`
`|http://anandtech.com/printarticle.aspx?i=3513
`InterNeT ARCHIVE
`
`
`
`
`Wwayoaelmeting1capture
`31 Dec 2009
`BRET Y About this capture
`
`
`
`PRANAUTECH
`
`~yourSOUICE for hardware analysisand news
`
`printthispage
`
`Intel's 32nm Update: The Follow-on to Core i7 and More
`Date: Feb 11, 2009
`Type: CPU & Chipset
`Manufacturer: Intel
`Author: Anand Lal Shimpi
`Page 1
`Sevenbillion dollars.
`
`That’s the amountthatIntel is going to spend in the US alone on bringing up its 32nm manufacturing process in 2009 and 2010.
`
`Theseare the fabsIntel is converting to 32nm: Fab 32 Arizona - 2010
`
`In Oregon Intel has the D1D fab whichis already producing 32nm parts, and D1C whichis scheduled to start 32nm production at the end of this year. Then two
`fabs in Arizona: Fab 32 and Fab 11X. Both of them comeonline in 2010.
`
`By the endof next year the total investmentjust to enable 32nm production in the US will be approximately eight billion dollars. In a time where all we hear about
`are bailouts, cutbacks and recession, this is welcome news.
`
`If anything, Intel should have a renewed focus on competition given that its chief competitorfinally woke up. That focus is there. The show mustgo on. 32nm will
`happenthis year. Let’s talk about how.
`
`Page 2
`
`The Manufacturing Roadmap
`
`Thetick-tock cadence may have come aboutat the microprocessor level, but its roots have always been in manufacturing. As long as I’ve been running
`AnandTech, Intel has introduced a new manufacturing process every two years. In fact, since 1989 Intel has kept up this two year cycle.
`
`Process Name
`
`
`P1264
`
`Lithography
`
`65nm
`
`1% Production
`
`2005
`
`
`P1272
`
`2013
`
`16nm
`
`
`
`
`Wesawthefirst 45nm CPUs with the Penryn core back in late 2007. Penryn, released at the very high end, spent most of 2008 making its way mainstream. Now
`you can buy a 45nm Penryn CPUfor less than $100.
`
`The next process technology, which Intel refers to internally as P1268, shrinks transistor feature size down to 32nm. The table above showsyouthatfirst
`production will be in 2009 and, after a brief pause to check your calendars, that meansthis year. More specifically, Q4 of this year.
`I'll get to the products in a moment, butfirst let’s talk about the manufacturing processitself.
`Here we have our basic CMOStransistor:
`
`
`Google Ex. 1052 - Page 1
`Google Ex. 1052 - Page 1
`
`Google Exhibit 1052
`Google Exhibit 1052
`Google v. Singular
`Google v. Singular
`
`

`

`+*——__ Low resistance layer
`
`Depleted region —~* =} ,_ Gatedielectric (SiO)
`
`
`
`Silicon substrate
`
`
`Current flows from source to drain when thetransistor is on, and it isn’t supposed to flow whenit’s off. Now as you shrink the transistor,all of its parts shrink. At
`65nm Intel found that it couldn’t shrink the gate dielectric any more without leaking too much current through the gateitself. Back then the gate dielectric was
`1.2nm thick (about the thickness of 5 atoms), but at 45nm Intel’s switched from a SiO2 gate dielectric to a high-k one using Hafnium. That’s where the high-k
`comesfrom.
`
`The gate electrode also got replaced at 45nm with a metal to help increase drive current (more current flows when you wantit to). That’s where the metal gate
`comesfrom.
`
`
`The combination of the two changesto the basic transistor gave us!ntel’s high-k + metal gate transistors at 45nm, and at 32nm wehavethe second generation of
`those improvements.
`
`The high-k gate dielectric getsalittle thinner (equivalent to a 0.9nm SiO2 gate, but presumably thickersinceit’s Hafnium based, down from 1.0nm at 45nm ) and
`we'vestill got a metal gate.
`
`Standard
`Transistor
`
`HK+MG
`Transistor
`
`
`
`+— High-k gate oxide
`Low resistance layer —_-
`
`Polysilicon gate ia
`SiO, gate oxide ——>» '''‘’,
`
`
`
`
`_ Low resistance layer
`a Metal gate
`
`At 32nm thetransistors are approximately 70% the size of Intel’s 45nm hk + mgtransistors, allowing Intel to pack more in a smaller area.
`
`2"¢ generation high-k + metal gate transistors
`
`Immersion lithography on critical layers
`
`9 copper + low-k interconnectlayers
`
`e ~70%dimension scaling from 45nm generation
`
`Pb-free and halogen-free packages
`
`The big changehereis that Intel is using immersionlithography on critical metal layers in order to continue to use existing 193nm lithography equipment. The
`smaller your transistors are, the higher resolution your equipment hasto be in orderto actually build them. Immersionlithography is used to increase the resolution
`of existing lithography equipmentwithout requiring new technologies. It is a costlier approach, but one that becomes necessary as you scale below 45nm. Note
`that AMD to made the switch to immersion lithography at 45nm.
`
`e 24 generation high-k + metal gate
`
`- 0.9nm equivalent oxide thickness high-k
`(scaled from 1.0 nm on 45nm)
`
`Replacement Metal Gate process flow
`
`- 30nm gate length
`
`- 4% generation strained silicon
`
`e >22%performance increase
`
`
`e Tightest reported gate pitch
`
`Google Ex. 1052 - Page 2
`Google Ex. 1052 - Page 2
`
`

`

`e Highest reported drive currents
`
`Intel reported significant gains in transistor performance at 32nm; the graphs below help explain:
`
`32nm Transistor Performance
`
`1000 ;
`
`Ss
`= 100
`2
`&
`
`S 10
`
`1
`
`PMOS
`
`intel
`45nm
`
`intel
`32nm
`
`1000
`
`3
`= 100
`2
`&
`
`5 10
`
`1
`
` NMOS
`
`06 08 10 12 14 16 18
`08 10 12 14 16 18 2.0
`lon (mA/um)
`low (mA/um)
`32nm provides improved performance or reduced leakage
`Highest drive current of all reported 32nm technologies
`
`We're looking at the comparison of leakage currentvs. drive current for both 32nm NMOSand PMOStransistors. The newtransistors showcase a huge
`improvement in powerefficiency. You can either run them faster or run them at the same speedand significantly reduce leakage current by a magnitude of greater
`than 5 - 10x comparedto Intel’s 45nm transistors. Intel claims that its 32nm transistors boast the highest drive current of all reported 32nm technologiesatthis
`point, which admittedly there aren't many.
`
`The power/performancecharacteristics of Intel’s 32nm process makeit particularly attractive for mobile applications. But more onthatlater.
`
`Page 3
`
`Fat Pockets, Dense Cache, Bad Pun
`
`WheneverIntel introduces a new manufacturing processthe first thing we seeit used onis a big chip of cache. The good ol’ SRAMtest vehicle is a great way to
`iron out early bugs in the manufacturing process and at the end of 2007 Intel demonstratedits first 32nm SRAMtestchip.
`
`0.171 um? cell
`
`291 Mbit
`
`>1.9 billion transistors
`
`4 GHz operation
`
`First demonstrated Sep ‘07
`
`Intel's 32nm SRAMtest vehicle
`
`The 291Mbit chip was made up of over 1.9 billion transistors, switching at 4GHz, using Intel’s 32nm process. The important numberto lookatis the cell size,
`whichis the physical area a single bit of cache will occupy. At 45nm thatcell size was 0.346 um‘2 (for desktop processors, Atom usesa slightly largercell),
`compared to 0.370 um42 for AMD’s 45nm SRAMcell size. At 32nm you can cut the area nearly in half down to 0.171 um42 for a 6T SRAMcell. This meansthatin
`the samedie area Intel can fit twice the cache, or the same amountof cachein half the area. Given that Corei7 is a fairly large chip at 263 mm/‘2 I'd expectIntel
`to take the die size savings and run with them. Perhaps with a modest increase to L3 cachesize.
`SRAM Cell Size Scaling
`
`Cell Area
`
`(um?)
`
`10
`
`0.171 um? Cell 0.1
`
`32nm Generation
`
`2010
`2005
`2000
`1995
`
`A hia raacan we're avian aattina thie dieclacira tadaw ic hacaiica af haw healthy tha 29nm nracecc ic Ralawwe have a aranh af defact dancitv (nitmhar af dafacte
`
`Google Ex. 1052 - Page 3
`Google Ex. 1052 - Page 3
`
`

`

`PAY PUUUUr Wee Uru guuiy UNE UUIVUUTY WwuUy tu Be UUUSY Uo Fredy re Ge pruueuy ie. wer We Tle U6 gu Ur Were Uuuny (rer Ur ULEUL
`in silicon per area) vs time; manufacturing can’t start until you’re at the lowest part of that graph- thetail that starts to flatten out:
`
`(log scale)
`
`—
`ty
`
`Higher
`
`Chi
`Yield
`
`Intel’s 45nm process ramped and matured very well as you can see from the chart. The 45nm process reached lower defect densities than both 65nm and 90nm
`anddid it faster than either process. Intel’s 32nm processis on track to outperform eventhat.
`
`Two Different 32nm Processes?
`
`With Intel now getting into the SoC business (System on a Chip), each process nodewill now havetwoderivatives - one for CPUs and onefor SoCs. This started
`at 45nm with process P1266.8, used for Intel’s consumerelectronics and Moorestown CPUs and will continue at 32nm with the P1269 process.
`
`
`45 nm
`32 nm
`22 nm
`
`Process:
`
`P1266 P1266.8
`
`P1268] P1269
`
`P1270!
`
`(P1271
`
`Products:
`
`CPU
`
`SoC
`
`CPU
`
`SoC
`
`CPU
`
`SoC
`
`There are two majordifferences between the CPU and SoC versions of a given manufacturing process. One, the SoC version will be optimized for low leakage
`while the CPU versionwill be optimized for high drive current. Rememberthat graph comparing leakagevs. drive current of 45nm vs. 32nm? The P1268 process
`will exploit the arrows to the right, while P1269 will attempt to push leakage current down.
`
`The seconddifferenceis that certain SoC circuits require higher than normal voltages and thus you needa processthat can tolerate those voltages. Remember
`that with a SoCit’s not always just Intel IP being used, there are manythird parties that will contribute to the chips that eventually make their way into smartphones
`and otherultra portable devices.
`
`The buck doesn’t stop here, in two more years we'll see the introduction of P1270, Intel’s 22nm process. But before we getthere, there’salittle stop called Sandy
`Bridge. Let’s talk about microprocessorsfor a bit now shall we?
`
`Page 4
`
`Tick-Tock: U R Doin it Right
`
`Let’s check the stats; Conroe in July 2006, Penryn in October 2007, Nehalem in November 2008. That’s a tock, tick, and another tock, each about a year apart.
`Note that the cadence does appearto beslipping a bit, but we'll see exactly when in 2009 we get Westmere before making any accusations.
`
`aeeea
`a
`epeeait aitli)
`
`Future Intel®
`Microarchitecture
`
`Forecast
`
`fee)
`NEW
`Microarchitecture
`Lee]
`©
`TOCK
`
`Nehalem
`Penryn
`NEW
`a
`Process Technology Microarchitecture
`rye)
`45nm
`2)
`©
`ile
`TOCK
`
`Sandy
`See
`Westmere
`NEW
`st
`(Process Technology Microarchitecture
`EPA)
`32nm
`
`TICK
`
`TOCK
`
`The nexttick is, as | just mentioned, Westmere. It’s a 32nm shrink of Nehalem, much like Penryn was a 45nm shrink of Conroe/Merom. Andit’s due outin the
`fourth quarter of this year.
`
`Yesterday, Intel demonstrated working versionsof its 32nm processors in both desktops and notebooks. The notebook aspect of the demonstration is very
`important, whichI'll get to later. Both mobile and desktop versions of Westmerewill be shipping from Intel in Q4.
`
`Getting Complicated with Code Names
`
`Nehalem is the overall namefor Intel’s 45nm desktop/mobile/server product family. At the high end we have Bloomfield, which is the quad-core, eight-thread, Core
`i7 processor weall long for. That’s the only Nehalem derivative that’s launched thus far.
`
`(a Processor|Cores| Threads pease
`Process
`[DEAK]
`
`EhabEnd
`
`45nm
`
`
`coe Bloomfield
`
`Google Ex. 1052 - Page 4
`Google Ex. 1052 - Page 4
`
`mines
`
`

`

`1500
`vVeSsKtop
`,
`LGA-
`Mainstream
`2H 2009
`8
`4
`Lynnfield
`1156
`45nm
`Desktop
`
`
`Mobile 45nm—™S* Clarksfield 4 8 2H 2009
`
`
`LGA- Nehalem-
`1567
`EX
`LGA- Nehalem-
`1366
`EP
`face
`Lynnfield
`
`ZUUG6
`
`8
`
`4
`4
`
`16
`
`2H 2009
`
`8
`8
`
`1H 2009
`2H 2009
`
`4S Server
`
`2S Server
`1S Server
`
`45nm
`
`45nm
`45nm
`
`By the endofthis year we'll see Lynnfield and Clarksfield. These are both quad-core, eight-thread Nehalem processors but at lower TDPsandprice points. They
`will fit into Intel’s unannounced LGA-1156 socket and only support two channels of DDR3 memory (compared to LGA-1366 and 3-channels with Corei7).
`
`On the server side we'll have Nehalem-EX,an 8-core, 16-thread version. Nehalem EP a 4-core, 8-thread version. And Lynnfield again for the entry level servers.
`Theseare all 45nm parts and all due out by the end ofthis year.
`
`Note that there’s one name missing: Havendale. Havendale was supposedto be a 2-core Lynnfield + on-chip graphics, perfect for notebooks and low end
`desktops where quad-coreisn’t necessary. Unfortunately, Havendale got delayed until Q4 2009 with systems shipping in Q1 2010. That just happened to coincide
`with Intel’s 32nm ramp so a very significant decision was made: Havendale got scrapped.
`
`Page 5
`
`Enter the 32nm Lineup
`
`Instead of Havendale in Q4, we'll get Clarkdale and Arrandale. These are both dual-core, quad-thread processors, and both have on-package graphics. The CPU
`coreswill be built on Intel’s 32nm processandin fact, they will be the first Westmere CPUs shipping into the market.
`
`Nehalem/Westmere Client Roadmap
`2009
`2010+
`
`pee Di)
`
`2010 HEDTPlatform
`
`32nm Westmere extends Nehalem through the mainstream
`
`Intel® Core 17 Extreme Processor (4C/8T)|32nmGulftownProcessor(6C/_Pome LeetdtT) scree Caltuowen Processes (OU
`
`
`
`te ClESdeyig welBorgarg
`
`
`barteB)ye
`
`Now notethat the dual-core marketis the largest slice of the processorpie. Intel must be incredibly confident in its 32nm processto start shipping it into these
`demand marketsfirst. Rememberthat both 65nm and 45nm initially launched on the high end desktop, but 32nm is making its debut in mainstream notebooks and
`desktops. The 32nm rampis going to be a good onefolks.
`
`eaeany
`Process
`
`Processor|Cores| Threads pais
`Date
`
`LGA-
`High End
`1H 2010
`Gulftown
`1366
`32nm
`Desktop
`As
`4
`2
`ee Clarkdale
`te 32nm
`
`
`Mobile 32m=PCA prrandale &|geay2
`
`
`
`
`4S Server sear|a?32nm 2 2 2010
`
`
`2S Server
`32nm | ae
`2
`2
`2010
`1S Server
`32nm
`ae Clarkdale
`2
`4
`2010
`
`Clarkdale/Arrandale have 32nm CPUs but their on-package GPUsarestill built on Intel’s 45nm process; these are the GPUsthat were supposed to be used for
`Havendale! It won't be until 2010 with Sandy Bridge that we see a 32nm CPU and 32nm GPUonthe same package.
`
`
`
`
`
`Google Ex. 1052 - Page 5
`Google Ex. 1052 - Page 5
`
`

`

`Westmere
`\ yf)
`.
`core
`
`45nm integrated
`graphics & integrated
`uN eeCle
`
`Peed seca cea
`Intel® Hyper-Threading technology (2 Cores, 4 threads)
`Integrated graphics, discrete / switchable graphics
`
`Aside effect of the Clarkdale/Arrandale architecture is that the memory controller is now located on the GPU and not the CPU, although botharestill on package
`and should still be quite low latency.
`
`Westmere Family
`Mainstream Desktop / Mobile Processor
`
`arUte
`* Intel microarchitecture codename Nehalem on 32nm
`BeeeeeTiepceeei
`.
`CeegPera rae Cc)
`euee Cee essy
`Sees ceates|og]
`controller built on 45nm high-k metal gate process
`PMeeeella
`BM eateeeeia ele Cemeoe Bay go)
`Ce a ea]
`= Integrated memory controller (IMC)
`«= Integrated, discrete / switchable graphics support
`= Advanced Encryption Standard (AES) acceleration
`PeeMm tee Rtg(Pelesgen
`Peete Bed eestsbidtS)
`
`Keepfollowing; if you want a quad-core Westmere, your only option will be in the LGA-1366 socket with Gulftown. Core i7 will get replaced with a six-core,
`twelve-thread processorin early 2010. There won’t be a 32nm quad-core part on the desktop until the end of 2010 with Sandy Bridge.
`
`Worlds First 32nm Based Processor
`
`Page 6
`
`The Server Roadmap
`
`Intel’s 32nm server roadmap is notably different from the desktop roadmap. Nehalem-EX will ship into the Xeon 7000 series as an 8-core, 16-thread part.It will
`eventually get replaced sometime in 2010 with a 32nm Westmere derivative.
`
`integrated Memory Controller (IMC) - 2ch DDR3
`
`MA haces
`
`Intel XeonEnterprise Roadmap
`-
`
`2009
`
`@
`Xeon
`
`2010+
`beee
`
`beee
`
`Patcrea nc)i
`@ Peersyee
`micro-
`=
`ae ee MEea
`See ed
`teeter
`Pty
`Tylersburg & Dual-OH Chipsets
`pan
`(rte) — DOaeg
`
`=
`a3znmClarkdaleProcesso
`alee z
`Meee)a
`32nm Upgrades AcrossAll Intel® Xeon* Segments:
`Entry (EN), Efficient Performance (EP) & Expandable (EX)
`Aa Gates, product Gescriptions, availabilty, and plans are
`=
`Petter tte
`
`We'll see a 32nm six-core Westmere based processorin the Xeon 5000 series in 2010.
`
`Finally Lynnfield and Clarkdale will be carried over to the entry level Xeon platforms at the end of this year and into 2010.
`
`What About Chipsets?
`
`Intel’s X58 chipsetwill remain the top dog through 2010. Chancesare that we won't seeit replaced until the next tock with Sandy Bridge. Nowthatisn’t to say that
`the six-core 32nm Gulftownwill work in existing X58 motherboards; while that would be nice, Intel does have a habit of forcing motherboard upgrades, we’ll have
`to wait and see.
`
`
`Google Ex. 1052 - Page 6
`Google Ex. 1052 - Page 6
`
`

`

`The rest of the Nehalem/Westmerefamily will rely on Intel’s upcoming P55 chipset:
`
`YTSctse-m erehyete
`2-Chip Solution
`
`Processor
`
`,
`‘ PCle
`iGFX
`Graphics
`Mitea
`Display
`aytzlm fee)
`
`ye
`
`ny)
`
`TsMgny
`meme)1)
`
`0 Gera
`STacl
`
`/0
`
`Originally both Lynnfield and Havendale were to have an on-package PCle controller, I’m not sure if that has changed with the Havendale cancellation but | see no
`reasonforit to have. In which case a Lynnfield system will still look like this:
`
`Eo =
`
`Tee Rai
`
`peeMemiteseiesl \yi=o ec
`
`
`DMI- 2.0GB/s
`
`Taeh)
`PCH
`
`Page 7
`
`Westmere’s NewInstructions
`
`Muchlike Penryn and its new SSE4.1 instructions, Westmere comeswith 7 new instructions added to those already in Core i7. These instructions are specifically
`focused on accelerating encryption/decryption algorithms. There’s a single carryless multiply instruction (PCLMULQD@Q...I love typing that) and 6 instructions of
`AES.
`
`- Enables full disk encryption
`
`Westmere: AES New Instructions
`
`¢ Use addiiofia) ijansistor budget to add new capabilities
`- Similar to adding SSE4.7 in Penryn (45nmtick)
`¢ 7 NeW iNSHUGHONS fo_ accelerating encryption/decryption
`algorithms
`~ Carryless multiply (PELMULQDQ)
`- 6 instructions for AES
`
`¢ Example client usage
`
`
`Intel gives the example of hardware accelerated full disk encryption as a need for these instructions. With the new instructions being driven into the mainstream
`
`Google Ex. 1052 - Page 7
`Google Ex. 1052 - Page 7
`
`

`

`first, we'll probably see quicker than usual software adoption.
`
`Final Words
`
`Whatis there to say otherthan: it’s a healthy roadmap. Theonly casualty I’ve seen is Havendale but |’d gladly trade Havendale for a 32nm version. Butlet’s get
`downto what this means for what you should buy and when.
`
`At the very high end, Core i7 users havelittle reason to worry. While Intel is expected to bumpi7 up to 3.33GHz in the near future, nothing below i7 looks
`threatening in 2009. Moving into 2010, the 6-core 32nm i7 successor should be extremely powerful. Intel’s strategy with LGA-1366 makesa lot of sense: if you
`want morecores, this is the platform you're going to have to be on.
`
`Nowalthough | said that nothing will threaten Core i7 this year, you may be able to get i7-like performance out of Lynnfield in the second half. A quad-core
`Lynnfield running near 3GHz, should offer much of the performanceof an i7 with a lower platform cost. Rememberbackto our original i7 review; we didn’t find a
`big performance benefit from three channels of DDR3versus two.
`
`Lynnfield is on track for a 2H 2009introduction and if you’re unable to makethe jumpto i7 now, you'll probably be able to get i7-like performance out of Lynnfield in
`about 6 months. Intel did mention that the most overclockable processors would go into the LGA-1366 socket. Combine better overclockability with the promise of
`6 coresin the future and it seems like LGA-1366 is shaping up to be a platform that’s going to stick around despite cheaperalternatives.
`
`The 32nm Clarkdale/Arrandale parts arriving by the end of this year really means onevery importantthing: the time to buy a new notebookwill be either in Q4
`2009 or Q1 2010. A 2-core, 4-thread 32nm Westmerederivative is not only going to put current Penryn cores to shame, it’s going to be extremely powerefficient.
`In its briefing yesterday, Intel mentioned that while Clarkdale/Arrandale clock speeds and TDPs would be similar to what we havetoday, you'll be getting much
`more performance. Seeing what we’ve seen thusfar with Nehalem, I’d say a 2-core, 32nm version in a notebookis going to be reason enough for you to want to
`upgrade.
`
`If | had to build a new desktop todayI’d go Core i7 and think about upgrading to a 6-core version sometime next year. If | couldn’t or didn’t need to build today,
`then the thing to wait for is Lynnfield. Four cores that should deliveri7-like performance just can’t be beat, and platform costs will be much cheaper by then (expect
`~$100 motherboards and near price parity between DDR3 and DDR2).
`On the mainstream quad-coreside, it may not make senseto try to upgrade to 32nm quad-coreuntil Sandy Bridge at the end of 2010. If you buy Lynnfield this
`year, chancesare that you won’t feel a need to upgrade until late 2010/2011.
`
`Onthe notebookside, if | needed one today I’d buy whatever | could keeping in mind that within a year I’m going to want to upgrade. | mentioned this in much of
`my recent Mac coverage; if you bought a new MacBook, it looks great, but the one you're going to really wantwill be here in about a year.
`
`WeoweIntel a huge thanksfor being so forthcoming with its roadmaps. It’s going to be a good couple of years for performance.
`
`
`
`Google Ex. 1052 - Page 8
`Google Ex. 1052 - Page 8
`
`