`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
` UNITED STATES PATENT AND TRADEMARK OFFICE
`
`
`
`
`
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`
`
`
`
`
`APPLE INC.,
`Petitioner,
`
`v.
`
`Zentian Limited
`Patent Owner.
`____________________
`
`Case IPR2023-00037
`Patent No. 10,971,140
`____________________
`
`
`
`PATENT OWNER’S SUR-REPLY
`
`
`
`
`
`
`
`
`
`
`Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`
`TABLE OF CONTENTS
`
`
`
`I.
`
`II.
`
`4
`
`1
`4
`8
`15
`18
`19
`19
`21
`
`21
`
`23
`
`
`
`A.
`
`B.
`
`Petitioner has failed to show obviousness as to limitations 1(d) and 1(e)
`The Petition’s combination lacks a reasonable expectation of success
`Patent Owner’s Response presented detailed evidence directed to the
`Petition’s combination and the abilities of the Petition’s POSA
`Petitioner’s Reply fails to rebut Patent Owner’s evidence, and instead
`relies on incorrect premises
`7
`Petitioner’s false logic 1: “a POSITA is assumed to have
`requisite knowledge” to enable the Petition’s combination
`Petitioner’s false logic 2: A POSA would have known how to
`achieve Petitioner’s combination through general knowledge of
`parallel processing architectures and various speech recognition
`systems
`Petitioner’s false logic 3: “Zentian does not argue that the
`combination is inoperable, [or] could not be done. . . .”
`Petitioner’s false logic 4: “Zentian’s alleged complications . . .
`would occur regardless of the type of information being
`processed.” Paper 22 at 14.
`Petitioner’s false logic 5: “Chen envisions its processors and
`memories are directly and adjacently accessible.” Paper 22 at
`16.
`
`1.
`2.
`3.
`4.
`5.
`III. The evidence proves no motivation to combine
`
`
`A.
`B.
`
`Patent Owner’s Response disproved Petitioner’s motivations
`Petitioner’s attempts to rehabilitate its motivations to combine are
`ineffective
`
`
`
`
`
`
`
`
`
`
`
`
`
`- i -
`
`
`
`Exhibit No.
`
`2017
`
`2018
`
`2019
`
`2020
`
`2021
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`
`EXHIBIT LIST
`
`
`Description
`
`Deposition of Christopher Schmandt dated September 6, 2023
`
`Intentionally left blank
`
`Intentionally left blank
`
`Declaration of David Anderson, Ph.D
`
`Binu K. Mathew et al., A Gaussian Probability Accelerator for
`SPHINX 3, (“Mathew I”)
`
`
`
`
`
`
`
`
`
`- ii -
`
`
`
`
`
`I.
`
`Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`Petitioner has failed to show obviousness as to limitations 1(d) and 1(e)
`Limitations 1(d) and 1(e) require that the “speech recognition circuit is
`
`
`
`configured to generate an initial score,” 1(d), which is “used to determine whether
`
`to continue processing to determine a final score via processing a larger amount of
`
`data than that was processed to generate the initial score.” 1(e).
`
`
`
`To demonstrate obviousness, Mr. Schmandt and Petitioner were required to
`
`explain how Jiang’s tree search 74 modified in view of Chen, i.e., the combination
`
`as a whole, would allegedly operate to meet limitations 1(d) and 1(e). ActiveVideo
`
`Networks, Inc. v. Verizon, 694 F.3d 1312, 1327 (Fed. Cir. 2012); Ex. 2017 at
`
`46:12-22 (Schmandt admitting his theory requires modifying Jiang’s tree search
`
`engine 74 to use Chen’s clustered processor architecture); Ex. 2017 at 68:17-25,
`
`83:10-14 (Schmandt admitting his theory relies on tree search engine 74 for
`
`limitations 1(d) and 1(e)).
`
`
`
`Mr. Schmandt admitted, however, that his declaration’s discussions of
`
`limitations 1(d) and 1(e) contains no such explanation. Ex. 2017 at 79:11-23, 81:8-
`
`22, 83:15-18. The Petition fails on that basis alone. ActiveVideo, 694 F.3d at 1327.
`
`
`
`Moreover, the only theory Mr. Schmandt offered at his deposition was that
`
`one processor in one cluster in Chen would have been used to generate the “initial
`
`
`
`
`
`
`
`
`
`- 1 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`score” of limitation 1(d) and to determine whether to continue processing as
`
`
`
`recited in limitation 1(e). Ex. 2017 at 102:5-103:3.
`
`
`
`
`
`As Patent Owner’s Response demonstrated, however, that theory would fail
`
`because it would require one processor in one of Chen’s clusters to receive
`
`necessary information from all four memories within Chen’s Fig. 4 embodiment,
`
`Paper 19 at 32-34, whereas Mr. Schmandt’s declaration admits that the four
`
`memories are not all “interconnected,” and that no one processor in one cluster can
`
`directly or adjacently access all of the data stored across the four memories. Paper
`
`19 at 34-35; Ex. 1003 ¶ 95; Ex. 2020 ¶¶ 50-51; Ex. 1005, 9:10-39.
`
`
`
`
`
`
`
`
`
`- 2 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`Accordingly, Patent Owner demonstrated that Petitioner has not proven that
`
`
`
`
`
`its combination would meet limitations 1(d) and 1(e).
`
`
`
`Petitioner’s Reply arguments only further prove that reality. Petitioner
`
`argues that “Jiang’s algorithm of generating an initial score and using the initial
`
`score to determine whether to continue processing (the functionality of claims
`
`1(d)-1(e)) would be replicated across Chen’s processors.” Paper 22 at 2 (emphasis
`
`Petitioner’s). That theory, however, does not solve Petitioner’s problems; it makes
`
`them worse.
`
`First, Petitioner’s Reply theory requires every processor in Chen’s four
`
`clusters to perform limitations 1(d) and 1(e), and is thus not Mr. Schmandt’s
`
`deposition theory, which required only one processor in one cluster to do so.
`
`Petitioner thus abandons its own expert’s testimony, discrediting Mr. Schmandt.
`
`Second, Petitioner’s Reply theory would require every processor in Chen’s
`
`four clusters to calculate the same initial score and make the same determination
`
`whether to continue processing towards a final score—a design proposal that
`
`makes no sense and for which Petitioner has provided neither a reasonable
`
`expectation of success nor a motivation.
`
`Third, Mr. Schmandt’s deposition testimony explained that there would be
`
`no advantage from using all of Chen’s processors to calculate the initial score and
`
`
`
`
`
`
`
`
`- 3 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`determine whether to continue processing because “[a]t that point, we have
`
`
`
`finished our parallel processing operation and we’re back to serial operation.” Ex.
`
`2017 at 102:5-103:3.
`
`Fourth, Petitioner’s Reply theory would still require each processor to
`
`obtain necessary information stored across all four memories 104(a)-(d) in Chen,
`
`Ex. 2020 ¶ 51, whereas none of Chen’s processors can directly or adjacently access
`
`all cluster memories. Ex. 1003 ¶ 95; Ex. 2020 ¶¶ 50-51; Ex. 1005, 9:10-39. Indeed,
`
`Petitioner’s Reply admits that “[o]nce the cumulative probability score is
`
`calculated, inter-cluster processing of the results for each tree search would be
`
`communicated[.]” Paper 22 at 16. Petitioner’s Reply theory is thus no more
`
`feasible than Mr. Schmandt’s deposition theory.
`
`Fifth, Petitioner’s Reply theory was not set forth in the Petition. Petitioner’s
`
`Reply citations are to aspects of the Petition discussing other limitations, not
`
`limitations 1(d) and 1(e). Paper 22 at 2 (citing Pet. at 1, 17-18, 24, 30, Ex. 1003 ¶¶
`
`74, 76-77, 83). Petitioner’s Reply theory is thus foreclosed.
`
`Accordingly, Petitioner has not proven unpatentability.
`
`II. The Petition’s combination lacks a reasonable expectation of success
`A.
`Patent Owner’s Response presented detailed evidence directed to
`the Petition’s combination and the abilities of the Petition’s POSA
`
`
`
`
`
`
`
`
`
`
`
`- 4 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`Patent Owner’s Response demonstrated through extensive evidence that the
`
`
`
`Petition’s combination lacks a reasonable expectation of success. Paper 19 at 7-16.
`
`Each aspect of Patent Owner’s argument was directed to problems and deficiencies
`
`arising out of the Petition’s specific combination, not the ’140 Patent. Id.
`
`
`
`In particular, Patent Owner’s Response proved:
`
`• The Petition’s combination requires performing the speech recognition
`
`techniques of Jiang on the clustered processor and memory architecture of
`
`Chen, Ex. 1003 ¶ 68;
`
`• Jiang does not enable performing its speech recognition on Chen’s clustered
`
`processors, nor does Chen enable using its hardware to perform Jiang’s
`
`speech recognition, Ex. 2017 at 86:9-17, Ex. 2020 ¶ 24;
`
`• Mr. Schmandt’s declaration did not provide any underlying details to
`
`support his conclusion as to the POSA’s capability, Ex. 2017 at 48:13-21;
`
`• Mr. Schmandt has never supervised anyone involved in mapping a speech
`
`recognition model to a clustered processor and memory architecture like
`
`Chen’s, Ex. 2017 at 145:20-24, nor has he built the processor to memory
`
`architecture of any speech recognition system identified in his background
`
`experience, Id. at 34:24-35:5, 32:3-9, 106:10-23.
`
`
`
`
`
`
`
`
`
`- 5 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`• Implementing the Petition’s combination of Jiang and Chen would have
`
`
`
`been highly complex and well beyond the skill level of the POSA prior to
`
`the ’140 Patent, Ex. 2020 ¶¶ 17-19, 27 (page 16), because:
`
`o porting Jiang’s speech recognition techniques to Chen’s clustered
`
`parallel processors and memories would have required coordinating
`
`multiple caches, avoiding memory conflicts, controlling task sharing
`
`in an efficient manner, resolving synchronous bottlenecks, addressing
`
`communication bandwidth and latency issues among the various
`
`hardware, and developing a messaging strategy to coordinate
`
`information sharing between and within the clusters, Ex. 2020 ¶ 27,
`
`whereas;
`
`o the ordinary artisan in the field of speech recognition was a specialist
`
`in digital signal processing and speech recognition, not parallel
`
`processing architectures or high performance computing, Ex. 2020 ¶
`
`28, and further because;
`
`o Jiang teaches the use of the Viterbi decoding algorithm, which
`
`requires extensive communication between computational
`
`components of a system, whereas Chen’s cluster memories are not all
`
`directly or adjacently accessible to the processors and memories in
`
`
`- 6 -
`
`
`
`
`
`
`
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`each other cluster, rendering it unlikely that Jiang’s Viterbi algorithm
`
`
`
`could successfully be run on Chen’s clustered processors in a practical
`
`system. Ex. 2020 ¶ 29;
`
`o It was known in the art that communication between parallel
`
`processing nodes can determine the feasibility of a parallel processing
`
`architecture for a given problem, and that a memory model in which
`
`memory is connected only within a cluster (as in Chen) may be
`
`appropriate “for applications that require little or no communication,”
`
`whereas Jiang’s speech recognition stage requires significant
`
`communication, and is thus poorly suited for Chen. Ex. 2020 ¶ 30;
`
`o Even Dr. Anderson, who is a person of extraordinary skill, along with
`
`a team of engineers and a DSP expert from Texas Instruments, could
`
`not successfully transition existing signal processing software to a
`
`parallel processing architecture after six months. Ex. 2020 ¶ 31.
`
`B.
`
`Petitioner’s Reply fails to rebut Patent Owner’s evidence, and
`instead relies on incorrect premises
`In the face of the extensive evidence summarized above, Petitioner does not
`
`
`
`present a supplemental declaration from Mr. Schmandt. Moreover, Petitioner
`
`
`
`
`
`
`
`
`
`- 7 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`abandoned its deposition of Dr. Anderson after merely one hour, and thus has
`
`
`
`nothing substantive to introduce from Dr. Anderson’s testimony. Ex. 1035.
`
`
`
`Instead, Petitioner’s Reply presents a series of incorrect conclusions built on
`
`inaccurate or inapplicable premises. Each of Petitioner’s logically false arguments
`
`is addressed and refuted in turn below.
`
`1.
`
`Petitioner’s false logic 1: “a POSITA is assumed to have
`requisite knowledge” to enable the Petition’s combination
`Petitioner attempts to conflate the Petition’s combination with the
`
`
`
`embodiments and claims of the ’140 Patent. Petitioner argues that if the patent did
`
`not discuss the problems with the Petition’s combination, it must be assumed that
`
`the POSA would have known how to overcome the problems with the Petition’s
`
`combination. Paper 22 at 4-6. That logic is false.
`
`
`
`“Skill in the art does not act as a bridge over gaps in substantive presentation
`
`of an obviousness case, but instead supplies an important guarantee of objectivity
`
`in the process.” Al-Site Corp. v. VSI Int'l, Inc., 174 F.3d 1308, 1324 (Fed. Cir.
`
`1999) (emphasis added).
`
`
`
`Here, the Petition’s combination begins with Jiang and Chen, and proposes
`
`to combine those references in particular ways that neither reference enables.
`
`Section II.A, supra. Petitioner has failed to demonstrate through actual evidence
`
`
`
`
`
`
`
`
`
`- 8 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`that the POSA, as defined in the Petition, would have the necessary skill or
`
`
`
`knowledge to in fact achieve the Petition’s combination with a reasonable
`
`expectation of success. Section II.A, supra. Patent Owner has demonstrated
`
`through detailed evidence that the POSA could not. Section II.A, supra.
`
`Petitioner’s assertion that the POSA should be “assumed to have requisite
`
`knowledge” to achieve the Petition’s contorted and highly complex combination,
`
`Paper 22 at 4, is the essence of hindsight bias. Petitioner cannot weaponize the
`
`level of ordinary skill in the art, which is a safeguard against hindsight bias, to fill
`
`substantive gaps in its obviousness case. Al-Site, 174 F.3d at 1324.
`
`Petitioner’s premise is that if the ’140 Patent did not teach how to solve the
`
`specific problems arising out of Petition’s combination, then the Board must
`
`conclude that the POSA would know how to solve Petitioner’s problems. Paper 22
`
`at 2-5. That reasoning is false: the ’140 Patent does not teach performing Jiang’s
`
`speech recognition techniques, which require extensive communication between a
`
`system’s computational components, within memory and processor clusters that
`
`are not all directly or adjacently accessible to each other, as Petitioner’s
`
`combination requires. Ex. 2020 ¶¶ 29-30. The ’140 Patent thus has no reason to
`
`teach solutions to problems arising out of Petitioner’s combination. Paper 22 at 4.
`
`The patent’s silence on those issues is certainly not evidence that the solutions
`
`
`
`
`
`
`
`
`- 9 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`were within the skill level of the Petition’s POSA. The consequence of the
`
`
`
`Petition’s absurd combination theory lies with Petitioner, not Patent Owner.
`
`Moreover, Dr. Anderson explained at his deposition that the ’140 Patent
`
`does provide important teachings and solutions that Petitioner has not shown were
`
`known in the prior art, and that are distinct in key respects from Petitioner’s
`
`combination of Jiang and Chen. Ex. 1035 at 23:6-27:3.
`
`Petitioner may not, of course, demonstrate obviousness based on the ’140
`
`Patent’s own teachings. “[T]he inventor’s own path itself never leads to a
`
`conclusion of obviousness; that is hindsight.” Univ. of Strathclyde v. Clear-Vu
`
`Lighting LLC, 17 F.4th 155, 162 (Fed. Cir. 2021) (emphasis added); TQ Delta,
`
`LLC v. Cisco Sys., 942 F.3d 1352, 1362 (Fed. Cir. 2019).
`
`Petitioner’s case law citations do not establish otherwise. In Uber Techs.,
`
`Inc. v. X One, Inc., 957 F.3d 1334, 1339 (Fed. Cir. 2020), the issue was whether a
`
`POSA would have been motivated to select between one of only two known, finite
`
`approaches. Because the challenged patent was directed to both solutions, the
`
`Federal Circuit held that the appropriate choice was within the level of skill of the
`
`POSA. Id. Here, the issue is expectation of success, not motivation to combine,
`
`there are not finite choices, and the ’140 Patent does not teach Petitioner’s
`
`combination.
`
`
`
`
`
`
`
`
`- 10 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`In In re Publicover, 813 F. App’x 527, 531-32 (Fed. Cir. 2020), the prior art
`
`
`
`and the challenged patent both taught identification of vestibulo-ocular movement,
`
`a type of eye movement. Patent Owner presented only “attorney argument,” not
`
`evidence, that the prior art’s disclosures failed to adequately explain how to
`
`identify that type of eye movement. Id. The Federal Circuit rejected that “attorney
`
`argument” because the challenged patent likewise taught identifying that type of
`
`eye movement but did not provide a detailed teaching of how to do so. Id. Here,
`
`neither Jiang nor Chen teaches implementing Jiang’s speech recognition
`
`techniques on a clustered parallel processing, shared cluster-memory
`
`architecture—much less in the manner Petitioner has proposed. Ex. 2017 at 86:9-
`
`17, Ex. 2020 ¶ 24. Accordingly, the prior art does not and cannot carry any
`
`presumption of enablement as to Petitioner’s combination. In re Antor Media
`
`Corp., 689 F.3d 1282, 1287-88 (Fed. Cir. 2012) (prior art only presumed enabling
`
`for what it teaches). Moreover, the ’140 Patent teaches approaches that are distinct
`
`in key respects from Petitioner’s combination of Jiang and Chen, and provides
`
`solutions that, in any case, have not been shown in the prior art. Ex. 1035 at 23:6-
`
`27:3. Thus, the ’140 Patent provides no assumption that Petitioner’s combination
`
`was within the POSA’s skill. Finally, Patent Owner’s demonstration of a lack of a
`
`
`
`
`
`
`
`
`
`- 11 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`reasonable expectation of success in this proceeding is based on extensive
`
`
`
`evidence, not “attorney argument.” Publicover is thus inapposite.
`
`In re Epstein, 32 F.3d 1559, 1568 (Fed. Cir. 1994), fails to support Petitioner
`
`for the same reasons. There, the Federal Circuit simply held that “one of skill in the
`
`art would have known how to implement the features of the references” as taught
`
`in each reference individually. Id. (emphasis added). Here, Petitioner’s
`
`combination does not simply require practicing what Jiang or Chen taught
`
`individually—it requires a combination that neither reference taught, and that a
`
`POSA would not have been able to achieve. Ex. 2020 ¶¶ 23-32 (pages 15-20).
`
`Petitioner’s reliance on the Board’s decisions in Walmart and Samsung is
`
`misplaced because those decisions do not carry legal authority. Nonetheless, both
`
`decisions also fail to support Petitioner’s argument.
`
`In Walmart, the Board relied on the challenged patent’s teaching that “a
`
`plurality of light panels 40 may be disposed end to end,” and the patent’s silence as
`
`to how to address challenges that would arise “when disposing the panels in
`
`electrical series.” IPR2022-00534, Paper 50 at 32 (emphasis added). By contrast,
`
`the problems Patent Owner has identified with Petitioner’s combination do not
`
`arise out of practicing what the ’140 patent itself teaches; they arise out of
`
`
`
`
`
`
`
`
`
`- 12 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`Petitioner’s combination design. The Board’s decision in Walmart is thus
`
`
`
`inapposite.
`
`In Samsung, the prior art disclosed wireless communication, and thus was
`
`entitled to a presumption of enablement for its own teachings. IPR2022-00324,
`
`Paper 33 at 36-37. Moreover, the challenged patent itself disclosed the same
`
`wireless communication techniques as the prior art without implementation details.
`
`By contrast, the problems with Petitioner’s combination here do not arise out of the
`
`individual teachings of any reference, but from a combination of teachings that is
`
`neither disclosed nor enabled by either reference. Section II.A, supra. Petitioner’s
`
`combination thus carries no presumption of enablement. Nor does the ’140 patent
`
`disclose Petitioner’s combination without implementation details. Ex. 1035 at
`
`23:6-27:3.
`
`Petitioner’s reliance on Keynetik is further misplaced. Paper 22 at 6. There,
`
`the patent owner had “cited no evidence” to contradict petitioner’s assertions of a
`
`reasonable expectation of success, and had instead relied only on “attorney
`
`argument.” Keynetik, Inc. v. Samsung Elecs. Co., Ltd., 2023 U.S. App. LEXIS
`
`3567, at *3-4 (Fed. Cir. 2023). The Federal Circuit held that even though
`
`petitioner’s alleged evidence was “brief, in the absence of any contradictory
`
`evidence, it constitutes substantial evidence to support the Board’s finding.” Id.
`
`
`
`
`
`
`
`
`- 13 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`Here, by contrast, Patent Owner has presented extensive evidence disproving a
`
`
`
`reasonable expectation of success, and showing that Petitioner’s combination
`
`would not have simply required routine and simple software programming within
`
`the POSA’s skill. Section II.A, supra. In particular, Dr. Anderson (one of
`
`extraordinary skill) has testified that he and a team of engineers failed to
`
`successfully port far simpler signal processing software to a parallel processing
`
`hardware architecture. Ex. 2020 ¶ 31. Keynetik is thus inapposite.
`
`Finally, the problems Patent Owner has identified with respect to
`
`Petitioner’s combination are not “generic,” as Petitioner contends. Paper 22 at 7.
`
`They are highly detailed and specific to Petitioner’s combination, and supported by
`
`the testimony of both experts. Section II.A, supra; Ex. 1035 at 23:6-27:3. For the
`
`same reasons, the challenged claims do not need to recite either the problems
`
`arising from Petitioner’s combination theory or the solutions to those problems, as
`
`Petitioner argues. Paper 22 at 7. The question of a reasonable expectation of
`
`success for the combination must be assessed based on the details Petitioner has
`
`proposed for the combination, not the words of the challenged claims. Arctic Cat
`
`Inc. v. Bombardier Rec. Prods., 876 F.3d 1350, 1360-61 (Fed. Cir. 2017)
`
`(requiring proof that “an ordinarily skilled artisan would [ ] believe that a
`
`
`
`
`
`
`
`
`
`- 14 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`particular combination would have a reasonable expectation of ‘anticipated
`
`
`
`success’) (emphasis added).
`
`2.
`
`Petitioner’s false logic 2: A POSA would have known how
`to achieve Petitioner’s combination through general
`knowledge of parallel processing architectures and various
`speech recognition systems
`Petitioner argues that because Jiang allegedly taught “using multiple
`
`processors,” and because references like Mathew taught parallel processing in
`
`speech recognition, it follows that the POSA would have known how to achieve
`
`Petitioner’s specific combination of Jiang and Chen. Paper 22 at 11-15.
`
`That logic is false. Prior art is only presumptively enabled for what it
`
`teaches. In re Antor Media Corp., 689 F.3d 1282, 1287-88 (Fed. Cir. 2012). Mr.
`
`Schmandt has admitted that neither Chen nor Jiang teaches Petitioner’s
`
`combination, and thus neither reference is entitled to a presumption of enablement
`
`with respect to Petitioner’s combination, as Petitioner contends. Compare Paper 22
`
`at 12-15 with Ex. 2017 at 86:9-17, Ex. 2020 ¶¶ 23-32 (p. 15-19); Raytheon Techs.
`
`Corp. v. GE, 993 F.3d 1374, 1380-81 (Fed. Cir. 2021). Even assuming Chen
`
`teaches an “operable clustered processor architecture,” Paper 22 at 15, Chen does
`
`not teach or enable using its architecture to run Jiang’s speech recognition
`
`
`
`
`
`
`
`
`
`- 15 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`techniques (much less in the manner the Petition has proposed), or to run any
`
`
`
`speech recognition. Ex. 2017 at 86:9-17.
`
`And Petitioner itself states that “Mathew is dissimilar architecture” to the
`
`combination, so Mathew could not have enabled the combination. Paper 22 at 21
`
`(emphasis added). The same is true for Hon and Ravishankar. Indeed, the evidence
`
`is uncontroverted that Petitioner’s combination appears nowhere in the prior art,
`
`but is instead a novel fabrication concocted from multiple modifications to the
`
`teachings of two entirely distinct references, one of which (Chen) is not even in the
`
`field of speech recognition, in a manner not taught or enabled by any reference.
`
`Section II.A, supra. The law does not permit assuming a combination to be
`
`enabled when the prior art of record does not enable the combination. Raytheon,
`
`993 F.3d at 1380-81.
`
`
`
`Moreover, Dr. Anderson’s detailed testimony refutes Petitioner’s conclusory
`
`assumptions of a reasonable expectation of success. Ex. 2020 ¶¶ 23-32. Indeed, Dr.
`
`Anderson has explained that, despite his extraordinary level of skill, and the help
`
`of multiple other engineers, he could not successfully transition existing signal
`
`processing software to a parallel processing architecture after many months. Ex.
`
`2020 ¶ 31.
`
`
`
`
`
`
`
`
`
`- 16 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`Petitioner argues that Dr. Anderson’s testimony merely “impugns Dr.
`
`
`
`
`
`Anderson’s credibility,” but “does not prove a POSITA would lack the knowledge
`
`to reasonably succeed in the combination.” Paper 22 at 10-11. That is incorrect. Dr.
`
`Anderson’s testimony demonstrates that even a person of extraordinary skill in the
`
`art “could not necessarily expect success with respect to far simpler combinations
`
`than the one the Petition has proposed,” Ex. 2020 ¶ 31, and thus supports the
`
`credibility of Dr. Anderson’s testimony that the Petition’s combination was beyond
`
`the level of ordinary skill. Ex. 2020 ¶¶ 23-32 (pages 15-20). Petitioner
`
`hypothecates that “[p]erhaps Dr. Anderson simply did not have the necessary
`
`skills,” Paper 22 at 11, but Dr. Anderson’s qualifications and experience far exceed
`
`Mr. Schmandt’s, and Petitioner did not even attempt to question Dr. Anderson
`
`regarding the challenges he faced in porting existing signal processing software to
`
`a parallel processing architecture. Ex. 1035.
`
`
`
`By contrast, Mr. Schmandt’s lack of experience in building the processor to
`
`memory architecture of any speech recognition system, Ex. 2017 at 34:24-35:5,
`
`32:3-9, 106:10-23, or supervising anyone involved in mapping a speech
`
`recognition model to a clustered memory and processor architecture, id. at 145:20-
`
`24, greatly undermines his opinions regarding a reasonable expectation of success.
`
`And Mr. Schmandt’s opinion is further discredited by his testimony that his
`
`
`
`
`
`
`
`
`- 17 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`declaration does not explain how a POSA could have achieved his combination,
`
`
`
`Ex. 2017 at 48:13-21.
`
`
`
`Petitioner argues that Mr. Schmandt’s testimony is admissible despite those
`
`flaws, Paper 22 at 9-10, but the issue here is not admissibility but rather credibility.
`
`Mr. Schmandt’s conclusory assertions of a reasonable expectation of success
`
`should be given no weight in view of the evidence of record, including his own
`
`admissions.
`
`3.
`
`Petitioner’s false logic 3: “Zentian does not argue that the
`combination is inoperable, [or] could not be done. . . .”
`Petitioner appears to suggest that Patent Owner must prove that Petitioner’s
`
`
`
`combination could never be achieved even with unlimited knowledge, skill, and
`
`resources. Paper 22 at 18. That is false. Patent Owner need only demonstrate that
`
`Petitioner has not proven that a person of ordinary skill in the art at the time of the
`
`invention would have reasonably expected to succeed in combining the prior art
`
`elements in the way Petitioner has proposed. KEYnetik, Inc. v. Samsung Elecs.
`
`Co., 841 F. App’x 219, 227 (Fed. Cir. 2021); Arctic Cat Inc. v. Bombardier Rec.
`
`Prods., 876 F.3d 1350, 1360-61 (Fed. Cir. 2017); In re Cyclobenzaprine
`
`Hydrochloride Extended-Release Capsule Patent Litig., 676 F.3d 1063, 1069 (Fed.
`
`Cir. 2012). Patent Owner has met and exceeded that requirement. Section II.A,
`
`
`
`
`
`
`
`
`
`- 18 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`supra. Petitioner’s arguments fundamentally ignore the crucial distinction between
`
`
`
`persons of ordinary skill and those of extraordinary skill.
`
`4.
`
`Petitioner’s false logic 4: “Zentian’s alleged complications
`. . . would occur regardless of the type of information being
`processed.” Paper 22 at 14.
`Petitioner contends that the problems arising out of its combination would
`
`
`
`occur in any clustered processing architecture. Paper 22 at 14. But Petitioner cites
`
`no evidence in support of that assertion. Id.
`
`
`
`Petitioner later attempts to flip its burden onto Patent Owner, arguing that
`
`Zentian never explains why certain problems “would be present in the combination
`
`but not” in the system of Claim 1. Paper 22 at 20. That argument also fails.
`
`Petitioner cannot shift its burden of proof to Patent Owner. Nonetheless, Patent
`
`Owner has demonstrated in detail that Petitioner’s problems arise out of the
`
`requirements Petitioner and its expert have identified for Petitioner’s combination,
`
`and the teachings of the prior art references they have relied upon, Ex. 2020 ¶¶ 23-
`
`32, not the ’140 Patent. Ex. 1035 at 23:6-27:3.
`
`5.
`
`Petitioner’s false logic 5: “Chen envisions its processors and
`memories are directly and adjacently accessible.” Paper 22
`at 16.
`Petitioner admits that its combination must communicate the probability
`
`
`
`scores calculated by the various processors and stored within the various memories
`
`
`
`
`
`
`
`
`
`- 19 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`of Chen’s clusters across the different clusters. Paper 22 at 16 (“Once the
`
`
`
`cumulative probability score is calculated, inter-cluster processing of the results of
`
`each tree search would be communicated”).
`
`
`
`Petitioner instead contends that Chen’s “processors and memories are
`
`directly and adjacently accessible.” Id. But Petitioner cites no testimony from its
`
`expert witness. Instead, it cites to Chen’s “Prior Art” discussion, where Chen
`
`theorizes an idealized parallel processing system in general terms. Id. (citing Ex.
`
`1005 at 1:56-2:1). That passage of Chen does not describe the embodiment
`
`Petitioner relies upon. Rather, Mr. Schmandt’s own declaration stated that in the
`
`relevant embodiment of Chen, “not all[ ] the clusters are interconnected,” and “the
`
`processors 102a in cluster 100a . . . cannot directly or adjacently access the
`
`memory 104d of cluster 100d, while processors 102d in cluster 100d . . . cannot
`
`directly or adjacently access the memory 104a.” Ex. 1003 ¶ 95. That fact is further
`
`set forth in Chen, Ex. 1005 at 9:10-39, and is demonstrated by Dr. Anderson. Ex.
`
`2020 ¶¶ 50-51. Petitioner’s Reply contention is thus false.
`
`
`
`Petitioner next argues that Chen’s “node switching mechanism” permits the
`
`necessary communications between different cluster processors and memories.
`
`Paper 22 at 17. Again, Petitioner does not cite to any testimony from Mr.
`
`Schmandt. And once again, Petitioner’s assertions are false. Petitioner cites to
`
`
`
`
`
`
`
`
`- 20 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`Chen at 5:54-6:3, which explains that the “two or more processors, and one or
`
`
`
`more input/output ports” within a single cluster are “symmetrically connected by a
`
`node switching mechanism to a homogenous cluster shared memory.” Id.
`
`(emphasis added). Thus, Chen’s “node switching mechanism” allows processors
`
`within one cluster to access the “cluster shared memory” of that same cluster, and
`
`thus does not support Petitioner or disprove Patent Owner. Petitioner’s citation to
`
`Chen at 10:36-55 contains the same teaching, describing access by processors 102a
`
`within cluster 104a to the memory 100a of that same cluster. Ex. 1005 at 10:36-
`
`55. Petitioner’s citation to Chen at 8:52-64 is likewise off the mark. That passage
`
`does not address or negate Chen’s express teachings at 9:34-39 that the processors
`
`in cluster 100a cannot directly or adjacently access the memory of cluster 100d,
`
`and the processors of cluster 100d cannot directly or access the memory of cluster
`
`100a. Ex. 1005 at 9:34-39, Ex. 1003 ¶ 95, Ex. 2020 ¶¶ 50-51.
`
`III. The evidence proves no motivation to combine
`A.
`Patent Owner’s Response disproved Petitioner’s motivations
`Patent Owner’s Response refuted each of the Petition’s motivations through
`
`
`
`
`
`the testimony of both experts. Paper 19 at 17-28.
`
`
`
`With respect to increasing processing power and speed, Mr. Schmandt’s
`
`declaration did not provide any analysis to explain why Chen would allegedly
`
`
`
`
`
`
`
`
`
`- 21 -
`
`
`
`
`
` Case IPR2023-00037
`PATENT OWNER’S SUR-REPLY
`speed up Jiang or relax its pruning threshold, or by how much. Ex. 2020 ¶ 35. To
`
`
`
`the contrary, Mathew I taught that a five-processor system parallel processing
`
`speech recognition system was slower than a two-processor system due to high
`
`synchronization