 lines 25-28, The test data pattern 255 may also include transient waveforms extracted from on-die measurements or simulations at the receiver's input.); training a machine learning model based on the plurality of data sets (see fig.2A-B, col.6 lines 27-32, The IC 214 also includes a machine learning system 230 configured to receive training waveforms and provide a trained machine learning model to predict waveforms at the receiver's output. Also see Col.7 lines 47-56, (29) The machine learning engine 235a is configured to perform machine learning (e.g., neural network training) using the training data patterns 245 to generate a trained machine learning model 250b. The trained machine learning model 250b may be able to predict millions of bits and generate high-correlation eye diagrams. In some embodiments, the trained machine learning model's complexity may be automatically adaptive according to the training data patterns 245.); and applying the trained machine learning model and an estimation model to one of the transmission model (202), the channel model 9206a-n), and the reception model (204), (see col.8 lines 20-39, The machine learning engine 235a is then configured to retrieve the test data pattern 255 and apply the test data pattern 255 to trained machine learning model 250a to generate predicted data pattern 260. The predicted data pattern 260 may include transient waveforms at the receiver's output. One or more software may be used to extract eye diagrams, and/or bit error rate (BER) bathtub characterizations from the predicted data pattern 260. col.7 lines 50-56, The trained machine learning model 250b may be able to predict millions of bits and generate high-correlation eye diagrams. the trained learning model’s complexity may be automatically adaptive according to the training data patterns 245. Col.9 lines 11-12, The trained machine learning model is used to predict waveforms at the receiver's output.), (see col.8 lines 31-34, The machine learning engine 235a is then configured to retrieve the test data pattern 255 and apply the test data pattern 255 to the trained machine learning model 250b to generate predicted data pattern 260.). However, he does not specifically provide for the noise simulation.
Dai et al. provide for a simulation module 248 for simulating and quantitating noise (see fig.2, 6, para [0083] All models of passive and active components of the receiver may be extracted from a circuit simulation tool such as SPICE (Simulation Program with Integrated Circuit Emphasis) under worst case PVT. From the SPICE simulation, a frequency behavior model is extracted using AC (alternating current) analysis and is captured in MATLAB. [0100], [0100] Take clock noise source for example. The reference clock is often generated by an off-chip crystal oscillator whose frequency can vary from 10 MHz to over 200 MHz. The SerDes high speed clock (multi-GHz) is then generated with a phase locked loop (PLL) circuit with respect to the reference clock. The PLL employs either voltage controlled ring oscillator (VCO) or LC oscillator (LCO). All forms of oscillators have phase noises which contribute to timing jitter. Phase noise is expressed as the ratio of signal power to noise power measured in 1 Hz bandwidth at a given offset from the desired signal. FIG. 13 shows a typical phase noise profile. A detailed study can be found in "Oscillator phase noise: a tutorial" by T. Lee and A. Hijimiri, IEEE Journal of Solid-State Circuits, Vol. 35, No. 3, pp. 326-336, March 2000, the disclosure of which is incorporated by reference herein. Once the phase noise is obtained in the frequency domain, it can be converted into timing jitter that will be used later to construct the internal eye mask. A mathematical relationship between phase noise and timing jitter (the conversion) is described in Maxim application notes 3359 "Clock (CLK) jitter and phase noise conversion", Sep. 23, 2004, the disclosure of which is incorporated by reference herein. [0106], For example, the DFE model includes imperfection with quantization noise in tap weights). Those settings are documented and provided to the factory (e.g., SerDes manufacturing facility) to be pre-set based on equipment applications and customer requests [0091].
Jiao et al. and Dai et al. are analogous art because they are from the same field of endeavor and that the model analyzes by Dai et al. is similar to that of Jiao et al. Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.12 As per claim 12, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the actual SerDes is an experimental SerDes (see Jiao et al. receiver 204 included in SerDes model 210, col.6 lines 25-26, The receiver 204 may also be part of a SerDes 210). Col.10 lines 28-52, FIG. 4A depicts a flow chart of an exemplary method to generate a machine learning model. A method 400A of generating a machine learning model includes, at 405, determining training data patterns (e.g., waveform u.sub.a(t) at SerDes receiver's input and waveform y.sub.a(t) at SerDes receiver's output). The waveforms u.sub.a(t) and y.sub.a(t) may be transient waveforms extracted from on-die measurements or simulations at the receiver's input and output,). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
5.13 Regarding claim 13, the combined teachings of Jiao et al. and Dai et al. teaches that wherein the trained machine learning model and the estimation model are applied to the reception model (see Jiao et al. fig.2A-B, 4, col.2 line 65-col.3 line 6, (c) applying the model order to the initial neural network model and training the model in response to the set of training data patterns to obtain parameters used in the neural network model, the trained neural network model is used to predict data pattern at the receiver's output, (d) providing a test data pattern, and, (e) processing, by the trained neural network model, the test data pattern to obtain a corresponding predicted data pattern at the receiver's output. Col.6 lines 18-32, FIG. 2A depicts a communication system with an exemplary machine learning system implemented to predict transient waveforms at receiver's output. In this depicted example, a serial communication system 200 includes a transmitter 202 coupled to a receiver 204 over one or more transmission channels 206a-206n. The transmitter 202 may be part of a serializer-deserializer (SerDes) 208 and the SerDes 208 is disposed in an integrated circuit (IC) 212. The receiver 204 may also be part of a SerDes 210 and the SerDes 210 is disposed in an IC 214. The IC 214 also includes a machine learning system 230 configured to receive training waveforms and provide a trained machine learning model to predict waveforms at the receiver's output. Bathtub characterizations and eye diagrams may be then extracted from the predicted output waveforms.). Therefore, it would have been obvious to a person of skilled in the art at the time of filing of the applicant’s invention to combine the method of Dai et al. with that of Jiao et al. because Dai et al. provide for improved system optimization (para [0028]).
Conclusion
6. The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
6.1 Yuan et al. (USPG_PUB No. 2010/0088676) teaches a method of performing a three-way merge includes receiving first, second, and third versions of a structured document containing first, second, and third pluralities of elements respectively; deserializing the first, second, and third versions to generate first, second, and third tree-structured data models respectively representing the first, second, and third versions.
6.2 Jiao et al. (U.S. Patent No. 11,423,303) teaches an apparatus and associated methods relate to providing a machine learning methodology that uses the machine learning's own failure experiences to optimize future solution search.
7. Claims 1-20 are rejected and this action is non-final. Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANDRE PIERRE-LOUIS whose telephone number is (571)272-8636. The examiner can normally be reached M-F 9:00 AM-5:00 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, EMERSON C PUENTE can be reached at 571-272-3652. 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.
/ANDRE PIERRE LOUIS/Primary Patent Examiner, Art Unit 2187 April 15, 2026
Prosecution Insights