`WESTERN DISTRICT OF TEXAS
`AUSTIN DIVISION
`
`AQUILA INNOVATIONS, INC., a
`Delaware Corporation,
`
`Plaintiff,
`
`v.
`ADVANCED MICRO DEVICES, INC.,
`a Delaware corporation
`Defendant.
`
`No. 1:18-cv-554-LY
`
`
`
`§
`§
`§
`§
`§
`§
`§
`§
`§
`
`
`AQUILA INNOVATIONS, INC.’S
`PRELIMINARY INFRINGEMENT CONTENTIONS
`Pursuant to the Court’s Scheduling Order, D.I.23, Plaintiff Aquila
`
`
`
`Innovations, Inc. (“Aquila”) submits the following preliminary infringement
`
`contentions for U.S. Patents 6,239,614 (“’614 Patent”) and 6,895,519 (“’519 Patent”).
`
`These preliminary infringement contentions were prepared without the benefit of
`
`the Court’s claim construction or the parties’ exchange of constructions. Discovery
`
`has been stayed, and AMD has not produced any information concerning the
`
`Accused Products. Thus, this chart is based on publicly available evidence, and
`
`based upon information and reasonable belief in light of such evidence. As such,
`
`Aquila reserves the right to amend or supplement its contentions to address any
`
`issues arising from the Court’s constructions or to account for new information that
`
`becomes available.
`
`
`
`AMD EX1012
`U.S. Patent No. 6,895,519
`
`0001
`
`
`
`(a) Identification of asserted claims
`
`’614 Patent: Claims 1, 2
`
`’519 Patent: Claims 1, 2, 3, 5, 6, 7, 10
`
`(b) Identification of accused products
`’614 Patent Accused Products:
`
`
`
`
`
`Aquila contends that all AMD processor products containing power gate rings
`
`infringe each of the asserted claims of the ’614 Patent. This specifically includes but
`
`is not limited to processors with cores having microarchitectures belonging to the
`
`following families:
`
`o AMD 12h Llano Fusion APUs
`o AMD 15h Bulldozer APUs
`o AMD 15h Piledriver APUs
`o AMD 15h Excavator APUs
`AMD products belonging to each product family are identified in Exhibits C.1
`
`through C.4. The identification of specific products was prepared without the
`
`benefit of discovery from AMD and may not include OEM or custom processor
`
`products. Aquila reserves the right to amend or supplement its identification of
`
`accused products as AMD provides more information.
`
`’519 Patent Accused Products:
`
`Aquila contends that all AMD processor products capable of entering/exiting
`
`the CC1, PC1, and PC6 states infringe each of the asserted claims of the ’519 Patent.
`
`This specifically includes but is not limited to processors with cores having
`
`microarchitectures belonging to the following families:
`
`-2-
`
`0002
`
`
`
`o AMD Family 12h
`o AMD Family 14h
`o AMD Family 15h
`o AMD Family 16h
`o AMD Family 17h
`AMD products belonging to each product family are identified in Exhibits C.1
`
`through C.5. The identification of specific products was prepared without the
`
`benefit of discovery from AMD and may not include OEM or custom processor
`
`products. Aquila reserves the right to amend or supplement its identification of
`
`accused products as AMD provides more information.
`
`(c) Claim Charts
`
`Claim charts for each asserted claim corresponding to each representative
`
`accused product are contained in the exhibits below.
`
`
`
`
`
`
`
`’614 Patent: Exhibit A
`
`’519 Patent: Exhibit B
`
`(d) Doctrine of Equivalents
`
`Aquila contends that each limitation in each asserted claim is met literally.
`
`The Court has not construed the asserted claims, and AMD has not yet provided
`
`discovery on the accused products or provided non-infringement contentions. Aquila
`
`reserves the right to respond if AMD provides non-infringement contentions, which
`
`response may include doctrine of equivalents contentions.
`
`
`
`-3-
`
`0003
`
`
`
`(e) Identification of Priority Date
`
`
`
`’614 Patent: Each asserted claim of the ’144 Patent is entitled to a priority
`
`date as late as January 14, 1999.
`
`
`
`’519 Patent: Each asserted claim of the ’519 Patent is entitled to a priority
`
`date as late as February 25, 2002.
`
`
`
`Dated: February 13, 2019
`
`
`
`
`
`Respectfully submitted,
`
`
`
`
`
`
`/s/Jing H. Cherng
`Robert E. Freitas (admitted pro hac vice)
`Jing H. Cherng (admitted pro hac vice)
`FREITAS & WEINBERG LLP
`350 Marine Parkway, Suite 200
`Redwood Shores, CA 94065
`Telephone: (650) 593-6300
`rfreitas@fawlaw.com
`gcherng@fawlaw.com
`
`
`Henry B. Gonzalez III
`State Bar No. 00794952
`Jeffrie B. Lewis
`State Bar No. 24071785
`GONZALEZ, CHISCANO, ANGULO,
`& KASSON, PC
`9601 McAllister Freeway, Suite 401
`San Antonio, Texas 78216
`Tel: (210) 569-8500
`hbg@gcaklaw.com
`jlewis@gcaklaw.com
`
`Attorneys for Plaintiff
`Aquila Innovations, Inc.
`
`-4-
`
`0004
`
`
`
`I hereby certify that on this 13th day of February, 2019, a true and correct
`copy of the foregoing was forwarded to the following:
`Jennifer Librach Nall
`Kevin J. Meek
`Aashish Kapadia
`Puneet Kohli
`jennifer.nall@bakerbotts.com
`kevin.meek@bakerbotts.com
`aashish.kapadia@bakerbotts.com
`puneet.kohli@bakerbotts.com
`DLWiLAN_AMD_BakerBotts@BakerBotts.com
`BAKER BOTTS LLP
`
`
`
`
`
`
`
`
`
`
`CERTIFICATE OF SERVICE
`
`Jing H. Cherng
`Jing H. Cherng
`
`
`
`
`-5-
`
`0005
`
`
`
`Accused Product: AMD Family 12h Fusion Processors
`
`Exhibit A.1: Preliminary Infringement Contention Claim Chart for U.S. Patent 6.239.614
`
`These preliminary infringement contentions were prepared without the benefit of the Court’s claim construction or the parties’ exchangeof constructions.
`As of the date of these contentions, AMD has not produced any information concerning the Accused Products. Thus, this chart is based on publicly available
`evidence, and based upon information and reasonablebelief in light of such evidence. As such, Aquila reserves the right to amend or supplementits
`contentions to address any issues arising from the Court’s constructions or to account for new information that becomesavailable.
`
`A semiconductor
`integrated circuit device,
`
`To the extent the ortis a2 ere fs Accused Products are semiconductor integrated circuit devices.
`= —
`
`Suisse‘AMD’S “LLANO”FUSIONAPU, Hot Chips 23, 19th August 2011, page 6.
`
`Graphics SIMD
`Eee hY
`
`Thy 0) Fa]
`
`0 Controllers
`
`0006
`
`
`
`
`Limitation
`
`The Accused Products contain a plurality of first unit cells:
`
`a plurality of first unit
`cells each including a
`plurality of first MOS
`transistors, each of the
`first MOStransistors
`havinga first threshold
`
`voltage; Contention
`
`As is typical of multi-VT standard cell design methodologies and structures, the Llano Accused Products use a mix
`of low Vt (LVt), regular Vt (RVt), long channel regular Vt (LC-RVt) and long channel high Vt (LC-Hvt) standard
`cells.
`
`Graphics SIMD
`Nery]
`
`Display
`
`VO Controllers
`
`Source:AMD’S “LLANO”FUSIONAPU, Hot Chips 23, 19th August 2011, page6.
`
`0007
`
`
`
`
`
`ubMN. :
`
`Ce
`yt i
`
`The Accused Productscontain a plurality of first MOS transistors, each of the first MOStransistors havingafirst
`threshold voltage:
`
`0008
`
`
`
`
`
`JOTWANI ef al: AN 186-64 CORE IN 32 am SOLCMOS
`
`Contention
`
`a
`
`&
`
`(a)
`
`Post Swapping
`HVvt } Vt Types
`
`0% 20% 40% 60% 60%
`(a)
`=#45nm @&32nm
`
`PostSwapping
`
`(b)
`
`100%
`
`100%
`
`[ iStatic
`
`plurality of second MOS
`
`CustomMacros SND 26%
`StdCelis SD 25%
`Flops SB 17%
`Clock 38%
`s:.
`Gater @ 4%
`Static SD 2°.
`yi
`
`(a) Core device width histogram by Vt type. (b) Using Vi swaps to
`Fig. 10,
`reshape critical path timing opportunistically,
`
`| V. (b) Relative power improvement
`(a) TDP power distribution at
`Fig. IL.
`With respect to 45 nm generation core (normalized for performance}.
`
`Dynamic
`
`0009
`
`
`
`Contention
`.
`7
`eS
`
`Graphics SIMD
`
`Display
`
`| fe. Controllers
`
`Limitation
`
`transistors, each of the
`second MOStransistors
`having a second threshold
`
`
`
`deesbreeedOF68pte+ Graton
`tteestte
`HELL
`
`Array
`
`
`
`0010
`
`
`
`rcesLvt
`
`B
`
`Lvt
`
`uei
`
` Limitation
`
`Contention
`
`Modern multi-VT standard cell design methodologies and structures such as those used in the Llano Accused
`Products use a mix of low Vt (LVt), regular Vt (RVt), long channel regular Vt (LC-RVt) and long channel high Vt
`(LC-Hvt) standardcells.
`
`Lyt
`
`ecute compute-intensive code and will therefore be a high-performance device.
`
`DVFS Characteristics Modern computer chips are designed using multiple
`types of transistors, i.e. a mixture of low-, medium-, and high-threshold transis-
`tors, to target different design tradeoffs, e.g. high-performance vs. low power.
`Low-threshold voltage (Low-Vt) devices are used in timing-critical paths,
`but have high leakage power. High-threshold voltage (High-Vt) devices have
`
`0011
`
`
`
`The Accused Products contain a plurality of second MOStransistors, each of the second MOStransistors having a
`second threshold voltage:
`
`
`
`
`
`Limitation
`
`Contention
`
`Source: Scrbak et al., DVFS Space Exploration in Power Constrained Processing-in-Memory Systems, pp. 4-5.
`
`=z
`am 2s.
`aa iy
`ga.
`gi
`O4%
`
`aG
`
`ap:
`0% a 40% 60% 80%
`
`(a
`
`CustomMacros
`
`StdCells
`Flops
`Clock
`Repeater
`Gater
`Dynamic
`Static
`
`5
`
`Post S
`Swecring
`vn} Peet
`Vt Types
`
`i
`
`Post Swapping
`
`PreSwapping
`100
`125
`150
`
`75
`
`#45nm #32nm
`
`100%
`
`100%
`
`68%
`
`fi
`
`84%
`
`i
`
`Timing Slack (ps)
`
`Static
`
`Dynamic
`
`0012
`
`
`
`
`
`Limitation
`
`comprisedofsaid first and
`second unit cells laid in
`
`array form; Contention
`VO Controllers
`
`Graphics SIMD
`PNgeht
`
`Display
`
`0013
`
`
`
`
`
`Contention
`
`Modern multi-VT standard cell design methodologies and structures such as those used in the Llano Accused
`Products use a mix of low Vt (LVt), regular Vt (RVt), long channel regular Vt (LC-RVt) and long channel high Vt
`(LC-Hvt) standardcells.
`
`
`
`Lvt [svt
`
`DVFS Characteristics Modern computer chips are designed using multiple
`types of transistors, i.e. a mixture of low-, medium-, and high-threshold transis-
`tors, to target different design tradeoffs, e.g. high-performance vs. low power.
`Low-threshold voltage (Low-Vt) devices are used in timing-critical paths,
`
`0014
`
`
`
`
`
`Limitation
`
`Contention
`
`a power switch disposed
`aroundsaid unitcell
`array and comprisedof a
`plurality of third MOS
`transistors, each of the
`third MOStransistors
`having the second
`threshold voltage; and
`
`=
`Poe rereWwererrere
`939559935599395555
`aOetetanictl
`
`Virtual voltage spreads uniformly
`Bumps nearhot spots can exceed max limits
`
`Footeves
`ls
`
`low leakage power but are slower, and are typically used in circuits that are
`off the timing-critical paths. Medium-threshold voltage (Mid-Vt) devices offer
`a tradeoff between High-Vt and Low-Vt devices by having medium power re-
`quirements and medium delay. In general, low power chips are designed using a
`larger percentage of High-Vt devices and high-performance chips with a larger
`percentage of Mid-Vt and Low-Vt devices. ‘The host processor is assumed to ex-
`ecute compute-intensive code and will therefore be a high-performance device.
`Source: Scrbak et al., DVFS Space Exploration in Power Constrained Processing-in-Memory Systems,pp. 4-5.
`
`15 0) My Pn og Source: PRACTICAL POWER GATING AND DYNAMIC VOLTAGE/FREQUENCYSCALING by Stephen
`
`The Accused Products contain a power switch disposed aroundsaid unit cell array:
`LLANO CPU CORERING GATING
`WITH PACKAGE LAYER ASSIST
`cE
`eo er eee aeee
`ese es
`——
`©,sosane88
`
`”
`
`—susan
`
`a
`
`“ree
`
`vesfo
`
`0015
`
`
`
`Kosonocky, page 50.
`
`Contention
`
`Virtual eaecies uniformly
`
`Bumps near hot spots can exceed max limits
`
`The Accused Products contain a plurality of third MOStransistors:
`LLANO CPU CORE RING GATING
`WITH PACKAGE LAYER ASSIST
`
`
`
`CMOS,SooliespeaclngetCorefn32oemSO High Ryocancreatenoiseissues sl
`
`
`edt PRACT.TCALPOWER GATING AND DYNAMIC VOLTAGE/FREQUENCYSCALINGby Stephen
`
`Raw Jotwani, Sram Sundar:
`
`0016
`
`
`
`
`
`Limitation
`
`Contention
`
`Wedefined a low-power mode called core-level C6 (CC6) to
`allow core-level power gating [5] during periods of inactivity,
`The core is isolated fromthe supply during CC6 by a power-gate
`ring surrounding the CPU and L2 cachepair, allowing core level
`power down in a chip with multiple cores attached to a common
`power supply. The SOI process enables the gating of VSS (not
`
`VDD). constructing thdpower-gatewithregularVtaMOS}logic
`devices without the need for extra processing steps to reduce
`on-state resistance [6]. Fig. 12(a) describes the core operations
`controlled by the power management system for CC6 entry and
`exit sequences.
`Fig. 12(b) details the connections of the power-gate ring with
`respect to the core and the C4 bumps. In addition to two 16X
`M10 and MI! on-die metal layers, a low-impedance package
`layer connected to the die by C4 bumpsis dedicated for use as a
`virtual-groundlayer, eliminating the need for any ultra-thick sil-
`icon metallization layer [6]. The low-impedance package layer
`
`said unit cell array.
`
`VII. POWER GATING
`
`Source: An x86-64 Core in 32nm SOI CMOS, IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 46, NO. 1,
`JANUARY 2011, page 6.
`
`0017
`
`
`
`
`
`Contention
`
`LLANO CPU CORE RING GATING
`WITH PACKAGE LAYER ASSIST
`ss
`
`aesssoess.
`eseeeeease
`seenaneen
`
`SeesawSseanse#
`
`Virtual voltage spreads uniformly
`Bumps near hot spots can exceed max limits
`
`:
`
`
`
`Source: PRACTICAL POWER GATING AND DYNAMIC VOLTAGE/FREQUENCYSCALING by Stephen
`Kosonocky, page 50.
`
`MO—iSy&&&& 11.tt
`worst
`ov
`:
`;
`222eeeereWeeroeren
`A
`A
`4
`= (a
`.
`
`Ryss
`
`VSS,
`5
`Ravi Jotwani, Snram Sundaram, Stephen Kosonocky,
`Trop
`Alex Schaeter, Victor F. Andrade, Amy Novak,
`"
`‘
`i
`Samuel Naffziger, “An x66-64 Core in 32 nm SO!
`High Ryss can create noise issues
`culos’Jac
`201
`tet ee Cay ant pe 1 ee © ey Conny tneere dame 1" 2D‘ Beet Cet
`
`Res = 1m Q =
`
`.
`
`0018
`
`
`
`
`
`Limitation
`
`Contention
`
`Two major knobs have emerged for controlling power
`
`1. Dynamic Voltage and Frequency Scaling
`— Optimize performance for the application while it’s
`running
`2. Power Gating
`— Gate powerduring idle periods
`
`Each present unique challengesfor
`implementation and optimization
`
`
`
`
`
`Kosonocky, page 8.
`
`
`
`
`
`aoPee catersondDera met)|SechenteewaeFreceeecyDcahegtewews Pacant he wos
`
`
`
`Source: PRACTICAL POWER GATING AND DYNAMIC VOLTAGE/FREQUENCYSCALINGby Stephen
`
`0019
`
`
`
`Accused Products: AMD Family 15h Bulldozer/Piledriver APU Processors
`
`Exhibit A.2: Preliminary Infringement Contention Claim Chart for U.S. Patent 6,239,614
`
`These preliminary infringement contentions were prepared without the benefit of the Court’s claim construction or the parties’ exchange of constructions.
`As of the date of these contentions, AMD hasnot produced any information concerning the Accused Products. Thus, this chart is based on publicly available
`evidence, and based upon information and reasonablebelief in light of such evidence. As such, Aquila reserves the right to amend or supplementits
`contentions to address any issues arising from the Court’s constructionsor to account for new information that becomesavailable.
`
`Limitation
`
`
`| A-semiconductor integrated circuit
`device, comprising:
`
`Contention
`=
`| To the extent the preamble is a limitation, the Accused Products are a semiconductor integrated device.
`
`
`
`Vy
`
`sma) a) |
`
`0020
`
`
`
`The Accused Productdiscloses a
`
`plurality of first unit cells.
`
`Contention
`
`Limitation
`
`a plurality offirst unit cells each
`including a plurality of first MOS
`transistors, each of the first MOS
`transistors having a first threshold
`
`THE DIE | Photograph
`rr Corr
`
`es
`
`0021
`
`
`
`Limitation
`
`Contention
`
`THE DIE | Floorplan (315 mm7)
`
`
`
`
`Source: Sean White, HIGH-PERFORMANCE POWER-EFFICIENTX86-64 SERVER AND DESKTOP PROCESSORS:Using the
`
`2M
`
`Etrae
`
`fest =(01cm
`aoe
`
`
`
`0022
`
`
`
`
`Limitation
`Contention
`
`Modern multi-VT standard cell design methodologies and structures such as those used in the Llano Accused
`Products use a mix of low Vt (LVt), regular Vt (RVt), long channel regular Vt (LC-RVt) and long channel high Vt
`(LC-Hvt) standard cells.
`
`The Bulldozer module contains 84 unique custom macros and
`317,000 scannable flops. Module-level VSS power gating (C6)
`is used to reduce leakage power by approximately 95% when
`both cores are idle [4]. The 32 nm SOI process provides three
`transistor V7 types (low, regular, and high), with longer channel
`lengths used to achieve even finer-grained trade-offs between
`leakage and delay. V7’s used across the design consist mostly
`of regular (47%) and long-channelregular (46%), with less than
`1% low-V 7 used for the mostcritical paths.
`Source: McIntyre et al., Design Of The Two-Core x86-64 AMD “Bulldozer” Module In 32 nm SOI CMOS, IEEE Journal of Solid-
`State Circuits, Vol. 47, No. 1, January 2012, page 165.
`
`uteflwdfagfoudubefea
`
`
`0023
`
`
`
` Limitation
`
`Contention
`
`voltage.
`
`DVFS Characteristics Modern computer chips are designed using multiple
`types of transistors, i.e. a mixture of low-, medium-, and high-threshold transis-
`tors, to target different design tradeoffs, e.g. high-performance vs. low power.
`Low-threshold voltage (Low-Vt) devices are used in timing-critical paths,
`but have high leakage power. High-threshold voltage (High-Vt) devices have
`low leakage power but are slower, and are typically used in circuits that are
`off the timing-critical paths. Medium-threshold voltage (Mid-Vt) devices offer
`a tradeoff between High-Vt and Low-Vt devices by having medium power re-
`quirements and medium delay. In general, low power chips are designed using a
`larger percentage of High-Vt devices and high-performance chips with a larger
`percentage of Mid-Vt and Low-Vt devices. ‘The host processor is assumed to ex-
`ecute compute-intensive code and will therefore be a high-performance device.
`Source: Scrbak et al., DVFS Space Exploration in Power Constrained Processing-in-Memory Systems,pp. 4-5.
`
`The Accused Products contain a plurality of first MOStransistors, each of the first MOStransistors havinga first threshold
`
`0024
`
`
`
`Limitation
`
`Contention
`
`THEDIE| Process Technology
`
`RattNecatLe
`
`* Low-k dielectric
`
`* 32-nm Silicon-On-insulator (SOl) Hi-K Metal Gate
`(HKMG)process from GlobalFoundries
`
`transistors.
`
`* Dual strain liners and eSiGe to improve
`performance.
`
`* Multiple VT (HVT, VT, LVT) and long-channel
`
`0025
`
`
`
` JOTWANI et al: AN 586-64 CORE IN 32 am SOIL CMOS
`
`MOStransistors, each of the
`
`§(
`
`a)
`
`RVt} LCRVt
`
`HVt }
`
`Post Swapping
`Vt Types
`
`@32nm
`
`Post Swapping
`
`PreSwapping
`
`100%
`
`tli
`
`Dynamic
`
`Static
`
`Timing Slack (ps)
`(b)
`
`(a) Core device width histogram by Vt type. (b) Using Vi swaps to
`Fig. 10.
`reshape critical path timing opportunistically
`
`(a) TDP power distribution at | V. (b) Relative power improvement
`Fig, 11.
`with respect to 45 amgeneration core (normalized for performance).
`
`0026
`
`
`
`Limitation
`
`second MOStransistors having a
`second threshold voltage;
`
`Contention
`
`THE DIE | Photograph
`
`Source: Sean White, HIGH-PERFORMANCE POWER-EFFICIENTX86-64 SERVER AND DESKTOP PROCESSORS:Using the
`core code named “Bulldozer”, August 18, 2011, page 3.
`
`0027
`
`
`
`THE DIE | Floorplan (315 mm?)
`
`
`
`Ped eneat
`
`Source: Sean White, HIGH-PERFORMANCE POWER-EFFICIENTX86-64 SERVER AND DESKTOP PROCESSORS:Using the |
`
`Limitation
`
`Contention
`
`
`
`0028
`
`
`
`Limitation
`
`Contention
`
`THE DIE | Process Technology
`
`* 11-metal-layer-stack
`
`* Low-k dielectric
`
`* 32-nm Silicon-On-insulator (SOl) Hi-K Metal Gate
`(HKMG)process from GlobalFoundries
`
`transistors.
`
`* Dual strain liners and eSiGe to improve
`eatgitoe
`
`* Multiple VT (HVT, RVT, LVT) and long-channel
`
`0029
`
`
`
`
`Limitation
`Contention
`
`The Bulldozer module contains 84 unique custom macros and
`317,000 scannable flops. Module-level VSS powergating (C6)
`is used to reduce leakage power by approximately 95% when
`both cores are idle [4]. The 32 nm SOI process provides three
`transistor V7 types (low, regular, and high), with longer channel
`lengths used to achieve even finer-grained trade-offs between
`leakage and delay. V 7’s used across the design consist mostly
`of regular (47%) and long-channelregular (46%), with less than
`1% low-V 7 used for the mostcritical paths.
`Source: McIntyre et al., Design Of The Two-Core x86-64 AMD “Bulldozer” Module In 32 nm SOI CMOS, TEEE Journal of Solid-
`State Circuits, Vol. 47, No. 1, January 2012, page 165.
`
`
`
`Modern multi-VT standard cell design methodologies and structures such as those used in the Llano Accused
`Products use a mix of low Vt (LVt), regular Vt (RVt), long channel regular Vt (LC-RVt) and long channel high Vt
`(LC-Hvt) standardcells.
`
`LutHTHee
`
`0030
`
`
`
` Limitation
`
`Contention
`
`a second threshold voltage:
`
`DVFS Characteristics Modern computer chips are designed using multiple
`types of transistors, i.e. a mixture of low-, medium-, and high-threshold transis-
`tors, to target different design tradeoffs, e.g. high-performance vs. low power.
`Low-threshold voltage (Low-Vt) devices are used in timing-critical paths,
`but have high leakage power. High-threshold voltage (High-Vt) devices have
`low leakage power but are slower, and are typically used in circuits that are
`off the timing-critical paths. Medium-threshold voltage (Mid-Vt) devices offer
`a tradeoff between High-Vt and Low-Vt devices by having medium power re-
`quirements and medium delay. In general, low power chips are designed using a
`larger percentage of High-Vt devices and high-performance chips with a larger
`percentage of Mid-Vt and Low-Vt devices. ‘The host processor is assumed to ex-
`ecute compute-intensive code and will therefore be a high-performance device.
`Source: Scrbak et al., DVFS Space Exploration in Power Constrained Processing-in-Memory Systems,pp. 4-5.
`
`The Accused Products contain a plurality of second MOStransistors, each of the second MOStransistors having
`
`0031
`
`
`
`'
`ve 7}PostSwapping
`Post S
`RVt | LCRVt
`Vt Types
`
`#245nm &32nm
`
`n
`
`
`
`The Accused Products contain a unit cell array comprised ofsaid first and second unit cells laid in array form
`
`Post Swapping
`
`75
`
`Timing Slack (ps)
`(b)
`
`{00%
`
`;i
`
`Static
`
`100%
`
`Dynamic
`
`(b)
`
`(a) Core device width histogram by Vt type. (b) Using Vt swaps to
`Fig. 10.
`reshape critical path timing opportunistically,
`
`(a) TDP power distribution at 1 V. (b) Relative power improvement
`11.
`Fig.
`with respect to 45 nm generation core (normalized for performance).
`
`0032
`
`
`
`Limitation
`
`Contention
`
`THE DIE | Photograph
`
`CVSSSeeteed
`
`Source: Sean White, HIGH-PERFORMANCE POWER-EFFICIENTX86-64 SERVER AND DESKTOP PROCESSORS:Using the
`core code named “Bulldozer”, August 18, 2011, page 3.
`
`0033
`
`
`
`
`
`Source: Scrbaket al., DVFS Space Exploration in Power Constrained Processing-in-Memory Systems,pp. 4-5.
`
`DVFS Characteristics Modern computer chips are designed using multiple
`types of transistors, i.e. a mixture of low-, medium-, and high-threshold transis-
`tors, to target different design tradeoffs, e.g. high-performance vs. low power.
`Low-threshold voltage (Low-Vt) devices are used in timing-critical paths,
`but have high leakage power. High-threshold voltage (High-Vt) devices have
`low leakage power but are slower, and are typically used in circuits that are
`off the timing-critical paths. Medium-threshold voltage (Mid-Vt) devices offer
`a tradeoff between High-Vt and Low-Vt devices by having medium power re-
`
`0034
`
`
`
`Limitation
`
`Contention
`
`a powerswitch disposed around
`said unit cell array and comprised
`of a plurality of third MOS
`transistors, each of the third MOS
`transistors having the second
`
`The Accused Products contain a powerswitch disposed aroundsaid unit cell array and comprised ofa pluraility of third MOS
`transistors.
`
`POWER MANAGEMENT| CoreC6State (CC6)
`
`threshold voltage; and
`
`if a core isn't active,
`*Core C6:
`remove power
`
`*Implementedin this physical
`design by a power gating ring
`that isolates the Core VSS for
`each Bulldozer module from the
`“Real” VSS
`
`»"CC6 entry: when both Bulldozer
`cores in the module areidle,
`flush caches and dump register
`state to CC6 save space, then
`gate Core VSS
`
`j
`
`Bulldozer
`Module
`
`hz Power Gating FETS
`
`0035
`
`
`
`Limitation
`
`Contention
`
`Source: Sean White, HIGH-PERFORMANCE POWER-EFFICIENTX86-64 SERVER AND DESKTOP PROCESSORS:Using the
`core code named “Bulldozer”, August 18, 2011, page 21.
`
`Each of the third MOStransistors of the Accused Products has the second threshold voltage:
`
`» 32-nm Silicon-On-insulator (SOl) Hi-K Metal Gate
`(HKMG)process from GlobalFoundries
`
`transistors.
`
`THE DIE | Process Technology
`
`MeCercatolee
`
`* Low-k dielectric
`
`* Dual strain liners and eSiGe to improve
`performance.
`
`* Multiple VT (HVT, RVT, LVT) and long-channel
`
`0036
`
`
`
`
`
`Limitation
`
`Contention
`
`core code named “Bulldozer”, August 18, 2011, page 6.
`
`Each of the third MOStransistors of the power gate ring has the second threshold voltage.
`VII. POWER GATING
`
`We defined a low-power mode called core-level C6 (CC6) to
`allow core-level power gating [5] during periods of inactivity.
`The coreis isolated from the supply during CC6 by a power-gate
`ring surrounding the CPU and L2 cachepair, allowing core level
`power down in a chip with multiple cores attached to a common
`power supply. The SOI process enables the gating of VSS (not
`
`VDD), constructing thdpower-gatewithregularVinMOSllogic
`devices without the need for extra processing steps to reduce
`on-state resistance [6]. Fig. 12(a) describes the core operations
`controlled by the power managementsystem for CC6 entry and
`exit sequences.
`Fig. 12(b) details the connections of the power-gate ring with
`respect to the core and the C4 bumps. In addition to two 16X
`M10 and M11! on-die metal layers, a low-impedance package
`layer connected to the die by C4 bumpsis dedicated for use as a
`virtual-ground layer, eliminating the need for anyultra-thicksil-
`icon metallization layer [6]. The low-impedance package layer
`
`alray.
`
`0037
`
`
`
`THE DIE| Floorplan (315 mm?)
`
`peek ES
`
`STLS ACoeee
`
`he A semiconductor integrated circuit
`+ The powergate ring is turned off during standby and turned on whenthe core exits C6state.
`core code named “Bulldozer”, August 18, 2011, page 5.
`Limitation Contention
`
`Bulldozer
`
`Fr
`
`
`
`0038
`
`
`
`Limitation
`
`device accordingto claim 1,
`wherein said powerswitch is turned
`off during standby and turned on
`
`
`
`1%low-V 7 used for the mostcritical paths.
`
`The Bulldozer module contains 84 unique custom macros and
`317,000 scannable flops. Module-level VSS power gating (C6)
`is used to reduce leakage power by approximately 95% when
`both cores are idle [4]. The 32 nm SOIprocess provides three
`transistor V7 types (low, regular, and high), with longer channel
`lengths used to achieve even finer-grained trade-offs between
`leakage and delay. V p's used across the design consist mostly
`of regular (47%) and long-channel regular (46%), with less than
`
`Contention
`
`Source: McIntyre et al., Design Of The Two-Core x86-64 AMD “Bulldozer” Module In 32 nm SOI CMOS, TEEE Journal of Solid-
`State Circuits, Vol. 47, No. 1, January 2012, page 165.
`
`
`
`0039
`
`
`
`Limitation
`
`Contention
`
`|
`
`- Bulldozer
`Module
`
`~~
`
`Power Gating FETs
`
`interrupts, etc.)
`
`*"CC6 entry: when both Bulldozer "3°"
`coresin the module areidle,
`flush caches and dump register
`state to CC6 save space, then
`gate Core VSS
`
`if acore isn’t active,
`*Core C6:
`remove power
`
`*Implementedin this physical
`design by a power gating ring
`that isolates the Core VSSfor
`each Bulldozer module from the
`“Real” VSS
`
`*"CC6 exit: ungate Core VSS,
`reload CC6 savedstate, resume
`execution (ex: service
`
`0040
`
`
`
`Accused Product: AMD APUscontaining Excavator cores (Representative Product: AMD Carrizo APU Processors)
`
`Exhibit A.3: Preliminary Infringement Contention Charts for U.S. Patent 6,239,614
`
`These preliminary infringement contentions were prepared without the benefit of the Court’s claim construction or the parties’ exchange of constructions.
`As of the date of these contentions, AMD has not produced any information concerning the Accused Products. Thus, this chart is based on publicly available
`evidence, and based upon information and reasonablebelief in light of such evidence. As such, Aquila reserves the right to amend or supplement its
`contentions to address any issuesarising from the Court’s constructions or to account for new information that becomesavailable.
`
`
`To the extent the preambleis a limitation, the Accused Products are semiconductorintegratedcircuits. circuit device, comprising:
`
`A semiconductor integrated
`
`Exhibit A.3
`
`0041
`
`-1-
`
`0041
`
`
`
`AMD 6' GENERATION
`A-SERIES PROCESSOR
`
`DDR/PHY.
`
`X86
`Welsfetta
`
`NET
`Wen
`
`Southbridge
`
`Northbridge
`
`welETEN
`
`Limitation Contention
`
`Source: Krishnan, Energy Efficient Graphics and Multimedia in 28nm Carrizo APU | Hot Chips27,p. 2.
`
`To the extent that the powergate ring implemented in this product operates similarly to the Bulldozer or Llano
`powergate rings, for example in terms of the threshold voltages of the MOStransistors in the various unit cells
`and the powerswitch, Aquila contends that the charts applicable to those products apply equally to this product,
`
`Exhibit A.3
`
`0042
`
`-2-
`
`0042
`
`
`
`Ieies andincorporates those chartsbyreference.
`
`The Accused Productsdisclose a plurality of first unit cells:
`
`voltage;
`
`
`Limitation
`Contention
`
`a plurality of first unit cells
`each including a plurality of
`first MOStransistors, each of
`the first MOStransistors
`havinga first threshold
`
`Exhibit A.3
`
`0043
`
`-3-
`
`0043
`
`
`
`
`Limitation
`
`Modern multi-VT standard cell design methodologies and structures such as those used in the Llano Accused
`Products use a mix of low Vt (LVt), regular Vt (RVt), long channel regular Vt (LC-RVt) and long channel high Vt
`(LC-Hyvt) standardcells.
`
`[Lut] bt]
`Lvt
`{uve} t/t
`vt
`GeVYfiter}fillerfOVY U-VETLVfiller filler
`
`Source: Scrbaket al., DVFS Space Exploration in Power Constrained Processing-in-Memory Systems,pp. 4-5.
`
`DVFS Characteristics Modern computer chips are designed using multiple
`types of transistors, i.e. a mixture of low-, medium-, and high-threshold transis-
`tors, to target different design tradeoffs, e.g. high-performance vs. low power.
`Low-threshold voltage (Low-Vt) devices are used in timing-critical paths,
`but have high leakage power. High-threshold voltage (High-Vt) devices have
`low leakage power but are slower, and are typically used in circuits that are
`off the timing-critical paths. Medium-threshold voltage (Mid-Vt) devices offer
`a tradeoff between High-Vt and Low-Vt devices by having medium power re-
`quirements and medium delay. In general, low power chips are designed using a
`larger percentage of High-Vt devices and high-performance chips with a larger
`percentage of Mid-Vt and Low-Vt devices. ‘The host processor is assumed to ex-
`ecute compute-intensive code and will therefore be a high-performance device.
`
`Exhibit A.3
`
`0044
`
`-4-
`
`0044
`
`
`
`
`
`DYNAMIC UVD POWER GATING PN|oP
`
`:
`“Kaveri” UVD busy and burning power the whole time
`
`
`Beinn 44 Dynamic inter frame power gating controlled by rtsEEEEEEEEUEUEEEEEEEEEE———————————__—___——___
`microcontroller firmware
`VJ
`)
`I
`i
`ad
`lad
`rom
`v
`
`- Pipeline idle detection enables header/footer
`power gating of the entire IP
`
`Dynamic power gating along with low power
`hardening of the video decoder enables CZ to
`negate the bigger video decoder needed for
`H.265offload
`
`ee Sela metaa ee Le) cee oe
`
`“Carrizo” UVD and power gates
`
`Limitation
`
`Contention
`
`The Accused Products disclose a plurality of first MOStransistors, each of the first MOS transistors having a
`first threshold voltage, e.g., RVt:
`
`teeoff ond PUTS DRAMinto low power made
`
`Source: ENERGY EFFICIENT GRAPHICS AND MULTIMEDIA IN 28NM CARRIZO APU, page 21 HOT CHIPS
`27 —- AUGUST2015
`
`Exhibit A.3
`
`0045
`
`-5-
`
`0045
`
`
`
`The Accused Products disclose a plurality of second unit cells:
`
`a plurality of second unit cells
`each including a plurality of
`second MOStransistors, each
`of the second MOStransistors
`having a second threshold
`voltage;
`
`
`Limitation
`Contention
`
`
`
`Modern multi-VT standardcell design methodologies and structures such as those used in the Llano Accused
`Products use a mix of low Vt (LVt), regular Vt (RVt), long channel regular Vt (LC-RVt) and long channelhigh Vt
`
`Exhibit A.3
`
`0046
`
`0046
`
`
`
`
`Limitation
`Contention
`
`(LC-Hyvt) standardcells.
`
`HeeEee
`
`Source: Scrbak et al., DVFS Space Exploration in Power Constrained Processing-in-Memory Systems, pp. 4-5.
`
`DVFS Characteristics Modern computer chips are designed using multiple
`types of transistors, i.e. a mixture of low-, medium-, and high-threshold transis-
`tors, to target different design tradeoffs, e.g. high-performance vs. low power.
`Low-threshold voltage (Low-Vt) devices are used in timing-critical paths,
`but have high leakage power. High-threshold voltage (High-Vt) devices have
`low leakage power but are slower, and are typically used in circuits that are
`off the timing-critical paths. Medium-threshold voltage (Mid-Vt) devices offer
`a tradeoff between High-Vt and Low-Vt devices by having medium power re-
`quirements and medium delay. In general, low power chips are designed using a
`larger percentage of High-Vt devices and high-performance chips with a larger
`percentage of Mid-Vt and Low-Vt devices. ‘The host processor is assumed to ex-
`ecute compute-intensive code and will therefore be a high-performance device.
`
`Exhibit A.3
`
`0047
`
`-7-
`
`0047
`
`
`
`Limitation
`
`Contention
`
`The Accused Products disclose a plurality of