`571-272-7822
`
`Paper 53
`Entered: March 1, 2022
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`INTEL CORPORATION and XILINX, INC.,1
`Petitioner,
`v.
`FG SRC LLC,
`Patent Owner.
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`
`
`
`
`
`
`
`
`
`
`
`
`Before KALYAN K. DESHPANDE, GREGG I. ANDERSON, and
`KARA L. SZPONDOWSKI, Administrative Patent Judges.
`SZPONDOWSKI, Administrative Patent Judge.
`
`
`
`JUDGMENT
`Final Written Decision
`Determining All Challenged Claims Unpatentable
`Denying Patent Owner’s Motion to Amend
`35 U.S.C. § 318(a)
`
`
`1 Xilinx, Inc. filed a motion for joinder and a petition in IPR2021-00633,
`which were granted, and, therefore, Xilinx, Inc. has been joined as petitioner
`in this proceeding.
`
`
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`INTRODUCTION
`I.
`We instituted an inter partes review of claims 1–19 of U.S. Patent
`7,149,867 B2 (Ex. 1001, “the ’867 patent”), in response to a Petition (Paper
`1, “Pet”) filed by Intel Corporation (“Petitioner”). Paper 13 (“Dec.”).
`During the trial, FG SRC LLC (“Patent Owner”) filed a Response (Paper 34,
`“PO Resp.”), Petitioner filed a Reply (Paper 40, “Reply”), and Patent Owner
`filed a Sur-reply (Paper 44, “Sur-reply”).
`Patent Owner also filed a Motion to Amend the claims of the ’867
`patent. Paper 26. After considering Petitioner’s Opposition to the Motion to
`Amend (Paper 36), we issued Preliminary Guidance on Patent Owner’s
`Motion (Paper 38). Patent Owner subsequently filed a Revised Motion to
`Amend the claims of the ’867 patent that includes proposed substitute claims
`20–38. Paper 41 (“Mot. Amend”). Petitioner opposed Patent Owner’s
`Revised Motion to Amend (Paper 45, “Opp. Amend”), Patent Owner replied
`(Paper 49, “Reply Amend”), and Petitioner filed a Sur-reply (Paper 50, “Sur-
`reply Amend”).
`An oral hearing was held on January 6, 2022, and a copy of the
`transcript was entered into the record. Paper 52 (“Tr.).
`We have jurisdiction under 35 U.S.C. § 6. This Decision is a Final
`Written Decision under 35 U.S.C. § 318(a) as to the patentability of the
`claims on which we instituted trial. Based on the complete record, Petitioner
`has shown, by a preponderance of the evidence, that claims 1–19 of the ’867
`patent are unpatentable. We also deny Patent Owner’s Revised Motion to
`Amend, because Patent Owner has not met its burden in asserting that
`proposed substitute claims 20–38 have written description support in the
`original application that issued as the ’867 patent.
`
`2
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`II. BACKGROUND
`Real Parties in Interest
`A.
`Petitioner identifies Intel Corporation as the sole real party in interest.
`Pet. 2. Patent Owner identifies FG SRC LLC as the sole real party in
`interest. Paper 4, 2.
`Related Matters
`B.
`The parties advise that the ’867 patent is the subject of the following
`district court litigations:
`FG SRC LLC v. Intel Corporation, 6:20-cv-00315-ADA (W.D. Tex.),
`filed April 24, 2020 (“the co-pending district court litigation”);
`FG SRC LLC v. Xilinx, Inc., 1:20-cv-00601-LPS (D. Del.), filed April
`30, 2020; and
`SRC Labs, LLC et al. v. Amazon Web Services, Inc., et al., 2:18-cv-
`00317-JLR (W.D. Wash.), filed February 26, 2018.
`Pet. 2; Paper 4, 2. Petitioner also advises that the ’867 patent was the
`subject of IPR2019-00103 (institution denied on May 10, 2019). Pet. 2.
`The ’867 Patent (Ex. 1001)
`C.
`The ’867 patent issued from Application No. 10/869,200 filed June
`16, 2004, and claims the benefit of Provisional Application No. 60/479,339,
`filed June 18, 2003. Ex. 1001, codes (21), (22), (60). The ’867 patent is
`titled “System and Method of Enhancing Efficiency and Utilization of
`Memory Bandwidth in Reconfigurable Hardware” and is generally directed
`to “enhancing the efficiency and utilization of memory bandwidth in
`reconfigurable hardware” and “implementing explicit memory hierarchies in
`reconfigurable processors that make efficient use of off-board, on-board,
`on-chip storage and available algorithm locality.” Id. at code (57), 1:15–24.
`
`3
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`According to the ’867 patent, there was a growing need to develop
`improved memory hierarchies that limited overhead of a memory hierarchy
`without also reducing bandwidth efficiency and utilization. Ex. 1001, 3:57–
`60. The ’867 patent describes a system including a memory hierarchy and a
`reconfigurable processor that includes a data prefetch unit. Id. at 4:4–10,
`5:60–62, 6:9–13, 7:34–48. The ’867 patent states that a “Reconfigurable
`Processor” is “a computing device that contains reconfigurable components
`such as FPGAs [(field programmable gate arrays)] and can, through
`reconfiguration, instantiate an algorithm as hardware.” Id. at 5:26–29. The
`’867 patent states that a “Data prefetch Unit” is “a functional unit [a set of
`logic that performs a specific operation] that moves data between members
`of a memory hierarchy [a collection of memories],” where such “movement
`may be as simple as a copy, or as complex as an indirect indexed strided
`copy into a unit stride memory.” Id. at 5:34–43.
`Figure 1 of the ’867 patent, reproduced below, shows a reconfigurable
`processor (RP) 100 of the claimed invention. Id. at 4:38–40.
`
`4
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`
`Figure 1 depicts a reconfigurable processor (RP) 100. Id. at 4:38–40.
`Figure 1 depicts reconfigurable processor 100, which “may be
`implemented using field programmable gate arrays (FPGAs) or other
`reconfigurable logic devices, that can be configured and reconfigured to
`contain functional units and interconnecting circuits, and a memory
`hierarchy comprising on-board memory banks 104, on-chip block RAM 106,
`registers wires, and a connection 108 to external memory.” Id. at 6:5–11. In
`addition, “[o]n-chip reconfigurable components 102 create memory
`structures such as registers, FIFOs, wires and arrays using block RAM.” Id.
`at 6:11–14. “Dual-ported memory 106 is shared between on-chip
`reconfigurable components 102.” Id. at 6:14–15. “The reconfigurable
`processor 100 also implements user-defined computational logic . . .
`constructed by programming an FPGA to implement a particular
`interconnection of computational functional units.” Id. at 6:15–19. “In a
`
`5
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`particular implementation, a number of RPs 100 are implemented within a
`memory subsystem of a conventional computer, such as on devices that are
`physically installed in dual inline memory module (DIMM) sockets of a
`computer.” Id. at 6:19–23. “In this manner the RPs 100 can be accessed by
`memory operations and so coexist well with a more conventional hardware
`platform.” Id. at 6:23–25. The ’867 patent explains that “[u]nlike
`conventional static hardware platforms . . . the memory hierarchy provided
`in a RP 100 is reconfigurable” and “through the use of data access units and
`associated memory hierarchy components, computational demands and
`memory bandwidth can be matched.” Id. at 7:17–22.
`One or more data prefetch units are used to improve the memory
`hierarchy and bandwidth efficiency and utilization. Id. at 3:58–60, 8:62–65.
`Fig. 4 of the ’867 patent, reproduced below, depicts a logic block 300 with
`an addition of a data prefetch unit 401. Id. at 4:44–46.
`
`6
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`
`Figure 4 illustrates a logic block 300 (a block composed of computational
`functional units capable of taking data and producing results with each clock
`pulse) with the addition of a data prefetch unit 401. Id. at 7:6–8, 7:34–35.
`
`Logic block 300 includes computational functional units
`(computational logic) 301, 302, and 303, a control, and data access
`functional units 403 that present data to computational logic 301, 302, and
`303. Id. at 7:25–48, Fig. 4. Data prefetch unit 401 moves data from one
`member of the memory hierarchy 305 to another 308 (a block RAM
`memory). Id. at 7:34–37, Fig. 4. Data prefetch unit 401 operates
`“independently of other functional units 301, 302, and 303 and can therefore
`operate prior to, in parallel with, or after computational logic.” Id. at 7:37–
`40. In addition, data prefetch unit 401 may be “operated independently of
`logic block 300 that uses prefetched data.” Id. at 7:45–48. Data prefetch
`
`7
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`unit 401 deposits data into the memory hierarchy, where computational logic
`301, 302, and 303 can access it through data access units. Id. at 7:42–44.
`The ’867 patent explains:
`An important feature of the present invention is that many
`types of data prefetch units can be defined so that the
`prefetch hardware can be configured to conform to the
`needs of the algorithms currently implemented by the
`computational logic. The specific characteristics of the
`prefetch can be matched with
`the needs of
`the
`computational logic and the format and location of data in
`the memory hierarchy.
`Id. at 7:49–55. The ’867 patent provides examples of configuring a data
`prefetch unit depending on the needs of the computational logic. Id. at 7:52–
`62, 8:3–21, Figs. 9A–9B.
`Illustrative Claims
`D.
`Among the challenged claims, claims 1, 9, and 13 are independent.
`Independent claims 1, 9, and 13 are reproduced below, with brackets noting
`Petitioner’s identifiers.
`1. [preamble] A reconfigurable processor that instantiates
`an algorithm as hardware comprising:
`[1(a)] a first memory having a first characteristic memory
`bandwidth and/or memory utilization; and
`[1(b)] a data prefetch unit coupled to the first memory,
`[1(c)] wherein the data prefetch unit retrieves only computational
`data required by the algorithm from a second memory of second
`characteristic memory bandwidth and/or memory utilization and
`places the retrieved computational data in the first memory [1(d)]
`wherein the data prefetch unit operates independent of and in
`parallel with logic blocks using the computional [sic] data, and
`[1(e)] wherein at least the first memory and data prefetch unit are
`configured to conform to needs of the algorithm, and [1(f)] the
`data prefetch unit is configured to match format and location of
`data in the second memory.
`
`8
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`
`
`
`
`9. [preamble] A reconfigurable hardware system,
`comprising:
`[9(a)] a common memory; and
`[9(b)] one or more reconfigurable processors that can
`instantiate an algorithm as hardware coupled to the common
`memory, [9(c)] wherein at least one of the reconfigurable
`processors includes a data prefetch unit to read and write only
`data required for computations by the algorithm between the data
`prefetch unit and the common memory [9(d)] wherein the data
`prefetch unit operates independent of and in parallel with logic
`blocks using the computational data, and [9(e)] wherein the data
`prefetch unit is configured to conform to needs of the algorithm
`and [9(f)] match format and location of data in the common
`memory.
`
`13. [preamble] A method of transferring data comprising:
`[13(a)] transferring data between a memory and a data
`prefetch unit in a reconfigurable processor; and
`[13(b)] transferring the data between a computational unit
`and a data access unit, [13(c)] wherein the computational unit
`and the data access unit, and the data prefetch unit are configured
`to conform to needs of an algorithm implemented on the
`computational unit and transfer only data necessary for
`computations by the computational unit, and [13(d)] wherein the
`prefetch unit operates independent of and in parallel with the
`computational unit.
`Ex. 1001, 12:39–54; 13:13–26; 14:1–11.
`Evidence
`E.
`Petitioner relies on the following references (see Pet. 4–5).
`Reference Exhibit
`Patent/Printed Publication
`Zhang
`1003
`Xingbin Zhang et al., Architectural Adaptation
`of Application-Specific Locality Optimizations,
`published in the Proceedings of the International
`Conference on Computer Design - VLSI in
`Computers and Processors (IEEE, October 12–
`
`9
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`Gupta
`
`1004
`
`Chien
`
`1005
`
`
`
`15, 1997), 150–156
`Rajesh Gupta, Architectural Adaptation in
`AMRM Machines, Proceedings of the IEEE
`Computer Society Workshop on VLSI 2000
`(IEEE, April 27–28, 2000), 75–79
`Andrew A. Chien et al., MORPH: A System
`Architecture for Robust High Performance
`Using Customization (An NSF 100 TeraOps
`Point Design Study), Proceedings of Frontiers
`’96 – The Sixth Symposium on the Frontiers of
`Massively Parallel Computing (IEEE, October
`27–31, 1996), 336–345
`
`In addition, Petitioner relies on the Declarations of Rajesh K. Gupta,
`Ph.D. (Exs. 1010, 1030), Declarations of Jacob Robert Munford (Exs. 1012,
`1031), Declaration of Gordon MacPherson (Ex. 1027), Declaration of Eileen
`D. McCarrier (Ex. 1028), Declaration of Austin Schnell (Ex. 1029), and
`Declaration of Dr. Stanley Shanfield (Ex. 1006).
`Patent Owner relies on the Declaration of Dr. William Mangione-
`Smith (Ex. 2028) and Declaration of Ryan Kastner, Ph.D. (Ex. 2010).
`Deposition transcripts have been entered into the record for Dr. Gupta
`(Ex. 1039), Mr. MacPherson (Ex. 1040), Dr. Shanfield (Exs. 1043, 2029),
`and Dr. Mangione-Smith (Ex. 1044).
`
`10
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`Prior Art and Asserted Grounds
`F.
`Petitioner asserts that claims 1–19 are unpatentable on the following
`grounds (Pet. 5):
`
`Claims Challenged
`1, 2, 4–8, 13–19
`3, 9–12
`
`35 U.S.C.
`§2
`
`103
`103
`
`References
`Zhang, Gupta
`Zhang, Gupta, Chien
`
`
`
`III. ANALYSIS
`Legal Standards
`A.
`A claim is unpatentable under 35 U.S.C. § 103(a) if “the differences
`between the subject matter sought to be patented and the prior art are such
`that the subject matter as a whole would have been obvious at the time the
`invention was made to a person having ordinary skill in the art to which said
`subject matter pertains.” KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 406
`(2007). The question of obviousness is resolved on the basis of underlying
`factual determinations, including: (1) the scope and content of the prior art;
`(2) any differences between the claimed subject matter and the prior art; (3)
`the level of skill in the art; and (4) objective evidence of nonobviousness,
`i.e., secondary considerations. See Graham v. John Deere Co., 383 U.S. 1,
`17–18 (1966).
`A patent claim “is not proved obvious merely by demonstrating that
`each of its elements was, independently, known in the prior art.” KSR, 550
`
`
`2 The Leahy-Smith America Invents Act (“AIA”), Pub. L. No. 112-29, 125
`Stat. 284 (2011), amended 35 U.S.C. § 103, effective March 16, 2013.
`Because the application from which the ‘867 patent issued was filed before
`this date, the pre-AIA version of § 103 applies.
`
`
`
`11
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`U.S. at 418. An obviousness determination requires finding “both ‘that a
`skilled artisan would have been motivated to combine the teachings of the
`prior art references to achieve the claimed invention, and that the skilled
`artisan would have had a reasonable expectation of success in doing so.’”
`Intelligent Bio-Sys., Inc. v. Illumina Cambridge Ltd., 821 F.3d 1359,
`1367–68 (Fed. Cir. 2016) (citation omitted); see KSR, 550 U.S. at 418.
`Further, an assertion of obviousness “cannot be sustained by mere
`conclusory statements; instead, there must be some articulated reasoning
`with some rational underpinning to support the legal conclusion of
`obviousness.” KSR, 550 U.S. at 418; In re NuVasive, Inc., 842 F.3d 1376,
`1383 (Fed. Cir. 2016) (a finding of a motivation to combine “must be
`supported by a ‘reasoned explanation’” (citation omitted)).
`Level of Ordinary Skill in the Art
`B.
`Petitioner asserts a person of ordinary skill in the art “would have had
`an undergraduate degree in electrical engineering or related field with at
`least three years of experience in computer processor architecture and
`FPGAs, a master’s degree with two or more years of experience in those
`fields, or an equivalent combination of education and experience.” Pet. 14
`(citing Ex. 1006 ¶ 67).
`Patent Owner does not dispute Petitioner’s proposed level of skill.
`See generally PO Resp.
`We find Petitioner’s proposal is consistent with the level of ordinary
`skill in the art reflected by the prior art of record, and, therefore, adopt
`Petitioner’s proposed level of ordinary skill in the art for purposes of this
`Decision. See Okajima v. Bourdeau, 261 F.3d 1350, 1355 (Fed. Cir. 2001).
`
`12
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`C. Claim Construction
`We construe each claim “in accordance with the ordinary and
`customary meaning of such claim as understood by one of ordinary skill in
`the art and the prosecution history pertaining to the patent,” the same
`standard used to construe the claim in a civil action. 37 C.F.R. § 42.100(b)
`(2020).
`Agreed Constructions
`1.
`The parties agree to the construction of the following terms:
`Claim term
`Proposed Construction
`Reconfigurable
`a computing device that contains reconfigurable
`Processor
`components such as FPGAs and can, through
`reconfiguration, instantiate an algorithm as hardware
`Data Prefetch Unit a functional unit that moves data between members of
`a memory hierarchy. The movement may be as
`simple as a copy, or as complex as an indirect
`indexed strided copy into a unit stride memory
`a functional unit that accesses a component of a
`memory hierarchy, and delivers data directly to the
`computational logic
`a set of logic that performs a specific operation. The
`operation may for example be arithmetic, logical,
`control, or data movement. Functional units are used
`as building blocks of reconfigurable logic
`a collection of memories
`
`Data Access Unit
`
`Functional Unit3
`
`Memory
`Hierarchy4
`
`
`
`
`3 Petitioner states that although not directly recited in the claims, this term is
`used in the ’867 patent’s definition of “data prefetch unit” and “data access
`unit.” Pet. 15; Ex. 1001, 5:40–46.
`4 Petitioner states although not directly recited in the claims, this term is
`used in the ’867 patent’s definition of “data prefetch unit” and “data access
`unit.” Pet. 15; Ex. 1001, 5:40–46.
`
`13
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`Pet. 14–16 (citing Ex. 1001; Ex. 1006); PO Resp. 29–30. We adopt these
`agreed constructions for purposes of this Decision.
`Patent Owner also states that in the co-pending district court litigation,
`the parties have agreed to the following constructions:
`Claim term
`Agreed Construction
`(Preamble) A
`Preamble is limiting
`reconfigurable processor
`that instantiates an
`algorithm as hardware
`Common Memory
`
`Computational Unit
`
`the Data Prefetch Unit
`Receives Processed Data
`Configured To Conform to
`Needs of The Algorithm
`Reconfigurable Logic
`
`an external memory shared by processors in
`a multiprocessor system
`a functional unit of a reconfigurable
`processor that performs a computation
`the data prefetch unit receives the results of
`the algorithm
`configured in reconfigurable logic to
`conform to the needs of the algorithm
`reconfigurable logic is composed of an
`interconnection of functional units, control,
`and storage that implements an algorithm
`and can be loaded into a Reconfigurable
`Processor
`
`
`PO Resp. 30. We further adopt these agreed constructions for purposes of
`this Decision.
`Patent Owner’s Proposed Constructions
`2.
`Patent Owner proposes construction for two terms.
`“retrieves only computational data required by the
`a)
`algorithm from a second memory . . . and places the
`retrieved computational data in the first memory”
`(limitation 1(c))
`Patent Owner argues that this limitation should be construed as
`“retrieves from a second memory that computational data which is required
`by the algorithm and no other computational data … and places the
`
`14
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`retrieved computational data in the first memory.” PO Resp. 31–32
`(emphasis added).
`Patent Owner contends that “[t]he plain meaning requires that no
`superfluous computational data is transferred to the first memory.” Id. at 31.
`In support of its proposed construction, Patent Owner identifies column 9,
`lines 1–5 of the ’867 patent, which states “an important feature of the
`present invention is the ability to implement various kinds or styles of
`prefetch units to meet the needs of a particular algorithm being implemented
`by computational elements.” Id. Patent Owner also relies on column 9,
`lines 8–10, which states “in most cases the function being implemented by
`components 301 would change and therefore alter the decision as to which
`prefetch strategy is most appropriate.” Id. at 31–32. Patent Owner also
`relies on column 9, lines 35–40, which states “[g]ains are made by
`delivering only requested data from transfer buffer 1305 (not the remainder
`of a data block as in cache line oriented systems) by eliminating the need to
`transfer an index array either to the processor or to the memory controller.”
`Id. at 32.
`Petitioner contends that Patent Owner does not rely on this
`construction to overcome the prior art, and Patent Owner’s expert does not
`rely on (or even recite) this construction. Reply 2 (citing PO Resp. 40–44;
`Ex. 1044, 28:18–29:5, 32:10–14, 107:19–108:7). Petitioner also argues that
`Patent Owner’s proposed construction is unsupported by intrinsic evidence.
`Id. at 3–4.
`Patent Owner has not explained the impact of this proposed claim
`construction on its arguments supporting patentability of the claims. We
`disagree with Patent Owner that its proposed construction is the “plain and
`ordinary meaning,” because the claim language recites “retrieves only
`
`15
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`computational data required by the algorithm from a second memory,” and
`does not require the additional limitation “and no other computational data.”
`Under these circumstances, we decline to rewrite the claim language to
`include additional words that are not present in the claim, e.g., “and no other
`computational data.” See Hoganas AB v. Dresser Indus., Inc., 9 F.3d 948,
`950 (Fed. Cir. 1993) (quoting E.I. Du Pont de Nemours & Co. v. Phillips
`Petroleum Co., 849 F.2d 1430, 1433 (Fed. Cir. 1988)) (“It is improper for a
`court to add ‘extraneous’ limitations to a claim, that is, limitations added
`‘wholly apart from any need to interpret what the patentee meant by
`particular words or phrases in the claim.’”).
` Moreover, in light of the parties’ arguments, we determine that we
`need not expressly construe this term. See Nidec Motor Corp. v. Zhongshan
`Broad Ocean Motor Co., 868 F.3d 1013, 1017 (Fed. Cir. 2017) (“[W]e need
`only construe terms ‘that are in controversy, and only to the extent necessary
`to resolve the controversy.’” (quoting Vivid Techs., Inc. v. Am. Sci. & Eng’g,
`Inc., 200 F.3d 795, 803 (Fed. Cir. 1999))). We agree with Petitioner that
`Patent Owner does not rely on this construction to overcome the prior art;
`for example, Dr. Mangione-Smith testified that he had not been given or
`applied any particular claim construction. See Ex. 1044, 28:18–32:14;
`107:19–109:4; see PO Resp. 40–44; Ex. 2028 ¶ 74.
`“read and write only data required for
`b)
`computations by the algorithm between the data prefetch
`unit and the common memory” (claim 9)
`Patent Owner proposes that this term should be construed as “read,
`using the data prefetch unit, only data required for computations by the
`algorithm from common memory and write, using the data prefetch unit,
`only data required for computations by the algorithm.” PO Resp. 33. In
`
`16
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`support of its construction, Patent Owner states that “[t]his term should be
`accorded its plain and ordinary meaning.” Id.
`Petitioner contends that Patent Owner does not rely on this
`construction to overcome the prior art, and Patent Owner’s expert does not
`rely on (or even recite) this construction. Reply 2 (citing PO Resp. 40–44;
`Ex. 1044, 28:18–29:5, 32:10–14, 107:19–108:7). Petitioner also argues that
`Patent Owner’s proposed construction is unsupported by intrinsic evidence.
`Id. at 3–4.
`Patent Owner has not explained the impact of this proposed claim
`construction on its arguments supporting patentability of the claims.
`Moreover, Patent Owner does not explain why we should adopt this
`proposed construction, or provide any supporting intrinsic evidence.
`Moreover, given that Patent Owner relies on the same arguments for claim 9
`as it presented for claim 1 (which does not recite this limitation), we agree
`with Petitioner that Patent Owner does not rely on this construction in order
`to overcome the prior art. See PO Resp. 53–55; see, e.g., Ex. 1044, 28:18–
`32:14 (Dr. Mangione-Smith testifying that he had not been given or applied
`any particular claim construction). Accordingly, we determine that we need
`not expressly construe this term to resolve the parties’ dispute.
`The Asserted Prior Art References
`D.
`1.
`Zhang (Ex. 1003)
`Zhang is a paper published by the Institute of Electrical and
`Electronics Engineers, Inc. (hereafter “IEEE”) as part of the Proceedings of
`the International Conference on Computer Design - VLSI in Computers and
`Processors. Ex. 1003, 1–2, 4. 5 Zhang describes “a machine architecture
`
`
`5 Citations to Exhibit 1003 are to the page numbering provided by Petitioner.
`
`17
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`that integrates programmable logic into key components of the system with
`the goal of customizing architectural mechanisms and policies to match an
`application,” using application-specific hardware assists. Id. at 12
`(Abstract). Zhang “demonstrate[s] that application-specific hardware assists
`and policies can provide substantial improvements in performance on a per
`application basis.” Id. Zhang’s architecture “integrates small blocks of
`programmable logic into key elements of a baseline architecture, including
`processing elements, components of the memory hierarchy, and the scalable
`interconnect, to provide architectural adaptation—the customization of
`architectural mechanisms and policies to match an application.” Id. at 13.
`Zhang explains that architectural adaptation provides mechanisms for
`application-specific hardware assists to overcome rigid architectural choices
`that do not work well across different applications, as “integration of
`programmable logic with memory components enables application-specific
`locality optimizations.” Id. at 13–14.
`Zhang’s architecture is depicted in Figure 2 below:
`
`18
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`
`Figure 2 of Zhang depicts programmable logic integrated with CPU,
`cache, network interface, and memory.
`
`
`Zhang presents “two case studies of architectural adaption using
`application-specific knowledge to enhance latency tolerance and efficiently
`utilize network bisection on multiprocessors.” Ex. 1003, 14. The first case
`study uses architectural adaptation for prefetching and exploits application
`access pattern information. Id. at 15. Figure 4 of Zhang, reproduced below,
`depicts a prefetcher implementation for Zhang’s first case study, using
`programmable logic integrated with the L1 cache. Id.
`
`19
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`
`Figure 4 shows a prefetcher implementation using programmable logic
`integrated with the L1 cache.
`The prefetcher in Figure 4 requires two pieces of application-specific
`information: address ranges and memory layout of the target data structures.
`Id. at 15. The address range, which is application dependent, is needed to
`indicate memory bounds where prefetching is likely to be useful. Id.
`Prefetching can be enabled or disabled, and is triggered only by read misses.
`Id. Once the prefetcher is enabled, it determines what and when to prefetch
`by checking virtual addresses of cache lookups to check whether a matrix
`element is being accessed. Id. In one example, records spanning multiple
`cache lines are targeted to “prefetch[] all fields of a matrix element structure
`whenever some field of the element is accessed.” Id. Because each matrix
`element (which is padded to 64 bytes) spans two cache lines (in a cache with
`cache line size of 32 bytes), the prefetcher generates an additional L2 cache
`lookup address from the given physical address that prefetches the other
`cache line not yet referenced. Id. In a second example, particular pointer
`
`20
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`fields (those likely to be traversed when their parent structures are accessed)
`are targeted. Id. at 15–16. For example, in a sparse matrix-vector multiply,
`the record pointed to by the nextRow field is accessed close in time with the
`current matrix element. Id. A prefetcher generates an additional address
`after the initial cache miss is satisfied using the nextRow pointer value
`embedded in the data returned by the L2 cache. Id. at 16.
`Zhang’s second case study uses a sparse matrix-matrix multiply
`routine to show architectural adaptation that improves data reuse and
`reduces data traffic between the memory unit and the processor. Id. at 15–
`16. “The architectural customization aims to send only used fields of matrix
`elements during a given computation to reduce bandwidth requirement using
`dynamic scatter and gather.” Id. at 16. “The two main ideas are prefetching
`of whole rows or columns using pointer chasing in the memory module and
`packing/gathering of only the used fields of the matrix element structure.”
`Id. Figure 5 of Zhang, reproduced below, illustrates an architecture
`including a cache and a main memory module, and containing two units of
`logic, an address translation logic and a gather logic.
`
`21
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`
`
`Figure 5 shows a scatter and gather logic using two units of logic, an address
`translation logic and a gather logic. Id. at 16.
`
`
`
`Gupta (Ex. 1004)
`2.
`Gupta is a paper published by IEEE as part of the Proceedings of the
`IEEE Computer Society Workshop on VLSI 2000. Ex. 1004, 1, 3, 6. 6 Gupta
`describes an Adaptive Memory Reconfiguration Management (AMRM)
`prototype architecture that implements a board-level prototype designed to
`“simulate a range of memory hierarchies for applications running on a host
`processor” and “supports configurability of [a] cache memory via an on-
`board FPGA-based memory controller.” Id. at 8–10. The goals of Gupta’s
`AMRM prototype are that “it be adaptable to many different memory
`hierarchy architectures,” “be useful for running real time program execution
`or even memory simulations,” and demonstrate “a specific mechanism for
`
`
`6 Citations to Exhibit 1004 are to the page numbering provided by Petitioner.
`
`22
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`latency management . . . to provide significant performance boost for the
`class of applications characterized by frequent accesses to linked data
`structures scattered in the physical memory.” Id. at 9–10.
`Figure 1 of Gupta, reproduced below, illustrates an AMRM prototype
`board.
`
`
`Figure 1 illustrates the main components of an AMRM prototype board.
`
`As shown in Figure 1, an AMRM prototype board includes a general
`3-level memory hierarchy plus support for an AMRM ASIC chip
`implementing architectural assists within a CPU-L1 datapath. Id. at 10. The
`FPGAs on the board contain controllers for the SRAM, DRAM and L1
`cache. Id. The AMRM chip is positioned between the L1 cache and the rest
`
`23
`
`
`
`IPR2020-01449
`Patent 7,149,867 B2
`of the system and can be accessed in parallel with the L2 cache. Id. at 11.
`The AMRM chip can thus accept and supply data coming from or going to
`the L1 cache. Id. The AMRM chip may contain a write buffer or a prefetch
`unit to access L2, and also has access to the memory interface and can
`prefetch from memory. Id.
`Chien (Ex. 1005)
`3.
`Chien is a paper published by the IEEE Computer Society Press as
`part of the Proceedings of Frontiers ‘96–The Sixth Symposium on the
`Frontiers of Massively Parallel Computing. Ex. 1005, 1–2, 5. 7 Chien
`describes a design and architecture of a MultiprocessOr with Reconfigurable
`Parallel Hardware (MORPH) that “uses reconfigurable logic blocks
`integrated with the system core to control policies, interactions, and
`interconnections” and has “configurability [th