LONDON – Chip giant Intel and the research partnership clustered around IBM and STMicroelectronics are each set to report progress on their approaches to leading-edge IC manufacturing during the International Electron Devices Meeting (IEDM) in San Francisco in December.
Research teams are set to present on the FinFET approach--called tri-gate by Intel--on fully-depleted silicon-on-insulator (FDSOI) and on bulk planar processes at around 20 nm and beyond.
[Get a 10% discount on ARM TechCon 2012 conference passes by using promo code EDIT. Click here to learn about the show and register.]
Intel is set to deliver a paper on its 22-nm FinFET technology for SoC applications. In the same session, a research team drawn from CEA-Leti, STMicroelectronics, IBM, Globalfoundries and Renesas will present a paper on ultra-thin box and body (UTBB) FDSOI transistors for a multiple threshold voltage strategy at 20 nm and below.
ST will also report on switching energy efficiency in the UTTB process while IBM will describe a 22-nm SOI process. Meanwhile, Samsung researchers will deliver a research paper on the extensibility of its bulk 20-nm planar HKMG process.
Intel is already making processors using a 22-nm FinFET manufacturing process technology. It has described that process as a CPU process that was not optimized for lowest power consumption whereas the subject of the IEDM presentation is called an SoC process. Intel will provide engineering details of its 22-nm tri-gate SoC process and discuss its use of the approach to build a technology platform for SoC applications. That implies broad families of high-speed, low standby power and high voltage tolerant transistors, as well as RF and mixed-signal capabilities, according to the paper's abstract.
High-speed logic transistors have sub-threshold leakages ranging from 100-nA per micron to 1-nA per micron, while the low-power versions feature a leakage of less than 50-pA per micron. Nonetheless, the process retains 1.8- and 3.3-volt transistors for analog circuits, and legacy circuits.
The Intel 22-nm SoC platform also includes carbon-doped oxide interconnect and three different types of SRAM bit cell to provide options between density, performance and low voltage operation, according to the the abstract.
In another session on Dec. 11, a paper authored by a team from IBM, STMicroelectronics, Globalfoundries, Renesas, Soitec and CEA-Leti will report on another SOI process at 22-nm known as ETSOI for extremely thin silicon-on-insulator. This process has a silicon channel for n-type transistors and strained silicon-germanium channel for p-type transistors.
IEDM runs from Dec. 10 to 12 at the Hilton San Francisco Union Square.
My foundry contact has studied Intel's 22nm SOC FinFET manufacturing flow and claims that it is not cost effective (compared to foundry 28nm/32nm) to manufacture mobile chips hence even Intel over next 3 years will still make most of its Infineon mobile cell phone chipsets at TSMC.
Chips such as the 2 main chips in Apple's iphone 5: MDM9615 and A6 have an average ASP of $20 compared to similar die size X86 that Intel sells for $100-200)
I hope your foundry contact is an ex-intel guy and he knows what he is comparing to. Most likely he is comparing what their FINFET will be in the COST sense. I will never doubt Intel's manufacturing cost much or significantly lower than foundry offering. Intel's biggest expense might not be in wafer manufacturing, they also need to cover IP/EDA tool development/product design teams, in addition marketing cost..... You should consider Synopsys/Candence/ARM/Virage../TSMC/SPIL/ASE../Qualcomm/nVidia.... before you start calculating the selling price. Too bad Intel can't just open its technology/manufacturing capability for design houses. You might want to check the baseband/AP SOC chips Clover Trail are made at TSMC or not? How about even 32nm SOC Medfield chip in TSMC? Who cares about legacy chips at TSMC? Also x86 atom is in the range 20x2 =40 price range. Nobody is using server chip for mobile phone. The war is getting more excited finally.
I have seen others post this as well. Can you address why repeating the intel spin on "manufacturing show down" makes any sense when today most advanced atom SOC is on 32nm and will stay that way until end of 2013 ! while Qualcomm is shipping 50M units 28nm this year in just the iPhone 5 design alone and
(2) I thought all you guys were writing about intel winning mobile market with 22nm finfet last year (2011) and intel has not demo anything close to A6 CPU or graphic power/performance (at mobIle power level) .... Even demoed something close to what is already shipping !
I was not aware that Intel has a position on a "manufacturing show down" and therefore did not try to repeat it.
My intention was to tell readers that papers will be presented at IEDM on many of the leading-edge process technologies that will be producing chips in the near future. And therefore San Francisco in early December would be a good place to find out more.
The rigor of the IEDM review process means that these should be detailed, quantitative presentations albeit likely to be only on certain aspects of each of these processes.
From my perusal of the program the 14-nm logic node is being discussed mainly by academic groups looking at aspects of doping and strain in individual transistors.
To your second point. All what guys?
22nm not cost-effective ? I suppose that depends upon the level of integration. While it's probably not a cost-savings to replace small individual 32nm devices with 22nm, there is a definite cost-savings thru higher integration at smaller process geometries to reduce PCB area, total BOM, and possibly power-savings as well.
Confirmed--Intel baseband will use foundry (mostly TSMC) for next 2-3 years since Intel's internal cost is too high. This is because Intel's 22nm to get to yield has more restrictive design rules (increases die size). Restrictive design rules are on logic, analog, I/O, and back-end metal. Intel will never have competitive baseband chips in its 22nm SOC.
Same is true for Intel's atom line, 22nm SOC Intel chip name Valleyview)...it is that part again due to the restrictive design rules that will not be competitive on cost.
So why not mix and match these techniques using the best for the type of circuits.
Finfets for fast transievers with power gating when not in use. Others that do not need the speed or low power switching can be other types.
I was drawing finfets in my notebook in the early 90's sitting in ee classes.
I am little confused by the title “Intel, Rivals gird for IC manufacturing showdown” because IEDM (International Electron Device Meeting) has not been a forum for IC manufacturing showdown, instead mainly for new research devices, new transistor analysis techniques, device physics, scaling limits…etc. Furthermore, among the three major technologies for 22/20 nodes, FDFinFETs by Intel, FDSOI/UTTBB by IBM Alliances and planar bulk Si by Samsung to be presented here at IEDM, Intel is the only one manufacturing its FDFinFETs for several months now. IBM and Samsung have not announced yet when their technologies will be manufacturing. Therefore, in my opinion the word manufacturability would be more appropriate than “manufacturing” because manufactrability will become the determining factor for ultimate CMOS scaling for 22/20nm nodes and beyond. In order to have manufacturability assessments the transistor electrical characteristics such as VT, dId/dVg, dId/dVd, DIBL, and SS (sub-threshold slope) should be measured, and used also as minimum criteria for paper selection. I have been attending IEDM for over ten years. The paper selection has significantly deteriorated recently as indicated by a significant number of the papers presented don’t meet the minimum criteria. Skim