TSMC, SMIC, IBM, Samsung, GlobalFoundries, and UMC. Four uncertainties. Also Intel's SoFIA.
@
Martian,
So how many generation/years that China foundry left in term of technology by leading global competitors?
28nm (TSMC 2011, SMIC sometime in 2014) volume production
20nm (TSMC in January 2014) volume production
14/16nm (TSMC 16nm FinFET in fourth quarter 2014) risk production in Dec. 2013 with volume production in fourth quarter 2014
10nm (TSMC says they are ready to go in fourth quarter 2015) risk/trial production before ramp up
The second citation from Electronics360: "TSMC ramped the volume supply of 28nm wafers to customers in 2011 which
puts SMIC about three years behind TSMC but only a couple of years behind Globalfoundries and UMC."
Taiwan Semiconductor Manufacturing Company (TSMC) is the world leader in RISC-based ARM semiconductor chips for mobile phones and tablets. TSMC started volume production of 20nm in January 2014 (see post #3052 above). TSMC will begin volume production of 16nm FinFET (ie. 3D chip architecture) in the fourth quarter of this year.
I can only give you an answer based on the current snapshot. If SMIC is able to proceed as fast as TSMC then SMIC is three years behind TSMC, which introduced volume production of 28nm in 2011. GlobalFoundries and UMC did not start small-volume production of 28nm until early last year. They are both in the process of increasing their yield. This puts SMIC about two years behind GlobalFoundries and UMC.
SMIC, GlobalFoundries, UMC, and Samsung license the 28nm fabrication technology from IBM. IBM has shown test-wafers of 20nm and 14nm. However, it is not clear whether the IBM wafers were memory chips (which are simple repeated patterns) or the more advanced logic chips (which are more difficult to manufacture). With a few minor exceptions, TSMC generally does not manufacture memory chips due to the lower margin.
IBM's problem has always been taking the technology to build one cutting-edge wafer in the lab and transforming it into a reliable mass-manufacturing process. Ten years ago, IBM went into the foundry business to compete against TSMC. IBM wooed away TSMC's largest customer NVIDIA.
However, IBM failed miserably and was late to market by six months. Due to IBM's inability to reliably mass-manufacture the semiconductor chips on its wafers, ATI (which remained a TSMC customer) surpassed NVIDIA and became the world's largest graphics card seller for that year. NVIDIA learned its lesson and moved its orders back to TSMC.
The reason I can't give you a good answer is that there are too many uncertainties. IBM's 20nm and 14nm technology (both gate-last) are different from its 28nm technology (gate-first). IBM was stuck at 28nm for an extra two years in comparison to TSMC (which has been consistently gate-last). Can IBM make a smooth transition from gate-first 28nm to 20nm gate-last?
Without additional problems, IBM is currently two years behind TSMC. Technically speaking, IBM provides "consultation" services. However, the Common Platform Alliance members (ie. IBM, Samsung and GlobalFoundries) are all relying on IBM's expertise. IBM has individual licensing side-deals with UMC and SMIC.
The issue is further complicated by IBM's recent effort to sell its chip business.
In conclusion, there are four large uncertainties. Firstly, IBM has a history of encountering delays in transitioning from a single lab wafer to mass manufacture. Secondly, IBM's partners are currently two years behind TSMC and their yield (70%?) is probably still lower than TSMC (95%?). This means TSMC still has better technology at 28nm, lower cost for itself and customers per wafer, and higher profit margins. This gives TSMC an advantage to invest more money into R&D for the next generation of semiconductors.
Thirdly, can IBM make a smooth transition from 28nm gate-first to 20nm gate-last? Finally, will IBM's effort to sell its chip business result in delays for the Common Platform Alliance, UMC, and SMIC?
There is a fifth major problem. Let's say SMIC catches up to TSMC in manufacturing technology. Does that mean it's a level playing field? No. TSMC has 30 years of libraries, IP (intellectual property), and tools. That is the reason Intel's SoFIA chips will be manufactured at TSMC in 2015.
TSMC's advantages in addition to cutting-edge 20nm RISC-based process technology.
"TSMC is the world’s largest dedicated semiconductor foundry, providing the
industry’s leading process technology and the foundry’s
largest portfolio of process-proven libraries,
IP,
design tools and
reference flows."
----------
"Intel's SoFIA Could Be A Real Bombshell - Seeking Alpha
Dec 2, 2013 -
SoFIA is Intel's integrated applications processor, cellular baseband, and ... The shocker, however, is that this chip will be built at TSMC."
----------
"How Intel is buying, building a piece of the tablet market | PCWorld
Jan 18, 2014 - Intel has an ambitious goal for 2014: get its Atom chips into 40 million tablets, ... Bay Trail until new Atom chips code-named Broxton and
SoFIA come out in 2015. ... not in Intel's own fabs but by contract manufacturer TSMC."