It seems a bit dubious to generalize that TSMC is having loads of 40nm trouble yet ATI has a fair amount of cards out there. Sure yields could most likely be better, but I wonder if nVidia's late arrival to the party with buggy, over complicated design/masks could be skewing the statistics?
The kind of yield issues TSMC is now experiencing with their 40nm line are typical of production volume scaling...as you add new tools to the line to make it "wider" you end up with a mathematical permutation of the number of combinations of toolsets that can end up being used for any given lot (minimum group of wafers that travel thru the fab)...do to intrinsic tool-to-tool variations the same two tool models designed to do the same thing to a wafer will have a slight mismatch in the output...after multiple wafer passes thru the tools the end result can be serious yield loss.
Every fab and every company faces this issue when scaling up the production volume of a process technology. TSMC being at the leading edge in a very visible fashion like this is the only reason we are hearing of their trials and tribulations in the public domain.
At any rate, this kind of yield hit does impact AMD as the temporary solution (called "band-aids" in fab-speak) is to restrict the toolsets allowed for production on any given lot, which then increases CT (cycle-time) for the WIP (work-in-progress) in the fab and any increase in CT means an increase in defect-density which means a further reduction in yield.
In short this sort of thing is never good for anyone.
It is obviously not in production - it is scheduled to go so in December hence my long-standing prediction of a January-February shipping date.
As for ATI if they have contracted TSMC for X amount of production they are getting it, period and if TSMC is indeed the one who's having problem engineers will probably focus on production yields, not pilot runs...
Those are two functions that are already serviced by two discrete sets of resources (both budgetary and headcount)...they operate in parallel and aren't really amenable to being reallocated on the fly to assist the crisis that might be going on in the other team's module. (regarding production yield versus pilot runs)
Only in the most extreme of situations - a fab shutdown - do the production yield team's decisions impact the ongoing day-to-day workings of the pilot run team, and even in that case it doesn't really change the focus of the pilot team as much as it puts them and their efforts on hold (and generates much much much angst to the customer side of the business development equation) while the yield team scurries about in 24/7 crisis mode until the fab hold is lifted.
I've seen this happen twice, once in Houston and once in Singapore and both times it was actually a contamination issue (meaning device reliability is compromised, not so much a yield issue as we define yields) that made things go
that awry.
Facebook
Twitter