`571-272-7822
`
`
`
`
`Paper No. 49
`Date Entered: June 28, 2016
`
`
`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`____________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`____________
`
`LG ELECTRONICS, INC.,
`Petitioner,
`
`v.
`
`ATI TECHNOLOGIES ULC,
`Patent Owner
`____________
`
`Case IPR2015-00326
`Patent 6,897,871 B1
` ____________
`
`
`
`Before JONI Y. CHANG, BRIAN J. McNAMARA, and
`JAMES B. ARPIN, Administrative Patent Judges.
`
`McNAMARA, Administrative Patent Judge.
`
`
`
`
`FINAL WRITTEN DECISION
`35 U.S.C. § 318(a) and
` 37 C.F.R. § 42.73
`
`
`LG Ex. 1007, pg 1
`
`LG Ex. 1007
`LG v. ATI
`IPR2017-01225
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`
`BACKGROUND
`On July 10, 2015, we instituted an inter partes review of claims 1–3, 5, 6, 8–
`11, 13, 15, 17, 18, and 20 (“the challenged claims”) of U.S. Patent No. 6,897,871
`B1 (Ex. 1001, “the ’871 Patent”). Paper 13 (“Dec. to Inst.”). ATI Technologies
`ULC (“Patent Owner”) filed a redacted and an un-redacted Patent Owner Response
`and a Motion to Seal. Papers 21, 20, and 19, respectively. LG Electronics, Inc.
`(“Petitioner”) filed an Opposition to Patent Owner’s Motion to Seal (Paper 25),
`Petitioner’s own Motion to Seal (Paper 27), an un-redacted Reply, and a redacted
`Reply (Papers 28 and 29, respectively).1 We authorized Patent Owner to file a Sur-
`Reply. Paper 32. Patent Owner filed duplicate Sur-Replies on February 9, 2016.
`Papers 36 and 37. Petitioner also filed a Motion to Exclude (Paper 41), which
`Patent Owner opposed (Paper 42), to which Petitioner replied (Paper 44). An oral
`hearing was conducted on April 6, 2016, and a transcript entered into the record.
`Paper 48 (“Tr.”).
`We have jurisdiction under 35 U.S.C. § 6(c). This Final Written Decision is
`issued pursuant to 35 U.S.C. § 318(a). We base our decision on the preponderance
`of the evidence. 35 U.S.C. § 316(e); 37 C.F.R. § 42.1(d). Having reviewed the
`arguments of the parties and the supporting evidence, we conclude that challenged
`claims 1–3, 5, 6, 8–11, 13, 15, 17, and 18 are unpatentable. Petitioner has not
`demonstrated that claim 20 is unpatentable.
`
`
`
`
`
`
`1 Unless otherwise noted, all references herein are to Petitioner’s un-redacted
`Reply, Paper 28 (“Reply”).
`
`
`
`2
`
`LG Ex. 1007, pg 2
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`
`BACKGROUND AND RELATED PROCEEDINGS
`In this proceeding, we instituted inter partes review on the following
`grounds:
`
`Claims 1, 2, 5, 8, 10–11, 13, and 15 as anticipated under 35 U.S.C.
`§ 102(e) by Lindholm;
`Claims 3 and 6 as obvious under 35 U.S.C. § 103 over the
`combination of Lindholm and Open GL;
`Claims 9, 17, and 18 as obvious under 35 U.S.C. § 103 over the
`combination of Lindholm and Kizhepat;
`Claim 20 as obvious under 35 U.S.C. § 103 over the combination of
`Lindholm and Kurihara;
`Claim 15 as obvious under 35 U.S.C. § 103 over Rich; and
`Claim 20 as obvious under 35 U.S.C. § 103 over the combination of
`Rich and Kurihara.
`During the oral hearing, Patent Owner conceded that, if it failed to antedate
`Lindholm, claims 1, 2, 5, 8, 10–11, 13, and 15 are unpatentable as anticipated by
`Lindholm, claims 3 and 6 are unpatentable as obvious under 35 U.S.C. § 103 over
`the combination of Lindholm and Open GL, and claims 9, 17, and 18 are
`unpatentable as obvious under 35 U.S.C. § 103 over the combination of Lindholm
`and Kizhepat. Ex. 2126, slide 2; Tr. 26 (“[T]he only basis for patentability with
`respect to grounds 1 through 3 is the antedating of Lindholm.”).
`All of the members of the panel in this proceeding participated in the oral
`hearing in related case IPR2015-00325, during which Patent Owner’s arguments
`concerning antedating of Lindholm were heard. Tr. 5. Patent Owner relies the
`same evidence and substantially the same arguments in the present review and in
`IPR2015-00325 in support of its efforts to antedate Lindholm. Compare, e.g.,
`
`
`
`3
`
`LG Ex. 1007, pg 3
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`IPR2015-00326, Paper 20, at v–x, 20–32 with IPR2015-00325, Paper 21, at v–x,
`15–29. In the Final Written Decision in IPR2015-00325, the panel determined that
`Patent Owner had not antedated Lindholm. LG Elecs., Inc. v. ATI Techs. ULC,
`Case IPR2015-00325, slip op. at 12–53 (PTAB April 14, 2016) (Paper 62). In
`view of the determination that Patent Owner has not antedated Lindholm, and in
`view of Patent Owner’s concession, we conclude that claims 1, 2, 5, 8, 10–11, 13,
`and 15 are unpatentable as anticipated by Lindholm, claims 3 and 6 are
`unpatentable as obvious under 35 U.S.C. § 103 over the combination of Lindholm
`and Open GL, and claims 9, 17, and 18 are unpatentable as obvious under 35
`U.S.C. § 103 over the combination of Lindholm and Kizhepat.
`The only matters remaining before this panel are whether claim 15 is
`obvious over Rich and whether claim 20 is obvious over the combination of Rich
`and Kurihara and the combination of Lindholm and Kurihara.
`THE ’871 PATENT
`In computer graphics systems, a three-dimensional shape is represented by
`collection of simple polygons called “primitives.” Ex. 1001, col. 1, ll. 11–12.
`Primitives are formed by the interconnection of individual pixels. Id. at col. 1,
`ll. 15–17. Color and texture are applied to the individual pixels that comprise the
`shape based on their location within the primitive and the primitive’s orientation
`relative to the generated shape. Id. at col. 1, ll. 17–19.
`A three-dimensional shape represented by a wireframe collection of
`primitives is transformed into colored images by two graphics-processing
`operations: (i) vertex operations and (ii) pixel operations. Prelim. Resp. 2 (citing
`Ex. 1001, col. 1, ll. 11–59). To orient the wireframe model as desired, matrix
`transformations applied to vertices Vx, Vy, Vz of the primitives generate new
`vertices Vx′ Vy′, Vz′, which then are translated into pixels to generate a rendered
`
`
`
`4
`
`LG Ex. 1007, pg 4
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`object that can be displayed as a two-dimensional image. Id. at 3 (citing Ex. 1001,
`col. 1, ll. 36–49). Pixel operations performed on each pixel of the rendered object
`determine the pixel’s color and appearance. Id. (citing Ex. 1001, col. 1, ll. 49–53).
`Conventional graphics processors include “shaders” that specify how and
`with what corresponding attributes a final image is generated on a screen or other
`device. Ex. 1001, col. 1, ll. 24–27. Conventional graphics processors require both
`a vertex shader and a pixel shader to render an object. Id. at col. 1, ll. 61–62. A
`vertex shader accepts as inputs data representing the vertices Vx, Vy, Vz, applies
`the matrix transformation, and provides angularly-oriented vertices Vx′, Vy′, Vz′.
`A pixel shader operating at the pixel level provides the color value associated with
`each pixel of the rendered object. Id. at col. 1, ll. 50–54.
`The ’871 Patent employs a “unified shader” capable of performing both
`vertex operations and pixel operations. Id. at col. 2, ll. 37–39. A multiplexer
`receives vertex data at a first input, and pixel parameter data and attribute data
`from a rasterization engine at a second input. Id. at col. 3, ll. 60–65. In response
`to a control signal, an arbiter circuit selects one of a plurality of inputs for
`processing and a shader coupled to the arbiter performs vertex operations or pixel
`operations based on the selected one of the inputs. Id. at col. 2, ll. 40–49. A
`control signal generated by the arbiter determines which of the two multiplexer
`inputs is provided to the unified shader. Id. at col 3, l. 65–col. 4, l. 1. According to
`an arbitration scheme implemented in the arbiter, vertex data at the first input is
`transmitted to the unified shader if there are sufficient resources available in the
`unified shader to operate on the vertex data; otherwise interpolated pixel data on
`the second multiplexer input is passed to the unified shader. Id. at col. 4, ll. 2–8.
`The unified shader includes a general purpose register for storing the
`plurality of selected inputs, a sequencer for storing logical and arithmetic
`
`
`
`5
`
`LG Ex. 1007, pg 5
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`instructions used to perform vertex and pixel manipulation operations, and a
`processor capable of executing both floating point arithmetic and logical
`operations on the selected inputs according to the instructions stored in the
`sequencer. Id. at col. 2, ll. 50–57. According to the arbitration scheme, if the
`general purpose register in the unified shader does not have sufficient space to
`store incoming vertex data, the arbiter does not transmit the vertex data. Id. at col.
`5, ll. 23–26. Instead, instructions for pixel calculation operations are carried out in
`the unified shader until enough registers become available to perform vertex
`operations. Id. at col. 5, ll. 27–33. When vertex data is transmitted to the unified
`shader, the resulting vertex data is transferred to a render back end block that
`converts the resulting vertex data to a format suitable for display. Id. at col. 5, ll.
`54–59.
`
`CHALLENGED CLAIMS
`Claims 15 and 20, reproduced below, are the only claims remaining to be
`addressed in this Decision:
`15. A unified shader, comprising:
`a general purpose register block for maintaining data;
`a processor unit; and
`a sequencer, coupled to the general purpose register block and the
`processor unit, the sequencer maintaining instructions operative to
`cause the processor unit to execute vertex calculation and pixel
`calculation operations on selected data maintained in the general
`purpose register block.
`
`
`20. The shader of claim 15, wherein the processor unit executes vertex
`calculations while the pixel calculations are still in progress.
`
`
`
`6
`
`LG Ex. 1007, pg 6
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`
`ANALYSIS OF PRIOR-ART CHALLENGES
`Introduction
`A claim is unpatentable under 35 U.S.C. § 103(a) if the differences between
`the claimed subject matter 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.” We resolve
`the question of obviousness 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;2 and (4)
`objective evidence of nonobviousness, i.e., secondary considerations. See Graham
`v. John Deere Co., 383 U.S. 1, 17–18 (1966). In an obviousness analysis, some
`reason must be shown as to why a person of ordinary skill would have combined or
`modified the prior art to achieve the patented invention. See Innogenetics, N.V. v.
`Abbott Labs., 512 F.3d 1363, 1374 (Fed. Cir. 2008). A reason to combine or
`modify the prior art may be found explicitly or implicitly in market forces; design
`incentives; the “interrelated teachings of multiple patents”; “any need or problem
`known in the field of endeavor at the time of invention and addressed by the
`patent”; and the background knowledge, creativity, and common sense of the
`person of ordinary skill. Perfect Web Techs., Inc. v. InfoUSA, Inc., 587 F.3d 1324,
`
`
`2 Petitioner and Patent Owner rely on substantially similar assessments of the level
`of ordinary skill in the relevant art. Ex. 1003 ¶ 41; Ex. 2003 ¶ 30. Although the
`parties are critical of the experience level of each other’s expert witness (PO Resp.
`49, Reply 25), each of Petitioner’s and Patent Owner’s declarants appear to exceed
`the qualifications required for a person of ordinary skill in the relevant art (see
`Ex. 1003 ¶¶ 3–8; Ex. 2003 ¶¶ 6–17), and we credit their testimony as to the
`definition of a person of ordinary skill in the art.
`7
`
`
`
`LG Ex. 1007, pg 7
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`132–9 (Fed. Cir. 2009) (quoting KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398, 418–
`21 (2007)).
`Claim 15 as Obvious over Rich
`Claim 15, which is drawn to a unified shader, does not recite an arbiter or a
`selection function as described above. Petitioner argues that Rich discloses a
`processor memory and registers, an arithmetic logic unit (ALU) as the processor
`unit, and a processing element array control unit as the sequencer. Pet. 47, 49–50.
`Petitioner notes Rich discloses that during pixel processing, the pixel data is
`retrieved from the memory 34 and processed. Id. at 47 (citing Ex. 1005, col. 33, ll.
`60–64). Petitioner acknowledges that Rich is silent regarding executing vertex
`calculation operations on the selected data maintained in the general purpose
`register block. Id. at 47; Ex. 1003, Declaration of Dr. Nader Bagherzadeh
`(“Bagherzadeh Decl.”) ¶ 216. Petitioner also notes that, in Rich, databases
`containing vertex data may be loaded into video memory via interfaces or may
`reside in locally accessible memory, and pixel data may be stored and retrieved
`from local processor memory 34. Pet. 48 (citing Ex. 1005, col. 9, ll. 1–12). Thus,
`Petitioner contends it would have been obvious to a person of ordinary skill that
`vertex data could also be stored in local processor memory 34 and retrieved for
`subsequent processing, for example by modifying Rich to maintain or temporarily
`store both primitive and vertex data from a database in processor memory 34 for
`local access by the ALU to process and transform primitives and their vertices. Id.
`at 48. As a reason for this modification, Petitioner cites faster and more efficient
`access resulting from temporary storage of data just before processing operations
`would reduce stall time associated with directly accessing data from external
`memory. Id.
`
`
`
`8
`
`LG Ex. 1007, pg 8
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`
`As to claim 15, Patent Owner contends that Petitioner’s Rich-based grounds
`fail because Rich would not work for its intended purpose if vertex data were
`stored in an on-chip memory. PO Resp. 33. Patent Owner emphasizes that Rich’s
`pipelined architecture stores vertex data (i.e., the transformed vertex data after the
`completion of vertex processing) in external memory to make the output of vertex-
`processing phase accessible to processing elements 32 during the pixel processing
`phase. Id. at 35. Patent Owner notes that, in Rich, processing element 32 that
`operates on a piece of data during the vertex processing phase is not necessarily the
`same processing element that operates on the data during the pixel-processing
`phase. Id.
`Patent Owner states that both Rich and the system recited in claim 15 must
`address input-routing because these systems have a single type of computational
`resource that operates on two different types of inputs—vertex and pixel inputs
`that compete for access to the computation resource. Id. at 37. According to
`Patent Owner, the systems recited in claim 15 and Rich address this issue in
`different ways. Id. at 38. Patent Owner argues that the claimed sequencer
`maintains instructions for both vertex and pixel operations that allow the processor
`to work with the general purpose register block to execute the appropriate
`instructions on the appropriate data at the appropriate time. Id. Patent Owner
`contrasts this approach to that of Rich, in which vertex and pixel operations are
`performed in discrete phases, such that vertex data stored in external, shared
`memory is accessible for routing to appropriate processing element 32 for pixel
`processing. Id. Patent Owner asserts that this difference eliminates the need in
`Rich for a general purpose register block or sequencer, as claimed, and that if Rich
`were modified to store vertex data locally (e.g., in on-chip memory 34), its
`processing elements would not be able to access the transformed data efficiently,
`
`
`
`9
`
`LG Ex. 1007, pg 9
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`meaning that Rich would not work for its intended purpose. Id. at 39–40. Patent
`Owner also argues that Rich “teaches away” because a person of ordinary skill
`would have understood that the Rich’s external memory is a shared memory for all
`processing elements, whereas its on-chip memory 34 is dedicated to one
`processing element. Id. at 41–42. Patent Owner also disputes the testimony of
`Petitioner’s expert that a person of ordinary skill would be motivated to store
`vertex data on chip for faster and more efficient access because Rich discloses that
`processing elements 32 have only a small amount of memory 34 in their own
`dedicated circuitry. Id. at 40–41.
`Claim 15 recites a “unified shader,” i.e., “a shader that is configured to
`perform both vertex and pixel operations” (Ex. 1001, col. 2, ll. 57–58), comprising
`a general purpose register block, a processor unit and sequencer coupled to both
`that maintains instructions to perform vertex calculation operations and pixel
`circulation operations on selected data maintained in the general purpose register
`block. The language of claim 15 does not limit the shader to one that follows any
`particular operational sequence; nor is claim 15 limited to a shader having an
`“on-chip” general purpose memory block.
`Petitioner acknowledges that Rich does not disclose explicitly vertex
`processing on data retrieved from processor memory 34. Bagherzadeh Decl. ¶ 216.
`However, Petitioner cites a disclosure in Rich that suggests making vertex data
`locally accessible, thereby recognizing that Rich could be operated using local
`on-chip memory to store vertex data. Reply 17 (citing Ex. 1005, col. 9, ll. 1–12).
`Rich discloses that the image generation system accesses the database of primitives
`defining the image to be generated and assigns the primitives to processing
`elements 32 of processing element array 30 and that “the database may reside in
`memory accessible through the host processor 22 or locally accessible.” Ex. 1005,
`
`
`
`10
`
`LG Ex. 1007, pg 10
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`col. 9, ll. 3–7, 10–12. Processing elements 32 transform the primitives defined in
`model space to x, y coordinates or screen coordinates. Id. at col. 9, ll. 17–21.
`Thus, we agree with Petitioner that Rich suggests making vertex data locally
`accessible. In addition, although Rich also discloses external memory circuitry
`may be necessitated when processing elements 32 have only a small amount of
`memory 34, we credit the testimony of Petitioner’s declarant, Dr. Bagherzadeh,
`that, at the time of the ’871 Patent, it would have been within the ability of one of
`ordinary skill to add more memory to the processing elements and that the use of
`on-chip memory was well known and commonly used in the art of data processing
`for fast access. Reply 20 (citing Bagherzadeh Decl. ¶ 223).
`Petitioner further argues that simply moving the vertex data to an on-chip
`memory would not render Rich’s system inoperable. Reply 17. As Petitioner
`notes, one of ordinary skill seeking to gain faster access to vertex data would have
`had reason to trade off pixel processing efficiency, without rendering the system
`inoperable for its intended purpose. Thus, we conclude that Petitioner has
`demonstrated that a person of ordinary skill would have had reason to modify Rich
`to arrive at the subject matter broadly recited in claim 15.
`Claim 20 as Obvious over Rich and Kurihara
`Claim 20 further limits claim 15, reciting that “the processor unit executes
`vertex calculations while the pixel calculations are still in progress.” Discussing
`the meaning of this limitation, Patent Owner explains that “[o]ne of the hallmarks
`of the unified shader is the ability to flexibly switch between vertex calculations
`and pixel calculations.” Tr. 24–25. By way of example, Patent Owner explains
`that in a ten instruction thread, one may execute five pixel instructions (1–5), stop
`and execute the vertex calculations from another thread, and sometime later
`execute the remaining instructions (6–10) of the original thread. Id. at 23. Thus,
`
`
`
`11
`
`LG Ex. 1007, pg 11
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`we understand Patent Owner to contend that claim 20 does not require the
`processor perform vertex and pixel calculations simultaneously, but that a
`processor have the ability to switch between vertex and pixel calculations as data is
`available. Id.; PO Resp. 45–46. Petitioner contends that this limitation would
`cover at least vertex and pixel calculations that are carried out simultaneously. Tr.
`18; Pet. 58–59. In the absence of further definition in claim 20, we agree that this
`limitation would include switching between calculations or calculations that take
`place simultaneously.
`Petitioner does not address whether it would have been obvious to one of
`ordinary skill to execute some instructions in one thread, pause execution of that
`thread, for example when data is not yet available, and execute instructions from
`another thread before returning to the first thread. Instead Petitioner contends that
`it would have been obvious in view of Rich and Kurihara to execute vertex and
`pixel calculations simultaneously. Pet. 58–59.
`Petitioner acknowledges that Rich does not disclose explicitly that both
`vertex and pixel processing occur simultaneously, but notes that, because Rich
`allows parallel processing and sharing of data among processing elements, Rich
`provides for simultaneous processing of both vertex and pixel data. Pet. 58.
`Petitioner argues that Kurihara discloses a graphics data parallel processing system
`that determines the type of graphics data to be transferred to first-in, first-out
`(“FIFO”) memories that send graphic data to respective processors for processing.
`Id. at 59. Noting that the system does not have to wait until all FIFO memories are
`emptied to transfer data, Petitioner argues that Kurihara allows for simultaneous
`processing of different types of graphics data. Id.
` As a basis for its combination with Kurihara, Petitioner states that the
`advantages of simultaneously processing pixel and vertex data were known in the
`
`
`
`12
`
`LG Ex. 1007, pg 12
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`art and that Rich discloses a number of parallel processing elements 32 allowing
`for simultaneous execution of various processing functions. Reply 21. Petitioner
`argues that it would have been known to one of ordinary skill to process pixel data
`in one of Rich’s processing units 32 at the same time as processing vertex data in
`another of Rich’s processing units 32, as taught by Kurihara. Id.
`Patent Owner contends that Kurihara and Rich are incompatible because
`Rich’s disclosure of vertex and pixel processing in discrete, sequential phases is
`the opposite of simultaneously performing different types of graphics operations,
`as disclosed in Kurihara. PO Resp. 43. Patent Owner also contends that even if
`the references could be combined, Kurihara does not suggest a single “processor
`unit” executes vertex calculations while pixel calculations are in progress. Id. a t
`44–45. Patent Owner argues that, in claim 20, the single processor unit executes a
`first type of graphics processing operation, e.g., vertex processing, while a second
`type of graphics processing, e.g., pixel processing, is in progress by stalling one
`type of processing in order to perform the other type of processing. Id. at 45.
`According to Patent Owner, none of Kurihara’s graphics processors can perform
`vertex calculations while pixel calculations are in progress. Id. at 46. Thus, Patent
`Owner’s arguments incorporate two assumptions: (1) that claim 20 extends only to
`diverting processing resources to processing a different type of data, while
`processing a first type of data; and (2) that the processing unit in claim 20 is a
`single processor.
`Petitioner responds that nothing in claim 20 limits the processing unit to a
`single processing element and that Kurihara’s group of graphics processors
`corresponds to the processing unit recited in claim 20. Reply 22. Claim 20 recites
`a processor unit, but does limit the processor unit to a single computational or
`processing element. Although Patent Owner’s declarant, Dr. Andrew Wolfe,
`
`
`
`13
`
`LG Ex. 1007, pg 13
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`asserts that Kurihara concerns how information is loaded into FIFO memories and
`does not disclose features of the processor (Ex. 2003, Declaration of Dr. Andrew
`Wolfe (“Wolfe Decl.”) ¶¶ 316–319), we note that claim 20 does not recite any
`specific features of the processor unit, other than that the processor unit executes
`vertex calculations while pixel calculations are still in progress.
`As discussed above, Petitioner contends that claim 20 includes
`simultaneously processing vertex and pixel calculations. Patent Owner contends,
`however, that Kurihara does not disclose that feature. PO Resp. 46. Kurihara
`claims a graphic data parallel processing apparatus having “a plurality of
`processors for simultaneously processing graphic data and defining a
`corresponding graphic image” and for displaying the graphic image. Ex. 1007, col.
`9, ll. 10–14 (claim 1). Claim 1 of Kurihara also recites “the graphic data for each
`graphic image being of first and second different types and of relatively and
`respectively larger and smaller quantities and requiring relatively and respectively
`smaller and greater processing times.” Id. at col. 9, ll. 14–18. Although claim 1 of
`Kurihara recites a plurality of processors for simultaneously processing graphic
`data and first and second types of graphic data requiring different processing times,
`the claim does not recite explicitly processing the first and second data types at the
`same time.
`Patent Owner contends that, in Kurihara, FIFO memories simultaneously
`transfer vertex or pixel data to graphics processors and each processor performs
`one type of graphics processing on the data type it receives. PO Resp. 44 (citing
`Ex. 1007, col. 4, ll. 56–65, col. 5, l. 37). Kurihara teaches that it was known in the
`art for graphics processors to process coordinate data independently by selecting an
`available FIFO memory according to half-full flags, but that, because attribute data
`must be transferred simultaneously through FIFO memories after all the FIFO half-
`
`
`
`14
`
`LG Ex. 1007, pg 14
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`full flags were set to 0, it was necessary to wait until the slowest graphics
`processor is released from the half-full state. Ex. 1007, col. 2, ll. 49–66, Fig. 5. As
`a result, other processors would have no data to process resulting in wasted time.
`Id. at col. 2, l. 67–col. 3, l. 2.
`Kurihara discloses that the graphics processors process data in parallel.
`Ex. 1007, col. 5, l. 37. However, Kurihara treats coordinate (vertex) and attribute
`(pixel) data differently. In Kurihara, controller 1 determines whether or not
`graphic data must be transferred to selected ones of the FIFO memories according
`to flags. Id. at col. 6, ll. 53–55. Controller 1 detects the type of data to be
`transferred and, based on flags set by data quantity detectors for each type of data,
`determines whether FIFO memories have free space to receive the coordinate or
`attribute data to be sent to the respective graphic processor. Id. at. col. 5, ll. 15–36;
`col. 8, ll. 19–26. Coordinate data is loaded into the FIFO memory if any one of the
`first flags is set to zero. Id. at col. 6, l. 56–61, col. 8, ll. 2–6, 31–34, Fig. 3. In
`contrast, attribute data is loaded into the FIFO only if all of the second flags are set
`to zero. Id. at col. 6, l. 61–col. 7, l. 3, col. 8, ll. 14–15, Fig. 3. In this way,
`Kurihara improves processing speed by reducing the suspension period in which
`no data are transferrable due to fluctuations in coordinate data (shape data). Id. at
`col. 8, l. 66–col. 9, l. 2. Separately processing coordinate data with the graphics
`processors, Kurihara minimizes processing periods in the graphics processing due
`to fluctuations in the quantities of coordinate data and quickly and simultaneously
`transfers attribute data to all of the graphic processors. Id. at col. 9, ll. 3–8.
`Petitioner’s declarant, Dr. Bagherzadeh, states that Kurihara discloses that
`the parallel processing system does not have to “wait until all of the FIFO
`memories 5 are emptied to transfer graphics data” and, based on this disclosure, he
`asserts “Kurihara’s system therefore allows for the simultaneous processing of
`
`
`
`15
`
`LG Ex. 1007, pg 15
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`different types of graphics data.” Bagherzadeh Decl. ¶¶ 287–288 (citing Ex. 1007,
`col. 4, ll. 56–62). However, the text cited by Dr. Bagherzadeh concerns
`“simultaneously transferring the attribute data with only a short wait time to the
`FIFO memories 5 because the system does not wait until all of the FIFO memories
`5 are emptied.” Ex. 1007, col. 4, ll. 56–60 (emphasis added). As discussed above,
`Kurihara discloses transferring the coordinate data to a FIFO memory where the
`flag indicates space is available, but “[i]f any one of the FIFO memories 5 has
`insufficient space to receive the attribute data, the controller suspends the write
`operation until every one of the FIFO memories 5 has sufficient free space to
`receive the attribute data.” Id. at col. 6, l. 67–col. 7, l. 3. Thus, although Kurihara
`discloses providing coordinate data to the FIFO memories as they become
`available, Kurihara does not disclose providing pixel data to the FIFO memories
`until they are all available. Kurihara further discloses that the attribute data is
`transferred from the FIFO memories to the graphics processors simultaneously. Id.
`at col 4, ll. 61–62, col. 9, ll. 5–8. The simultaneous transfer of attribute data to
`graphics the processors suggests that attribute processing occurs simultaneously,
`but does not provide any insight into whether coordinate processing and attribute
`processing occur simultaneously. Although the simultaneous transfer of attribute
`data to the processors does not necessarily preclude that vertex processing can
`proceed in one processor at the same time pixel processing takes place in another
`processor, Petitioner has not cited an adequate description of this feature in
`Kurihara.
` In consideration of the above, we are unable to conclude that the subject
`matter cited by Petitioner demonstrates that Kurihara discloses executing vertex
`calculations and pixel calculations simultaneously.
`
`
`
`
`16
`
`LG Ex. 1007, pg 16
`
`
`
`IPR2015-00326
`Patent 6,897,871 B1
`
`
`Claim 20 as Obvious over Lindholm and Kurihara
`As discussed above, in IPR2015-00325, we determined that claim 15 is
`unpatentable over Lindholm. Petitioner acknowledges that Lindholm does not
`disclose explicitly that both vertex and pixel processing occur simultaneously, and
`cites Kurihara as disclosing this feature. Pet. 36–37. For the reasons discussed
`above, Petitioner has not demonstrated that Kurihara discloses this limitation.
`Secondary Considerations
`Having determined that Petitioner has not established obviousness of claim
`20 based on the prior art cited in its challenges, we limit our analysis of Patent
`Owner’s secondary consideration arguments to claim 15. Patent Owner contends
`that the release of its Xenos chip, as the first commercially available unified
`shader, was a watershed event that overcame industry skepticism and resulted in a
`substantial change in the industry’s approach to the architecture of graphics
`products. PO Resp. 47. Petitioner’s Reply states that Patent Owner’s expert, Dr.
`Wolfe, testified that claim 15 recites a specific type of unified shader and that
`Patent Owner failed to explain what claimed features it alleges the industry has
`embraced. Reply 22–23.
`In its secondary consideration arguments, Patent Owner states only that the
`“hallmark of a unified shader is the ability to perform either vertex operations or
`pixel operations on inputs maintained in a shared repository.” PO Resp. 46. Dr.
`Wolfe’s testimony on behalf of Patent Owner discusses concerns that an
`architecture suitable for a pixel shader would result in hardware being idle most of
`the time during vertex shading, and that load balancing would be challenging in a
`unified shader. Wolfe Decl. ¶¶ 326–329. None of these features is recited in claim
`15, nor does Dr. Wolfe connect the