一些有關20nm製程的資訊:
1. GPU (Nvidia, AMD) 可能會跳過 20nm, 直接從28nm跳到14/16nm FinFET.
2. 20nm製程的漏電流(導致過熱,功耗比變差),還有大晶片的良率不佳.
3. 20nm本來就是過渡製程, Fabless原先就有準備會28nm 直跳 14/16 nm.
4. 所以14/16nm會是往後的新戰場. 28nm跟14/16nm會成為主流. 20nm比例會下降.
- In related bad news for the TSMC 20 nm process folks who are struggling
to make planar fab still work:
"We have learned that high performance discrete GPUs will never
make it to 20 nm, simply as process has too much power leaks and
poor yields with big cores."
- Fuad Abazovic (Fudzilla 02/09/15)
Which, if true, means that GPU houses like AMD and Nvidia will wait
until 2016 for TSMC 16 nm or GlobalFoundries/Samsung at 14 nm. Planar
is pretty much dead. It means those smaller TSMC customers like AMD
and Nvidia are waiting and might jump. And it gets worse:
"Apple is likely to switch its chipset for its next generation
iPhone to Samsung on 14 nm. Qualcomm could transition to 14/16 nm
quickly, making 20 nm node just a transitional node due to several
challenges."
- Warren Lau, analyst at Maybank (Barron's 01/11/15)
Big customers like Apple and Qualcomm make up 30% to 40% of TSMC's
business are also waiting and might jump