- Dec 25, 2013
- 3,899
- 193
- 106
Pretty spectacular.
http://seekingalpha.com/news/317518...ithdraws-resignation-street-debates-amd-china
http://fortune.com/2016/04/22/intel-aicha-evans-staying
PS: To Ashraf Eassa. About Moore's Law and tech leadership.
Yesterday I crunched the numbers and here are my conclusions.
* In short, TSMC and Intel both do something different: Intel is going for >= 2x density improvement, and thus sacrificing TTM because scaling is so hard. TSMC OTOH is going for less aggressive scaling, which has shorter TTM.
* They are not better or worse strategies from the data that I have available. Intel's leadership should still be *at least* 1 year. But that is IF and only IF TSMC operates perfectly (which is doubtful) without yield difficulties, and that's probably also 1 yr against Apple because they will be first, so versus other companies it will be more.
* The big unknown is what technologies the process nodes will adopt. Even if Intel's lead has shrunk to 1 year (density), if they have better transistors, the real lead will still be many years. Intel already has air gaps at 14nm, so that shows how good they are. TSMC's 10nm is finFET but 7nm is unknown, most likely SIGe. Intel's introduction of III-V/Ge and quantum well could be at 10nm, but that is unconfirmed.
We simply don't know.
* Okay, now back to scaling.
TSMC 16nm: 5760 nm²
Intel 14nm: 3650 nm²
(Difference: 1.55x denser)
TSMC 10nm (claim 0.53x so): 3053 nm² (2017) (finFET)
Intel 10nm (claim ~0.5x, let's take 0.52x): 1898 nm² (H2'17) (unknown) [if they achieve 2.1x like expected it will be 1740 nm²]
TSMC 7nm (claim 1.6x = 0.652x so): 1908 nm² (without EUV) (probably H2'18 for iPhone although claim is H1 -- don't forget TSMC was 2Q behind SS at 14/16nm)
So Intel 10nm will be *at least* as dense as TSMC 7nm AND have *at least* 1 year TTM advantage AND have *most likely* a much better transistor.
http://seekingalpha.com/news/317518...ithdraws-resignation-street-debates-amd-china
http://fortune.com/2016/04/22/intel-aicha-evans-staying
PS: To Ashraf Eassa. About Moore's Law and tech leadership.
Yesterday I crunched the numbers and here are my conclusions.
* In short, TSMC and Intel both do something different: Intel is going for >= 2x density improvement, and thus sacrificing TTM because scaling is so hard. TSMC OTOH is going for less aggressive scaling, which has shorter TTM.
* They are not better or worse strategies from the data that I have available. Intel's leadership should still be *at least* 1 year. But that is IF and only IF TSMC operates perfectly (which is doubtful) without yield difficulties, and that's probably also 1 yr against Apple because they will be first, so versus other companies it will be more.
* The big unknown is what technologies the process nodes will adopt. Even if Intel's lead has shrunk to 1 year (density), if they have better transistors, the real lead will still be many years. Intel already has air gaps at 14nm, so that shows how good they are. TSMC's 10nm is finFET but 7nm is unknown, most likely SIGe. Intel's introduction of III-V/Ge and quantum well could be at 10nm, but that is unconfirmed.
We simply don't know.
* Okay, now back to scaling.
TSMC 16nm: 5760 nm²
Intel 14nm: 3650 nm²
(Difference: 1.55x denser)
TSMC 10nm (claim 0.53x so): 3053 nm² (2017) (finFET)
Intel 10nm (claim ~0.5x, let's take 0.52x): 1898 nm² (H2'17) (unknown) [if they achieve 2.1x like expected it will be 1740 nm²]
TSMC 7nm (claim 1.6x = 0.652x so): 1908 nm² (without EUV) (probably H2'18 for iPhone although claim is H1 -- don't forget TSMC was 2Q behind SS at 14/16nm)
So Intel 10nm will be *at least* as dense as TSMC 7nm AND have *at least* 1 year TTM advantage AND have *most likely* a much better transistor.