UNITED STATES PATENT AND TRADEMARK OFFICE
`______________________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`______________________
`MICROSOFT CORPORATION,
`Petitioner,
`v.
`DIRECTSTREAM, LLC,
`Patent Owner.
`_______________________
`Cases IPR2018-01605, IPR2018-01606, IPR2018-01607
`Patent 7,620,800 B2
`__________________________
`
`PATENT OWNER DIRECTSTREAM, LLC’S RESPONSE
`UNDER 37 CFR §42.120
`
`
`
`
`
`
`
`
`
`

`

`TABLE OF CONTENTS
`
`
`I. INTRODUCTION .................................................................................................. 1
`II. TECHNICAL BACKGROUND OF THE ’800 PATENT ................................... 3
`A. Need for Cost-Effective HPC ..................................................................... 4
`B. Direct Streams’s HPC Advancements ........................................................ 9
`1. Architectural Innovations ................................................................ 10
`2. Invention of MAP Technology ....................................................... 11
`3. DirectStream’s Continued Improvement ........................................ 12
`4. DirectStream’s Continued Success ................................................. 14
`C. The ’800 Patent: SRC Invents Methods for Enhancing Parallelism
`and Performance in Reconfigurable Computing Systems. ...................... 16
`III. PETITIONER’S EXPERT TESTIMONY IS CONCLUSORY AND IS
`NOT RELIABLE TO EXPLAIN THE TEACHINGS OF THE PRIOR
`ART OR SUPPORT A FINDING OF OBVIOUSNESS .............................. 18
`A. Dr. Stone Fails to Understand the Teachings of the Patent and the
`Prior Art .................................................................................................... 18
`B. Dr. Stone’s Opinions are Grounded in Hindsight Bias............................. 19
`C. Dr. Stone Fails to Provide Facts and Data to Support His Opinions ........ 22
`IV. PERSON OF ORDINARY SKILL IN THE ART ............................................ 29
`V. DIRECTSTREAM’S CLAIM CONSTRUCTIONS .......................................... 30
`A. “seamless” vs. “systolic” and “data driven” ............................................. 34
`1. “pass computed data seamlessly between said computational
`loops” ............................................................................................ 35
`2. Systolic and Data Driven ................................................................. 43
`B. “stream communication” .......................................................................... 45
`
`
`
`ii
`
`

`

`1. Petitioner’s Construction is Deeply Flawed and Illogical
`Under BRI ..................................................................................... 45
`2. DirectStream’s Construction is Reasonable and Consistent
`with the Plain and Ordinary Meaning. .......................................... 47
`C. “computational loop” … [wherein only functional units needed to
`solve the calculation are instantiated] ...................................................... 65
`VI. THE CHALLENGED CLAIMS ARE PATENTABLE OVER THE
`PRIOR ART. .................................................................................................. 69
`A. Petitioner’s Burden of Proof to Invalidate the Patent. .............................. 69
`1. Anticipation. .................................................................................... 69
`2. Obviousness. .................................................................................... 71
`B. The Challenged Claims are Patentable over Splash2. .............................. 73
`1. The Prior Art Does Not Disclose Stream Communication. ............ 73
`a)
`Splash2 Does Not Disclose Stream
`Communication............................................................... 73
`Based on its Petition, Petitioner Cannot Now
`Argue Splash2 Discloses Stream Communication,
`as Properly Construed ..................................................... 74
`Petitioner’s Other Prior Art Do Not Disclose
`Stream. ............................................................................ 75
`2. The Prior Art Does Not Disclose Two Computational Loops. ....... 76
`a)
`Splash2 Does Not Disclose Two Computational
`Loops. ............................................................................. 76
`Splash2 Also Does Not Disclose Forming the Two
`Computational Loops in the FPGAs. .............................. 83
`Petitioner’s Other Prior Art Do Not Disclose Two
`Computational Loops. .................................................... 85
`
`b)
`
`c)
`
`b)
`
`c)
`
`
`
`iii
`
`

`

`b)
`
`c)
`
`d)
`
`e)
`
`3. The Prior Art Does Not Unambiguously Disclose Seamlessly
`Passing Computed Data Between Computational Loops. ............ 91
`a)
`Splash2 Does Not Unambiguously Disclose
`Seamlessly Passing Computed Data Between
`Computational Loops. .................................................... 91
`Using Petitioner’s Own Construction, Splash2
`Would Still Fail to Invalidate the ’800 Patent. ............... 95
`RaPiD Confirms the Prior Art Teaches Using
`Memory Between Processing Elements to Store
`Results. .......................................................................... 100
`Roccatano also Cannot Disclose “Seamless”
`Because it Requires Multiple Processors. .................... 101
`Petitioner’s Other Prior Art Just as Ambiguous as
`Splash2. ......................................................................... 102
`C. Claims are not rendered obvious by Combining Prior Art. .................... 104
`1. Prior Art Combinations Still Missing Claim Elements ................. 105
`2. A POSITA Would Not Have Been Motivated to Combine the
`Prior Art ...................................................................................... 106
`3. Petitioner and its Expert Failed to Consider Whether it
`Would Be Feasible to Modify the Teachings of the Prior
`Art to Combine with Each Other ................................................ 108
`4. Petitioner and its Expert Improperly Rely on Hindsight
`Reasoning to Combine Prior Art ................................................ 114
`D. The Objective Indicia in this Case Indicate Nonobviousness ................ 116
`VII. CONCLUSION .............................................................................................. 120
`
`
`
`
`
`
`
`
`iv
`
`

`

` TABLE OF AUTHORITIES
`
`CASES:
`ActiveVideo Networks v. Verizon Comm.,
`694 F.3d 1312 (Fed. Cir. 2012) .................................................................... 23, 28
`
`
`Am. Piledriving Equip., Inc. v. Geoquip, Inc.,
`637 F.3d 1324 (Fed. Cir. 2011) ...........................................................................65
`
`
`Asia Vital Components Co., Ltd. v. Asetek Danmark A/S,
`377 F. Supp.3d 990 (N.D. Cal. 2019) ..................................................... 20, 82, 85
`
`
`Brookhill-Wilk 1, LLC v. Intuitive Surgical, Inc.,
`334 F.3d 1294 (Fed. Cir. 2003) ...........................................................................33
`
`
`CAE Screenplates, Inc. v. Heinrich Fiedler GmbH & Co.,
`224 F.3d 1308 (Fed. Cir. 2000) ...........................................................................35
`
`
`Chef America, Inc. v. Lamb-Weston, Inc.,
`358 F.3d 1371, 69 USPQ2d 1857 (Fed. Cir. 2004) .............................................34
`
`
`Chemical Separation Tech., Inc. v. U.S.,
` 51 F. Cl. 771 (Fed. Cl. 2002) ................................................................. 20, 82, 85
`
`Crocs, Inc. v. International Trade Commission,
`598 F.3d 1294 (Fed. Cir. 2010) .........................................................................117
`
`
`Forest Labs, LLC v. Sigmapharm Labs., LLC,
`918 F.3d 928 (Fed. Cir. 2019)..............................................................................71
`
`
`Graham v. John Deere Co.,
`383 U.S. 1 (1966) ........................................................................................ 72, 116
`
`
`Grp. One, Ltd. v. Hallmark Cards, Inc.,
`407 F.3d 1297 (Fed. Cir. 2005) ...........................................................................29
`
`
`Helmsderfer v. Bobrick Washroom Equipment, Inc.,
`527 F.3d 1379 (Fed. Cir. 2008) ............................................................................. 2
`
`
`
`v
`
`

`

`Hill-Rom Servs. v. Matal,
`716 Fed. Appx. 996 (Fed. Cir. 2017) ...................................................................35
`
`
`Hitachi Metals Ltd. v. All. of Rare-Earth PermanentMagnet Indus.,
`699 Fed. Appx. 929 (Fed. Cir. 2017) ...................................................................43
`
`
`In re Bond,
`910 F.2d 831, 15 USPQ2d 1566 (Fed. Cir. 1990) ...............................................70
`
`
`In re Buszard,
`504 F.3d 1364 (Fed. Cir. 2007) ...........................................................................33
`
`
`In re Cortright,
`165 F.3d 1353 (Fed. Cir. 1999) .................................................................... 33, 65
`
`
`In re Cyclobenzaprine,
`676 F.3d 1063 (Fed. Cir. 2012) .........................................................................117
`
`
`In re Dow Chem. Co.,
`837 F.2d 469 (Fed.Cir.1988)................................................................................28
`
`
`In re Gordon,
`733 F.2d 900 (Fed. Cir. 1984)............................................................................107
`
`
`In re Kahn,
`441 F.3d 977 (Fed. Cir. 2006)..............................................................................23
`
`
`In re Lee,
`277 F.3d 1338 (Fed. Cir. 2002) ...........................................................................24
`
`
`In re NTP, Inc.,
`654 F.3d 1268 (Fed. Cir. 2011) ...........................................................................24
`
`
`In re Nuvasive, Inc.,
`841 F.3d 966 (Fed. Cir. 2016)..............................................................................29
`
`
`In re Suitco Surface, Inc.,
`603 F.3d 1255 (Fed. Cir. 2010) ...........................................................................33
`
`
`
`
`
`vi
`
`

`

`In re Translogic Tech., Inc.,
`504 F.3d 1249 (Fed. Cir. 2007) ...........................................................................32
`
`
`In re Zletz,
`893 F.2d 319, 13 USPQ2d 1320 (Fed. Cir. 1989) ...............................................34
`
`
`Innogenetics, N.V. v. Abbott Labs.,
`512 F.3d 1363 (Fed. Cir. 2008) .............................................................. 20, 82, 85
`
`
`Intelligent Bio Systems, Inc. v. Illumina Cambridge, Ltd.,
` 821 F.3d 1359 (Fed. Cir. 2016) .................................................................. passim
`
`InTouch Techs., Inc. v. VGo Commun’s, Inc.,
`751 F.3d 1327 (Fed. Cir. 2014) .................................................................... 26, 28
`
`
`J.T. Eaton & Co., Inc. v. Atlantic Paste & Glue Co.,
`106 F.3d 1563 (Fed. Cir. 1997) .........................................................................117
`
`
`KSR Int’l Co. v. Teleflex Inc.,
`550 U.S. 398 (2007) .................................................................................... 72, 105
`
`
`Liebel-Flarsheim Co. v. Medrad, Inc.,
`358 F.3d 898 (Fed. Cir. 2004)..............................................................................34
`
`
`Motor Vehicle Mfrs. Ass’n v. State Farm Mutual Automobile Ins. Co.,
`463 U.S. 29 (1983) ...............................................................................................31
`
`
`Nike Inc. v. Adidas AG,
`812 F.2d 1326 (Fed. Cir. 2016) .........................................................................117
`
`
`Ortho-McNiel Pharm., Inc. v. Mylan Labs., Inc.,
`520 F.3d 1358 (Fed. Cir. 2008) .........................................................................117
`
`
`Otsuka Pharm. Co. v. Sandoz, Inc.,
`678 F.3d 1280 (Fed. Cir. 2012) .........................................................................106
`
`
`PAR Pharm., Inc. v. TWI Pharms., Inc.,
`773 F.3d 1186 (Fed. Cir. 2014) .................................................................. 72, 106
`
`
`
`
`
`vii
`
`

`

`Para-Ordnance Mfg. v. SGS Importers Int’l,
`73 F.3d 1085 (Fed. Cir. 1995)............................................................................119
`
`
`PersonalWeb Techs., LLC. v. Apple, Inc.,
`917 F.3d 1376 (Fed. Cir. 2019) .................................................................... 72, 94
`
`
`Pfizer, Inc. v. Teva Pharmaceuticals, USA, Inc.,
`429 F.3d 1364 (Fed. Cir. 2005) ............................................................................. 2
`
`
`Phillips v. AWH Corp.,
`415 F.3d 1303 (Fed. Cir. 2005) ................................................................... passim
`
`
`Plas-Pak Indus. v. Sulzer Mixpac AG,
`600 Fed.Appx. 755 (Fed. Cir. 2015) ..................................................................107
`
`
`Rexnord Indus. v. Kappos,
`705 F.3d 1347 (Fed. Cir. 2013) ...........................................................................71
`
`
`Richardson v. Suzuki Motor Co.,
`868 F.2d 1226, 9 USPQ2d 1913 (Fed. Cir. 1989) ...............................................70
`
`
`Schering Corp. v. Geneva Pharms.,
`339 F.3d 1373 (Fed. Cir. 2003) ...........................................................................71
`
`
`Smiths Indus. Med. Sys., Inc. v. Vital Signs, Inc.,
`183 F.3d 1347 (Fed. Cir. 1999) ...........................................................................71
`
`
`Stratoflex, Inc. v. Aeroquip Corp.,
`715 F.2d 1530 (Fed. Cir. 1983) .........................................................................117
`
`
`Sunrace Roots Enter. Co. v. SRAM Corp.,
`336 F.3d 1298 (Fed. Cir. 2003) ...........................................................................33
`
`
`Tandon Corp. v. USITC,
`831 F.2d 1017 (Fed. Cir. 1987) ...........................................................................42
`
`
`U.S. v. Nixon,
`418 U.S. 683 (1974) .............................................................................................31
`
`
`
`
`
`viii
`
`

`

`Uniroyal, Inc. v. Rudkin-Wiley, Corp.,
`837 F.2d 1044 (Fed. Cir. 1988) .........................................................................119
`
`
`Verdegaal Bros. v. Union Oil Co. of California,
`814 F.2d 628, 2 USPQ2d 1051 (Fed. Cir. 1987) .................................................69
`
`
`Verve, LLC v. Crane Cams, Inc.,
`311 F.3d 1116 (Fed. Cir. 2002) .................................................................... 69, 94
`
`
`W.L. Gore & Associates, Inc. v. Garlock, Inc.,
`721 F.2d 1540, 220 USPQ 303 (Fed. Cir. 1983) ................................ 19, 109, 114
`
`
`Wasica Fin. GmbH v. Cont’l Auto. Sys., Inc.,
`853 F.3d 1272 (Fed. Cir. 2017) .................................................................... 70, 94
`PTAB DECISIONS:
`Apple Inc. v. Sightsound Technologies, LLC,
`CMB2013-00023 Paper 101 (P.T.A.B. Oct. 7, 2014)........................................... 117
`
`
`Apple Inc. v. Uniloc Luxembourg, S.A.,
`IPR2017-02202, Paper 10 (P.T.A.B. Dec. 13, 2018) .............................................. 81
`
`
`Caterpillar Inc. v. Wirtgen America, Inc.,
`IPR2017-02188, Paper 71 (P.T.A.B. July 15, 2019) ................................... passim
`
`
`Commerce Bancshares, Inc. v. Intellectual Ventures II LLC,
`IPR2014-00793, Paper 9 (P.T.A.B. Feb. 6, 2015) ...............................................24
`
`
`Corning Inc. v. DSM LP Assets B.V,
`IPR2013-00048, Paper 96 (P.T.A.B. July 11, 2014) ...........................................69
`
`
`Hologic, Inc. v. Enzo Life Sciences, Inc.,
`IPR2018-00019, Paper 17 (P.T.A.B. April 18, 2018) .................................. 85, 100
`
`
`Infobionic, Inc. v. Braemer Manufacturing, LLC,
`IPR2015-01704, Paper 11 (P.T.A.B. Feb. 16, 2016) ............................................ 81
`
`
`Kinetic Techs. v. Skyworks Solutions, Inc.,
`IPR2014-00529, Paper 8 (P.T.A.B. Sept. 23, 2014) ............................................. 81
`
`
`
`ix
`
`

`

`
`Masabi Ltd. v. Bytemark, Inc.,
`IPR2017-01449, Paper 38 (P.T.A.B. Dec. 3, 2018) ......................................... 81
`
`
`TRW Automotive US LLC v. Magna Elecs., Inc.,
`IPR2014-00258, Paper 18 (P.T.A.B. Aug. 27, 2014) ........................................... 82
`
`
`World Bottling Cap, LLC v. Crown Packaging Technology, Inc.,
`IPR2015-01651, Paper 34 (P.T.A.B. Jan. 19, 2017) ...................................... 116
`
`STATUTES:
`5 U.S.C. § 706 ..........................................................................................................30
`
`35 U.S.C. §102 .........................................................................................................89
`
`35 U.S.C. §103 .................................................................................................. 69, 71
`
`35 U.S.C. §112 .........................................................................................................46
`
`35 U.S.C. §312 .........................................................................................................28
`
`35 U.S.C. § 316 .......................................................................................................... 1
`REGULATIONS:
`37 C.F.R. § 41.65 .....................................................................................................23
`
`37 C.F.R. § 42.104 ...................................................................................................90
`
`37 C.F.R. § 42.22 .....................................................................................................28
`
`OTHER AUTHORITIES:
`
`MPEP §2111 ..................................................................................................... 32, 33
`
`MPEP §2141 ............................................................................................................22
`
`MPEP §2143 ................................................................................................. 105, 106
`
`
`
`
`
`x
`
`

`

`I. INTRODUCTION
`For the reasons summarized below, Petitioner has failed to meet its burden of
`
`proving, by a preponderance of the evidence, that any of the challenged claims are
`
`unpatentable. 35 U.S.C. §316(e). Accordingly, the Petition must be denied for at
`
`least the following four reasons.
`
`First, Petitioner’s expert testimony fails to appreciate the then state-of-the-art
`
`and understood by any competent POSITA regarding high-performance computing
`
`systems. To arrive at his conclusions, Petitioner’s expert, Dr. Stone ignored all the
`
`problems and issues of microprocessors, ASIC chips, FPGA capabilities to perform
`
`floating point operations, high-performance computing design challenges, or web
`
`search algorithms. This lack of context resulted in Dr. Stone using impermissible
`
`hindsight bias to arrive at the Patentee’s solutions rather than to consider the
`
`problems the patent was solving. Worse, Dr. Stone’s conclusory testimony is
`
`undercut by the textbook he himself authored in 1987. In short, Dr. Stone’s
`
`unreliable and scientifically unsound testimony is not a valid basis for finding any
`
`of the patents claims invalid.
`
`Second, Petitioner and its expert Dr. Stone rely on incorrect or incomplete
`
`claim constructions that do not accurately convey the scope of the claims as
`
`understood by a POSITA. For example, Microsoft failed to offer any construction
`
`of the term “computational loop” despite it being present in every single challenged
`
`
`
`1
`
`

`

`claim. In addition, Petitioner and Dr. Stone have incorrectly construed the terms
`
`“systolic,” “data driven,” and “seamlessly” to be synonymous. Microsoft’s approach
`
`conflicts with binding Federal Circuit precedent holding that “claim terms are
`
`presumed to have different meanings” and claims should be construed in a manner
`
`that “renders the patent internally consistent.” See, e.g., Helmsderfer v. Bobrick
`
`Washroom Equipment, Inc., 527 F.3d 1379, 1382 (Fed. Cir. 2008); Pfizer, Inc. v.
`
`Teva Pharmaceuticals, USA, Inc., 429 F.3d 1364, 1373 (Fed. Cir. 2005). These
`
`incorrect, or absent, constructions underpin the entirety of Microsoft’s Petition.
`
`Third, Petitioner and Dr. Stone have misinterpreted, or misrepresented, the
`
`prior art and its teachings. None of the art cited by Microsoft teaches “computational
`
`loops,” a “stream communication,” or “seamlessly passing computed data between
`
`computational loops” as the claims require.
`
`Fourth, Petitioner and Dr. Stone have failed to articulate how or why a
`
`POSITA would combine any of the asserted prior art. Petitioner and Dr. Stone
`
`simply assume that the references would be feasible to combine without considering
`
`any of the problems associated with trying to do so. Simply put, a POSITA would
`
`have known of the need to analyze the costs and benefits associated with the various
`
`overhead from new designs, which even Dr. Stone stated in the textbook he wrote in
`
`1987. Nevertheless, neither Dr. Stone nor Petitioner undertook such an analysis.
`
`Instead, Petitioner’s only reason to combine is Dr. Stone’s conclusory testimony,
`
`
`
`2
`
`

`

`which uses hindsight to ignore the state of the art and other technical solutions
`
`available to a POSITA. This simply is not sufficient.
`
`DirectStream invented revolutionary HPC reconfigurable systems. Microsoft
`
`has not carried its burden of proof to demonstrate that these novel systems are
`
`anticipated or rendered obvious by any of the prior art cited in the Petition.
`
`Accordingly, the Board should find every claim of the ’800 Patent patentable.
`
`II. TECHNICAL BACKGROUND OF THE ’800 PATENT
`Each of the challenged DirectStream Patents address a technical problem that
`
`arises in the context of High-Performance Computing (“HPC”).1 Throughout
`
`
`1 DirectStream contends the state of the art of HPC architecture and the design of
`
`reconfigurable systems is an integral part of understanding all DirectStream patents
`
`subject to Microsoft’s IPRs, i.e., the IPR2018-01594, -01599, -1600, -1601
`
`consolidated, -1604, and -1605 consolidated actions. Chronologically, the -01599 and
`
`-01600 Patent Owner Responses, should be read first as those patents represent
`
`DirectStream’s earlier innovations in HPC architecture and engineering that set the
`
`stage for subsequent innovations and improvements that led to subsequent patents. The
`
`evidence submitted in those cases will provide the Board with a full, contextual
`
`understanding of HPC, computing architecture, constituent parts, and reconfigurable
`
`systems as applied to IPR2018-01594, -1601, and -1605.
`
`
`
`3
`
`

`

`DirectStream’s history, its business has focused on building and improving HPC
`
`systems.2 It was at the forefront of creating new, cost-effective HPC systems
`
`through the novel designs and techniques for incorporating FPGAs into the
`
`architecture of an HPC computer system. DirectStream has received more than 25
`
`patents related to its innovative use of FPGAs within HPC systems, and
`
`DirectStream’s technology has been adopted by the U.S. government, industry, and
`
`universities.
`
`A. Need for Cost-Effective HPC
`The first true high-performance computer, or supercomputer, was developed
`
`by Semyour Cray in 1965—the CDC-6600. EX2136 at 58. In 1976, Cray also
`
`developed the first true vector supercomputer—the Cray-1. Id. Until the early
`
`1990s, vector supercomputers remained the gold standard. EX2166¶¶17-18.
`
`Vector-based HPC, however, used custom-built microprocessors relying on
`
`expensive technologies such as gallium arsenide and static memories. Id. Starting
`
`in the early 1990s, there was a push to replace vector-based HPC by using massively
`
`parallel system that could use commercial, off-the-shelf processors to replace the
`
`
`2 DirectStream is SRC Computers’s successor company and was assigned the patents
`
`at issue on May 21, 2019. For continuity, Patent Owner will use the name
`
`DirectStream unless otherwise noted.
`
`
`
`4
`
`

`

`more expensive custom processors. Id. at ¶19. This move away from vector systems
`
`was also driven by the goal of improving usability. Id. at ¶20. For example, in 1992
`
`the Department of Defense Advanced Research Projects Agency (“DARPA”)
`
`launched an HPC initiative to reduce both the development time and program
`
`execution time for supercomputers. Id.
`
`Off-the-shelf processors were attractive as
`
`their performance had
`
`exponentially improved, which is conventionally known as “Moore’s Law,” as
`
`illustrated below:
`
`
`
`5
`
`
`
`

`

`EX2138 at 36. Additionally, processor bandwidth, (the amount of work a computer
`
`system and its components can do over a period of time), and processor clock rate
`
`also improved exponentially:
`
`
`
`6
`
`

`

`EX2138 at 55. With improvements in microprocessor and the move away from
`
`
`
`7
`
`

`

`vector systems, some, such as DirectStream, began to review the potential for the
`
`use of FPGAs to improve HPC. EX2101¶29-30.
`
`The first commercially available FPGA was introduced in 1984. EX2138.
`
`Numerous companies developed FPGAs in the mid-to-late 1980s, but with the
`
`relative lack of speed couple with high costs, many FPGA companies “simply
`
`vanished.” EX2139 at 6; EX2166¶22. However, with improvements in chip
`
`manufacturing in the 1990s, FPGA costs decreased and performance increased as
`
`represented below:
`
`EX2139 at 2, 6. With improvements to FPGA technology, a market for FPGAs
`
`
`
`began to emerge:
`
`
`
`8
`
`

`

`
`
`Id. at 8. Additionally, during this time, DARPA made a number of announcements
`
`recognizing the potential benefits and challenges for FPGA-based computer
`
`systems. EX2166¶27.
`
`B. DirectStream’s HPC Advancements
`SRC Computers, Inc., now DirectStream, was founded in June 1996 by
`
`Seymour Cray and joined by Jon Huppenthal—a named inventor on each of the
`
`challenged Patents. EX2101¶¶1, 8. DirectStream was founded to improve HPC in
`
`a more cost-effective way. EX2101¶29. As part of DirectStream’s business model,
`
`it went in a unique direction and began exploring the use of FPGAs to improve HPC.
`
`EX2101¶¶30-31. However, at the time, the technology did not exist to fully use the
`
`capabilities of FPGAs in HPC, so DirectStream began its work in with FPGAs that
`
`resulted in numerous patented inventions:
`
`
`
`9
`
`

`

`
`
`EX2109 at 2.
`
`1. Architectural Innovations
`DirectStream’s early innovations included development of its patented SRC-
`
`6 Hi-Bar® Crossbar Switch allowing its FPGA equal access to all memory in a
`
`symmetric multi-processor
`
`(SMP) computing system.
`
` EX2101¶¶32-36;
`
`EX2166¶¶23-26. It also developed a patented cache coherency system that
`
`overcame the restrictions imposed by Intel on third-party utilization of its memory
`
`architecture which limited DirectStream’s access to 25% of the chip’s overall
`
`capabilities. EX2101¶¶46-47. DirectStream also made architecture improvements
`10
`
`
`
`

`

`by using faster SRAM memory chips, opposed to the DRAM chips commonly used
`
`at that time, which improve performance and could be mounted in an array and
`
`stacked on top of each other—allowing variable amounts of memory based on
`
`customer preference. EX2101¶48.
`
`2. Invention of MAP Technology
`While DirectStream made improvements to HPC with its new architecture, it
`
`continued to look for ways to speed up performance. EX2101¶49. While on a
`
`customer trip to Milan, Italy in the summer of 1997 at McDonald’s, Huppenthal and
`
`his colleague, Paul Leskar, came up with an idea—using an FPGA to perform a pre-
`
`determined selection of memory-related operations. Id. This moment at
`
`McDonald’s evolved into U.S. Patent No. 6,076,152. Id.
`
`This invention alleviated one of the problems plaguing HPC known as “corner
`
`turn.” EX2101¶49. “Corner turn” related to a processor’s ability to quickly write
`
`data to a horizontal array, but its inefficiency in reading data in a vertical array. Id.
`
`By using a specialized FPGA, not only would DirectStream reduce “corner turn”
`
`delay, but also support vector-style processing functions used in HPC, but not
`
`efficiently supported by the microprocessors of the day. Id. DirectStream dubbed
`
`this invention MAP® based on the ’152 Patent description of “memory algorithm
`
`processors,” but today, MAP is simply a branding term used by DirectStream.
`
`EX2101¶58.
`
`
`
`11
`
`

`

`3. DirectStream’s Continued Improvement
`In April 1999, DirectStream’s first SRC-6 system contained a first-generation
`
`MAP and was tested at the Department of Energy’s Oak Ridge National Laboratory
`
`(“ORNL”). EX2101¶54. The ORNL testing continued for over a year, and
`
`DirectStream continued to internally develop improvements to the MAP. Id. When
`
`ORNL completed its testing, it confirmed what DirectStream already knew: the Intel
`
`25% limitation continued to hold back the performance of the MAP. EX2101¶56.
`
`However, ORNL confirmed that MAP could be very promising if the MAP could be
`
`programmed by HPC software engineers rather than coded by hardware engineers.
`
`EX2101¶56. At this time, software engineers sent specifications to FPGA hardware
`
`engineers, the FPGA would be coded, and then sent back to the software engineer to
`
`be tested and debugged. Id. While this practice was acceptable in the general
`
`computer industry, it was rejected by the HPC community—they were used to
`
`recompiling and debugging software in minutes and this process for FPGAs took
`
`hours or days. Id.
`
`Given the feedback from ORNL and its own testing, DirectStream continued
`
`to work on methods to overcome the Intel 25% limitation. EX2101¶59.
`
`DirectStream developed a modified chip set to allow third-party applications to have
`
`peer access to the processor’s memory bus. Id. DirectStream also developed an
`
`interface, known as SNAP™, to take advantage of the fastest, lowest latency
`
`
`
`12
`
`

`

`interconnect for a processor—its memory interface—due to the fact that memory
`
`interface circuit designs provided DirectStream’s FPGA vendor, Xilinx, could not
`
`act as part of the memory. EX2101¶¶60-61. DirectStream also developed an
`
`optimized polling method for its SNAP Interface. EX2101¶¶63-66.
`
`In 2001, based on feedback from ORNL, DirectStream formed a compiler
`
`group with the aim of developing C and Fortran compilers, which would allow
`
`application developers to program both microprocessors and the MAP processors in
`
`a software language the developers understood. EX2101¶69. This program was
`
`called Unified Executables and was delivered in 2002 as part of the first SRC-6E
`
`systems. Id. Three SRC-6e systems were purchased under NSA’s LUCITE program
`
`and installed at George Washington University, the Naval Post Graduate School, and
`
`the National Security Agency. EX2101¶70; EX2166¶¶28-29. These systems
`
`performed well and would lead to purchases of larger later model systems by all
`
`three customers. Id.
`
`Based on DirectStream’s success of its hardware and software tools, it
`
`continued to improve its technology with regards to specific needs in the
`
`marketplace—a process that often resulted in discovering continued ways to
`
`improve its technology. EX2101¶77. For example, it developed technology for use
`
`with websites and e-commerce that exponentially improved business’s abilities to
`
`retain customers through faster page loads and reduction of payment delays related
`
`
`
`13
`
`

`

`to credit card encryption technology. This resulted in U.S. Patent No. 6,434, 687.
`
`Id.
`
`DirectStream also continued to improve its technology allowing software
`
`programmers to write FPGA applications in language such as C and Fortran—
`
`something no other company could offer. EX2101¶78. This permitted a
`
`programmer to write standard, high-level language loops, which DirectStream’s
`
`technology turned into a collection of functional units, and use them in a way to
`
`stream data into and out of the units, as well as seamlessly between the loops. Id.
`
`This allowed the DirectStream FPGA to determine that an output variable from one
`
`loop is the input from another loop, and then implement this as a storage free,
`
`seamless connection between the two loops implemented on the FPGA—avoiding
`
`the decreased performance of other FPGAs that relied on a data registry, memory,
`
`or switch to move data. Id. No other FPGA system at the time had this HPC
`
`capability. Id. DirectStream’s technology also supported data streams which
`
`permitted concurrent and consumer loops to be interconnected rather than relying on
`
`the slower process of using SRAM, DRAM, or a data registered to connect the loops.
`
`EX2101¶79. The advances led to development of large scientific applications used
`
`in seismic processing, fluid flow analysis, structural analysis, and