Virage Logic already has
TSMC 28nm silicon and SRAM is a major part of GPU architecture. ATI uses Virage SRAM. I believe there are something like 300k+ instances on the 40nm ATI GPU product.
Nvidia is TSMC 28nm. Xilinx and Altera are TSMC 28nm. Qualcomm is both TSMC and GFI 28nm. My guess is that AMD/ATI will do both TSMC and GFI 28nm to cover capacity.
TSMC will have 2 Giga Fabs @ 100k wafers per month per fab available for 28nm. GFI has one Mega fab in Dresden @ 60k wafers per month and another Mega fab being built in NY.
I think the biggest GFI advantage is that they are not TSMC. Technology and capacity wise I still bet on TSMC.
I read and enjoy your blog, BUT... that TSMC piece reminded me stongly of a Theo Valich article ... stepping outside perceived boundries of objectivity is problematic in a tech blog.
Hence that siren call to rationalize that almost inevitably leads to planet tarbaby ...
1) "Virage Logic already has
TSMC 28nm silicon" ---- that referenced article's title is "Virage Logic Announces Availability of a Full Suite of 28nm SiWare(TM) Memory Compilers and Rollout of SiWare(TM) Logic Libraries for Leading Edge Customers" and I saw this statement "In late 2009, Virage Logic announced its first 28nm test chip tapeouts", but what I DIDN'T see was any statement Virage Logic actually has 28nm test silicon in hand (which would require TSMC having an actual 28nm line up and running). Nor did I see that asserted in your blog. They may have it and you may know it, but I have not seen it written ... or you may be wishing into existence an event that has not yet materialized.
2) "and SRAM is a major part of GPU architecture. ATI uses Virage SRAM. I believe there are something like 300k+ instances on the 40nm ATI GPU product" ---- I didn't reference 28nm memory, I referenced 22nm memory, which Intel is currently sampling.
3) "Nvidia is TSMC 28nm. Xilinx and Altera are TSMC 28nm. Qualcomm is both TSMC and GFI 28nm. My guess is that AMD/ATI will do both TSMC and GFI 28nm to cover capacity." - so? I said several customers, specifically ARM based design, 28nm low power customers have climbed into bed with GFI ---- I assume they did so based on information I most definitely and you most likely do not have access to - and there is no more time critical~absolute cutting edge processor market that that one. There's a reason ALL the major ARM chip designers are climbing into GFI's bed.
4) "TSMC will have 2 Giga Fabs @ 100k wafers per month per fab available for 28nm. GFI has one Mega fab in Dresden @ 60k wafers per month and another Mega fab being built in NY." ---- My assertion was who will be FIRST to production silicon, not who will eventually win the volume race ... though I wouldn't count GFI out there, depending on TSMC's execution and Global Foundries new Saratoga facility is specifically designed to easily accomodate a tripling of capacity. If TSMC stumbles on 28nm and GF doesn't, that reserve capacity will probably be activated posthaste.
5) "I think the biggest GFI advantage is that they are not TSMC. Technology and capacity wise I still bet on TSMC." ---- which, if true, can only mean TSMC is in hot water with a number of dissatisfied customers ... at 28nm and below, I will bet on GFI on technology and call it a toss-up on capacity.
Then there's SEMATECH's EMI Partnership lauched at CNSE’s Albany NanoTech Complex (the proximity to which was a major factor in GFI's new fabrication facility location - that and proximity to IBM's headquarters), formed to share the costs of developing future node technologies between several industry leaders, notably including IBM and Toshiba, and which GFI will be a key player in, and which will be a
V-E-R-Y formidable factor going forward in the fabrication industry. I have not read that TSMC is going to be a part of that effort.
That alone is going to make it very very tough for TSMC to successfully compete with GFI on future nodes.
... or 22nm memory chips ... 
Facebook
Twitter