There have been a few articles floating around [many just requoting the original] that suggest that nVidia is looking to move away from TSMC and move to GlobalFoundries for production of their GPUs.This is not really a big surprise to anyone that has been following the issues that have plagued the 40nm process at TSMC. What caught my eye was the fact that the announcement was not about an agreement, it was simply that nVidia is in “high-level” talks with the AMD Spin off. This usually means that two companies are discussing the generalities of what they want from each other. For nVidia I am sure it will be issues such as volume, process, and of course pricing. GlobalFoundries will be looking to get, from nVidia, times lines for implementation, expected volumes and of course pricing.
These are never the nuts and bolts of a deal; they are general expectations from each party. So why make an announcement? If this was GlobalFoundries making the statement I would say it is a psychological move aimed at showing other fables companies they can handle the big players. However, the statement was given by no other than Jen-Hsun Huang, in his interview with Chinese site Expreview.
"GlobalFoundries is a leading silicon foundry with advanced and outstanding processing technology. We’re seriously evaluating and discussing about the possibilities of working with them. As to TSMC, we do regard it as a world-class silicon OEM with flexible strategies. We’re working very closely together."
To me this signals a gauntlet thrown at TSMC’s feet, the challenge? Fix your issues or we can and will go to someone else. The fact that he was cited in saying "outstanding processing technology" speaks volumes of how highly AMD was regarded in terms of manufacturing – but the resources were never there. Now that GlobalFoundries is all on its own, things are changing. GlobalFoundries can have a 35% of the company owned by AMD, but if GlobalFoundries gets the chance to sign nVidia with a potential of manufacturing more than a billion chips per year [GeForce + chipsets + Tegra], there is no fiscally-aware company that would walk away.
As I mentioned 40nm is a disaster in terms of leakage. TSMC is spending money to find the problem and correct it; this is certainly going to drive up the costs associated with large scale manufacture, and if yields are also low then companies using the TCMS 40nm process are losing even more money. The problem with TSMC is the fact that the company announced that they’re going to spend 10 billion dollars to overtake Intel in terms of manufacturing. That is three billion more than Intel is spending, with a certain discrepancy between the real hard cash that Intel is spending and various promises on the TSMC side. We spoke with the tooling manufacturers and they still consider Intel to be number one in paying for the most advanced tools you can think off, with GlobalFoundries now fighting to claw that title – back at the day, tooling manufacturers even demanded that Bob Rivet, AMD’s CFO of the times – steps down due to money AMD owed for the tools in Fab36 facility.
nVidia is not in what I would call a good financial situation right now and they do not need to lose more money on a bad process. So this move and announcement is a psychological move towards TSMC and a testing of the waters with GlobalFoundries.
If things go nVidia’s way they should see one or both of two things from TSMC; a better 40nm process and reduced pricing. With GlobalFoundries they have options for beyond 40nm. After all if TSMC can’t get 40nm right, which is their half-node die shrink from the never uni then 32 and 28 are probably not going to be great either.