The challenges in the supply chain logistics for semiconductors continue to increase, but the 10nm node will give one more opportunity to gain large benefits from technology scaling.
In March 2012, International Business Strategies (IBS) projected that gate costs at 20nm and 16/14nm would be higher than previous generations of technology. The analysis of the gate cost for 10nm now exhibits a different pattern, shown in the figure below.
While IBS did not project that scaling would stop, it predicted that cost penalties will occur with the adoption of 20nm bulk CMOS HKMG and 16/14nm FinFETs. These predictions on gate cost have been shown to be correct, and while 20nm products are in high volume for Apple, wafer capacity at 20nm is much lower than at 28nm.
TSMC provides another example. Its wafer capacity at 28nm is 150,000 wafers per month (WPM), but its wafer capacity at 20nm is nearly a third as much – 60,000 WPM. Globalfoundries also has 20nm capacity in its Malta, New York fab but the primary emphasis of this facility is on FinFETs. As for Samsung Electronics and UMC, they have decided to bypass 20nm.
While 16/14nm wafer volumes are ramping, wafer capacity at 16/14nm is again lower than 28nm. The wafer volume at 16/14nm is also driven by Apple again, but the length of use for 16/14nm technology will be determined by how rapidly 10nm will occur.
The lower gate cost at 10nm is due to the higher gate density that can be obtained compared to the increase in wafer cost. To attain lower gate cost at 10nm, there will be the need for high systemic and parametric yields, but this is achievable.
The expectation is that 10nm will be a high volume and long lifetime technology node. TSMC and Samsung are projecting risk production for 10nm in Q4/2015, and the customer target is clearly Apple. If there is the ability to ramp up 10nm in 2016 or even in mid-2017, 16/14nm will be a short lifetime technology node.
The capital expenditure required for 10nm, however, is approximately $2 billion for 10,000 WPM, and a facility running 40,000 WPM will cost $8 billion. Also, the minimum cost for a design at 10nm will be $150 million, so if revenues for a chip need to be ten times higher than design costs to get a good return on the investment, 10nm chips will need to achieve sales of $1.5 billion.
After 10nm, there will likely be the need for extreme ultraviolet (EUV) technology, and there is steady progress on enhancing EUV throughput. While 450mm wafer technology continues to be worked on, its introduction will not likely occur before 2020.
— Handel Jones is founder and CEO of International Business Strategies Inc., which provides custom studies in multiple areas of the electronics industry.
- PTT 在德國法蘭克福 - 2007
- 原來我早就知道 Robert Scoble - 2006
- Nokia: BenQ, Siemens 'turkeys will not make an eagle' - 2005
- 為什麼我會有勇氣問妳？ - 2001