10nm chip foundry process coming to Apple partner TSMC ahead of Intel
Apple chip foundry TSMC has given hints for its future road map, and claims that it will start mass producing 10nm chips before the end of 2016, nearly a year ahead of Intel.
At a tech symposium in the northern Taiwanese city of Hsinchu, TSMC CEO Mark Liu declared that 16nm chips have been in production by the company for some time, and research and development for a 5nm process is well underway.
Additionally, the 3nm process has entered initial research phases with between 300 and 400 engineers working on the project. Liu also reports a plan has been developed to bring in academics for development of 2nm processes.
Liu confirmed previous reports of the start of 7nm technology "risk production" in early 2017. Competitor Intel's road map predicts 10nm chips launching the second half of 2017.
Also at the event, Liu noted that TSMC is allocating $10 billion for capital expenditure, and the company's research and development budget has tripled since 2009.
TSMC has supplied Apple with A-series chips since the A8's introduction in Sept. 2014. The company is said to be the sole supplier of the A10 Fusion chip in the new iPhone 7 family, and the "A11" for Apple's 2017 lineup.
In August, TSMC was rumored to be working on the "A11" chip for Apple. Additionally, around the same time, TSMC said that trials of 7nm chip fabrication will begin in the start of 2017, with wide use of the technology by 2018.
At a tech symposium in the northern Taiwanese city of Hsinchu, TSMC CEO Mark Liu declared that 16nm chips have been in production by the company for some time, and research and development for a 5nm process is well underway.
Additionally, the 3nm process has entered initial research phases with between 300 and 400 engineers working on the project. Liu also reports a plan has been developed to bring in academics for development of 2nm processes.
Liu confirmed previous reports of the start of 7nm technology "risk production" in early 2017. Competitor Intel's road map predicts 10nm chips launching the second half of 2017.
Also at the event, Liu noted that TSMC is allocating $10 billion for capital expenditure, and the company's research and development budget has tripled since 2009.
TSMC and Apple
TSMC has supplied Apple with A-series chips since the A8's introduction in Sept. 2014. The company is said to be the sole supplier of the A10 Fusion chip in the new iPhone 7 family, and the "A11" for Apple's 2017 lineup.
In August, TSMC was rumored to be working on the "A11" chip for Apple. Additionally, around the same time, TSMC said that trials of 7nm chip fabrication will begin in the start of 2017, with wide use of the technology by 2018.
Comments
Intel's 10nm process will unambiguously be better than what TSMC labels "10 nm".
The more interesting question is -- how does TSMC's 10nm compare to Intel's 14nm? I haven't been able to find a clear answer to that question.
Great job, TSMC!! You rock!
We could say that TSMC 7nm will be roughly equivalent to an Intel 10nm.
source: http://www.extremetech.com/computing/221532-tsmc-will-begin-10nm-production-this-year-claims-5nm-by-2020
For example, TSMC's 16FF is actually a 20nm BEOL, with Finfets underneath, at the density of 20nm. Hence TSMC's A9 was larger than Samsung's.
Not that Samsung's 14nm was 14nm, it was closer to 18nm.
But before you think Intel are blameless, their 22nm was closer to 26nm, and their 14nm is closer to 16nm at best. And Intel's SRAM is dense, but their other logic less so, hence why some 28nm chips sometimes appear to have transistor densities close to Intel still.
So you can see where we are going here with the "10nm" processes. TSMC's is more like a real 14nm process - denser than Intel's 14nm, but as I wrote, that was more like 16nm really.
But regardless, it's smaller, and this 10nm process is at least a proper shrink at last over the previous process.
Let's not start talking about GlobalFoundries' 7nm process... http://semiaccurate.com/2016/09/26/globalfoundries-7nm-process-isnt-even-close-name/
But my impression is that it's even more complicated than that (see Hattig's post, for example).
Another complication is that there are other variables that affect the ultimate performance of a SOC/CPU/GPU that aren't reflected in that table at all -- factors like TSMC's InFO packaging; IBM's SOI; etc.
intel will remain ahead for the next three or four years, at least. But that doesn't matter unless Apple moves to them.
However, it's not at all clear that Intel will actually make their most modern process available to foundry customers. My guess is that they won't. They'll probably only make the 14nm process available for foundry work and keep 10nm for their own products.
Apple A10 3.3B Transistors 125mm^2
The head of manufacturing has already done analysis about how Apple/TSMC has managed to obtain greater transistor density.
- Apple uses a higher percentage of SRAM in their SoCs (though other research indicates that Intel mis-categorized some of Apple's reg. files as SRAM)
- Intel uses a higher percentage of cells that are larger than the minimum process geometries for performance (tall cells).
- Intel acknowledges that the Apple A8 (TSMC 20nm) and Apple A9 (16nm TSMC and 14nm Samsung) have greater RAW TRANSISTOR DENSITY than similar Broadwell (14nm) or Skylake (14nm) SoCs
- Intel has created their own notion, a normalized transistor density, that shows them somewhat ahead, but requires some questionable reverse engineering of exactly what types of logic and memory Apple has included in each block of their SoCs. See foils 19 and 20.
http://files.shareholder.com/downloads/INTC/0x0x862743/F8C3E42B-7DA9-4611-BB51-90BED3AA34CD/2015_InvestorMeeting_Bill_Holt_WEB2.pdf