`____________
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`____________
`REALTEK SEMICONDUCTOR CORP.,
`Petitioner
`v.
`ATI TECHNOLOGIES ULC
`Patent Owner
`____________
`
`Case No.: IPR2023-00922
`
`U.S. Patent No. 8,760,454
`Issue Date: June 24, 2014
`Title: Graphics Processing Architecture Employing a Unified Shader
`____________
`PETITIONER REALTEK SEMICONDUCTOR CORP.’S REPLY TO
`PATENT OWNER’S PRELIMINARY RESPONSE
`____________
`
`
`
`TABLE OF CONTENTS
`
`Page
`
`I.
`II.
`
`The Previous Federal Circuit Decision Is Not Controlling ............................ 1
`The Mangione-Smith Declaration Does Not Address the Differences
`Between the Challenged Claims and the ’871 Patent Decision ..................... 4
`III. The Withheld Gruber Transcript Demonstrates a Lack of Conception
`and Reduction to Practice ............................................................................... 6
`
`i
`
`
`
`1008
`
`Exhibit # Reference Name
`1001
`U.S. Patent 8,760,454 to Morein et al.
`1002
`Prosecution History of U.S. Patent 8,760,454
`1003
`Declaration of Hanspeter Pfister, Ph.D.
`1004
`Curriculum Vitae of Hanspeter Pfister, Ph.D.
`1005
`U.S. Patent 7,038,685 to Lindholm et al. (“Lindholm ’685”)
`1006
`U.S. Patent No. 7,015,913 to Lindholm et al. (“Lindholm ’913”)
`1007
`John Amanatides and Edward Szurkowski, A Simple, Flexible,
`Parallel Graphics Architecture, In Proceedings of Graphics
`Interface at 155-160 (Canadian Information Processing Society
`1993) published in Proc. Graphics Interface ’93 in May 1993
`(“Amanatides”)
`Les Kohn and Neal Margulis, Introducing the Intel i860 64-bit
`Microprocessor, IEEE, Volume 9, Issue 4, pages 15-30, August
`1989 (“Kohn”)
`Harald Selzer, Dynamic Load Balancing within a High
`Performance Graphics System, In Proceedings of Rendering,
`Visualization and Rasterization Hardware (Eurographics' 91
`Workshop) at 37-53 (Springer-Verlag 1993) published in 1993
`(“Selzer”) [Library of Congress]
`Stuart Fiske and William J. Dally, Thread prioritization: A Thread
`Scheduling Mechanism for Multiple-Context Parallel Processors,
`In Proceedings of First Symposium on High-Performance
`Computer Architecture, 1995 at 210-221 (IEEE 1995) published in
`1995 (“Fiske”)
`IEEE Xplore bibliography page for Fiske, Thread prioritization: A
`Thread Scheduling Mechanism for Multiple-Context Parallel
`Processors, visited on May 11, 2023
`Declaration of Gordon MacPherson re Authentication of Fiske
`(May 11, 2023)
`IEEE Xplore bibliography page for Kohn, Introducing the Intel
`i860 64-bit Microprocessor visited on May 11, 2023
`
`1011
`
`1012
`
`1013
`
`PETITIONER’S EXHIBIT LIST
`
`ii
`
`1009
`
`1010
`
`
`
`1014
`
`1015
`
`1016
`
`1017
`
`Declaration of Gordon MacPherson re Authentication of Kohn
`(May 11, 2023)
`Initial Determination on Violation of Section 337, Certain
`Consumer Electronics and Display Devices With Graphics
`Processing and Graphics Processing Units Therein, 337-TA-932
`(October 9, 2015)
`Harald Selzer, Dynamic Load Balancing within a High
`Performance Graphics System, In Proceedings of Rendering,
`Visualization and Rasterization Hardware (Eurographics' 91
`Workshop) at 37-53 (Springer-Verlag 1993) published in 1993
`(“Selzer”) [University of California, Berkeley, Library]
`Redacted Deposition Transcript of Andrew E. Gruber, taken March
`17, 2015, in ITC Case No. 337-TA-932
`
`iii
`
`
`
`Pursuant to the Board’s authorization of October 6, 2023, Petitioner submits
`
`this Reply to Patent Owner’s Preliminary Response (“POPR”) to address the
`
`material deficiencies and omission of key evidence from the POPR. Once the
`
`withheld evidence is properly considered as part of the record, the POPR and
`
`Mangione-Smith declaration clearly do not show prior invention relevant to
`
`Ground 1 of the Petition.1 The subject matter in the omitted ATI 30(b)(6)
`
`deposition transcript also makes clear that a previous Federal Circuit decision did
`
`not decide the relevant issue here regarding whether the register allocation scheme
`
`in the Challenged Claims was conceived and reduced to practice. Much like the
`
`Mangione-Smith declaration filed with the POPR, it was not addressed.
`
`Petitioner’s Reply addresses three points: 1) irrelevance of the prior Federal
`
`Circuit decision, 2) deficiencies in the Mangione-Smith declaration related to the
`
`omitted evidence and differences between the Challenged Claims and claims in
`
`related patents, and 3) the omitted inventor and 30(b)(6) deposition transcript that
`
`contradicts the POPR and declaration and dooms Patent Owner’s position.
`
`I.
`
`The Previous Federal Circuit Decision Is Not Controlling
`
`The Challenged Claims include elements that were not addressed by the
`
`1 Patent Owner’s alleged prior invention argument has no bearing on Grounds 2
`
`and 3, which are not addressed in this Reply.
`
`1
`
`
`
`Federal Circuit’s decision of a related patent (U.S. Patent No. 6,897,871) in ATI
`
`Techs. ULC v. Iancu, 920 F.3d 1362 (Fed. Cir. 2019). In addition, the Federal
`
`Circuit and the PTAB did not address conception in that case because it was not
`
`challenged. Accordingly, the Federal Circuit decision does not dictate, much less
`
`address the issues presented in, this proceeding.
`
`First, the Challenged Claims (using claim 1 as an example below) include
`
`limitations related to the allocation of register blocks to be used for performing
`
`either vertex or pixel operations:
`
`The ’871 patent claims and the Federal Circuit decision mention nothing about
`
`allocation of or use of register space as a predicate for determining which type of
`
`operation to perform (e.g., vertex or pixel operations). See also Ex. 1017 at 168:5-
`
`23 (ATI witness confirming that the ’871 patent did not cover dynamic register
`
`allocation). Nor do the ’871 patent claims or the Federal Circuit decision address
`
`the failure of ATI to conceive of or reduce to practice the “dynamic register
`
`allocation” in the R400. Importantly, Patent Owner’s Preliminary Response does
`2
`
`
`
`not address the differences in the claims, and to the extent it attempts to do so in its
`
`Sur-Reply, this factual dispute should be resolved through a full proceeding.
`
`Second, conception was not contested in the ’871 patent IPR proceeding,
`
`and therefore, it was not addressed by the PTAB or the Federal Circuit. This is a
`
`critical distinction because ATI did not conceive of the use of register allocation in
`
`the context of the R400 or any other project, as recited in the Challenged Claims,
`
`prior to their effective filing date. Patent Owner failed to address this
`
`distinguishing factor in its Preliminary Reply. At most, ATI identified a problem,
`
`as confirmed by its witness testimony, but it did not identify a solution, which is
`
`required to establish prior inventorship. Singh v. Brake, 317 F.3d 1334, 1341 (Fed.
`
`Cir. 2003) (finding no conception for an unworking embodiment because “[e]ven
`
`if the entry expressed the problem, it did not provide the solution.”).
`
`The only court decision to address the lack of conception and reduction to
`
`practice related to the R400’s register allocation was the determination of the
`
`International Trade Commission in 337-TA-932. See Petition at 15-16. Indeed,
`
`the very evidence that Patent Owner withheld from its Preliminary Reply is the
`
`same evidence relied upon by Judge Pender in the 932 Investigation to find the
`
`lack of an “operative and complete invention” that is required to prove conception.
`
`Patent Owner withheld that testimony because it contradicts any argument it could
`
`attempt regarding conception and reduction to practice.
`
`3
`
`
`
`Accordingly, any attempted reliance on the Federal Circuit decision would
`
`fall short of addressing the specific, and different, claim elements at issue here.
`
`II.
`
`The Mangione-Smith Declaration Does Not Address the Differences
`Between the Challenged Claims and the ’871 Patent IPR Decision
`
`Patent Owner relies upon a declaration from Dr. Mangione-Smith to support
`
`its claim to previous conception and reduction to practice. Specifically, Dr.
`
`Mangione-Smith asserts that the ATI “inventors…created an implementation of the
`
`R400 in source code, e.g., emulation code and RTL.” Ex. 2001, ¶196. But the
`
`Mangione-Smith declaration does not perform any analysis of the R400 RTL code
`
`and relies only on the ’871 IPR proceeding and a declaration from Dr. Wolfe. Ex.
`
`2001 at ii (citing Ex. 2002, the Wolfe Declaration). This leaves the Mangione-
`
`Smith declaration deficient for at least three reasons.
`
`First, the Mangione-Smith declaration relies solely on draft specifications
`
`for the R400 while ignoring the corresponding RTL code and omitted inventor
`
`testimony. As ATI’s witness, Andrew Gruber, previously testified, “RTL is
`
`probably the best guide” and “anything not in the RTL code is not in the GPU.”
`
`Ex. 1017 at 52:11-53:3 (testifying that RTL is more definitive than the draft ATI
`
`specifications relied upon by Dr. Mangione-Smith). And Dr. Mangione-Smith
`
`cannot credibly claim ignorance of this prior testimony. That is, Dr. Mangione-
`
`Smith was one of ATI’s experts in the 1318 Investigation, and therefore, had
`
`access to the Gruber deposition transcript because it was marked as an exhibit and
`4
`
`
`
`produced by ATI. Dr. Mangione-Smith’s declaration in this IPR proceeding puts
`
`on blinders and presents an expert opinion that knowingly contradicts the sworn
`
`testimony of an ATI inventor on the ’454 patent. The Mangione-Smith declaration
`
`should be given no weight, especially in the absence of a more complete
`
`evidentiary record that may be developed following institution.
`
`Second, Dr. Mangione-Smith’s reliance on the R400 as being continuous
`
`with the R500 and Xenos projects at ATI is flatly contradicted by Mr. Gruber. His
`
`testimony confirmed that there was no “seamless extension” from the R400 to the
`
`other projects and that there were “substantial changes” in the R500 and R600,
`
`which became Xenos. Ex. 1017 at 7:23-8:22; compare with Ex. 2001 at 208
`
`(Mangoine-Smith labeling R400 to Xenos as “seamless extensions.”). Indeed, the
`
`“sequencer,” which is the component claimed in the ’454 Challenged Claims, “had
`
`substantial changes” and even had different teams working on it. Ex. 1017 at 9:1-
`
`10. Even worse, as Mr. Gruber testified, the R400 was never taped out and the
`
`dynamic register allocation never worked. Ex. 1017 at 104:20-25. Dr. Mangione-
`
`Smith ignored all of this testimony, which flatly contradicts the conclusions set
`
`forth in his declaration citing cherry-picked documents.
`
`Third, while the Mangione-Smith declaration ignored the contents of the
`
`R400 RTL code (as it needed to do in order to contradict the Gruber testimony),
`
`the Wolfe declaration from the ’871 proceeding actually reviewed RTL code and
`
`5
`
`
`
`concluded that the R400 had a fixed vertex priority scheme that did not depend on
`
`any amount of data in storage. But that is the opposite of what is required by the
`
`claims of the ‘454 patent. Accordingly, to the extent the Mangione-Smith
`
`declaration was incorporating by reference the opinion from Dr. Wolfe in the ’871
`
`IPR proceeding (which the declaration purports to do), then Patent Owner’s expert
`
`testimony conclusively establishes that ATI did not previously conceive of or
`
`reduce to practice what is required by the Challenged Claims.
`
`III. The Withheld Gruber Transcript Demonstrates a Lack of Prior
`Conception and Reduction to Practice
`
`All the Challenged Claims require a form of memory allocation that dictates
`
`whether vertices or pixels will be processed by the claimed shader. AMD’s
`
`30(b)(6) witness, and co-inventor on the ’454 patent, Andrew Gruber, provided
`
`testimony in the 932 ITC investigation—which was not considered in the ’871 IPR
`
`proceeding or by the Federal Circuit—demonstrating that the claims of the ’454
`
`patent were not conceived or reduced to practice as part of ATI’s R400 project.
`
`Ex. 1017. Petitioner discussed Judge Pender’s ruling that relied on Mr. Gruber’s
`
`testimony in the Petition at pages 15-16. ATI is well-aware of Mr. Gruber’s
`
`testimony and is in possession of Mr. Gruber’s deposition transcript from the 932
`
`Investigation, as it was produced and cited as an exhibit by ATI in a recent ITC
`
`Investigation. Yet, ATI produced over 200 exhibits with its Preliminary Response,
`
`6
`
`
`
`including another deposition transcript from Mr. Gruber, but conspicuously
`
`omitted Mr. Gruber’s testimony from the 932 Investigation. This was no mistake.
`
`The Challenged Claims require a register allocation method that was never
`
`conceived of in the R400 project and never made functional for use in Xenos. Ex.
`
`1017 at 103:11-104:25. Indeed, Xenos used a fixed register system that is not the
`
`dynamic method in the Challenged Claims. Id. at 14:18-15:10. Mr. Gruber’s
`
`testimony could not be clearer that, even years after the filing of the parent
`
`application to the ‘454 patent in 2003, the register allocation scheme did not work:
`
`Ex. 1017 at 104:20-25 (discussing the first tapeout of Xenos in late 2004/2005).
`
`Mr. Gruber went on to explain how this “aspirational” dynamic register allocation
`
`method included flaws that prevented it from being functional and may have never
`
`come close to working in any form until the later R600 (years after the patent
`
`filings). Id. at 104-107 (further testifying that the R400 “never got to the point” of
`
`even testing the flawed dynamic register allocation). Instead of being functional,
`
`the R400 and Xenos used a “fixed” register space widely available in the prior art
`
`that simply dedicated 50 percent of the register space to vertices and 50 percent of
`
`the register space to pixels. Id. at 107:10-16. With respect to the date that ATI
`7
`
`
`
`(including Mr. Gruber as a co-inventor) allegedly conceived of the register file
`
`allocation in the ’454 patent, Mr. Gruber testified that “I cannot provide a date”
`
`and it may have been in 2002. Id. at 115:115:19-116:8. Again, Mr. Gruber’s
`
`testimony contradicts the Mangione-Smith declaration and POPR.
`
`Mr. Gruber’s testimony regarding the RTL code, which the Mangione-Smith
`
`declaration ignores and does not analyze, further demonstrates that the R400 (and
`
`Xenos) cannot be an operative conception, much less a reduction to practice, of the
`
`Challenged Claims. That is, Mr. Gruber testified that in the R400, the dynamic
`
`register allocation mode was nonfunctional and the “static mode” did not utilize
`
`any priority and could not utilize any allocation of shared resources as required by
`
`the Challenged Claims. Id. at 167:13-21. Thus, there was no priority given to
`
`vertex or pixel threads based on the allocation of register space as required by the
`
`Challenged Claims. Dr. Mangione-Smith again ignores this relevant testimony and
`
`his declaration is contradicted by the unsubmitted evidence.
`
`By producing a different Gruber deposition transcript that did not address
`
`the register allocation scheme of the Challenged Claims, the POPR sought to hide
`
`ATI’s failure to invent earlier than Petitioner’s Ground 1 prior art from NVIDIA.
`
`The Federal Circuit also did not have this Gruber transcript, but also did not need
`
`to decide the issue of whether ATI conceived of the claimed register allocation
`
`method. ATI did not, and the Petition should be instituted as to Ground 1.
`
`8
`
`
`
`Dated: October 13, 2023
`
`Respectfully submitted,
`
`/Jeffrey L. Johnson/
`Jeffrey L. Johnson
`ORRICK, HERRINGTON & SUTCLIFFE LLP
`609 Main, 40th Floor
`Houston, TX 77002-3106
`Telephone: (713) 658-6400
`Facsimile: (713) 658-6401
`Email: 3J6PTABDocket@orrick.com
`
`Christopher J. Higgins
`ORRICK, HERRINGTON & SUTCLIFFE LLP
`Columbia Center
`1152 15th Street, N.W.
`Washington, D.C. 20005-1706
`Telephone: (202) 339-8400
`Facsimile: (202) 339-8500
`Email: OCHPTABDocket@orrick.com
`
`Steve Baik
`WHITE HAT LEGAL
`P. O. Box 382
`San Jose, CA 95002
`Telephone: (650) 618-5282
`Email: sbaik@whitehat.legal
`
`Counsel for Petitioner Realtek
`Semiconductor Corp.
`
`9
`
`
`
`CERTIFICATE OF SERVICE
`The undersigned certifies that on October 13, 2023, a copy of the following
`
`was served in its entirety via electronic mail, upon the following attorneys of
`
`record for the Patent Owner:
`
`William A. Meunier
`Michael T. Renaud
`Adam S. Rizk
`MINTZ, LEVIN, COHN, FERRIS, GLOVSKY AND
`POPEO, P.C.
`One Financial Center
`Boston, MA 02111
`
`wameunier@mintz.com
`mtrenaud@mintz.com
`arizk@mintz.com
`
`/Karen Johnson/
`Karen Johnson
`
`10
`
`