By: Jouni Osmala (firstname.lastname@example.org), August 18, 2014 12:21 pm
Room: Moderated Discussions
> I'm more sympathetic to the finance story than anything else I've heard so far --- the primary
> problem is simply that Intel's 14nm transistors cost a whole lot more than its 22nm transistors.
> I would say, however, that that agrees with my initial point --- that Intel and TSMC/Samsung
> are playing the same game here, and are not THAT far apart in capabilities.
> It's not useful (ie does not lead to an accurate understanding of the world) to claim that Intel has
> 14nm working in a way that is SUBSTANTIALLY different from TSMC/Samsung if Intel is not able to use
> its 14nm in an economically productive fashion. If Intel's choices for its new process are
> - Atom (low power is valued, but no-one is willing to pay Intel prices for it) OR
> - Broadwell mainstream (yield is so awful this is not viable) OR
> - manufacturing a part the demand for which is basically hypothetical,
> probably very small, and perhaps sold at a loss
> that does not realistically strike me as being way ahead of where the competition are at 16nm.
They really wanted high enough profit part that has small enough die size in their first
part when defect density wasn't low enough.
Haswell Y has 130mm^2 And mainstream haswell is about double that.
Broadwell Y has shrunk to 82mm^2
While I googled several different yield calculations from different sources the die size relative to defect density is risen to higher power in the function. So inorder to get same yield with mainstream broadwell as with broadwell Y the process needs to half the defect density. And if broadwell Y has mediocre yields then mainstream broadwell has TERRIBLE yields. On the other hand if you shrunk apple A7 to use this process you could have higher defect density than with broadwell Y and get away with it.