Global Foundries is having the proverbial ‘issues’ with their high end 32nm-SHP process. The knee-jerk reaction is to kick GloFo for the problems, but that doesn’t take in to account the good partss of the process.
To say this story is complex and nuanced is putting things mildly. The 32nm-SHP process is the first foundry process to ship High-K Metal Gate (HKMG) chips, and it is the first foundry to ship customer products on a sub-40nm process. They are also the only foundry shipping HKMG products with strain, aka a SiGe cap. That is the hard part, compared to strain, the rest of the HKMG process is easy. The fact that AMD has shipped almost 10 million Llano CPUs by now says that something is going right. GloFo is currently making things that no one else can, and with a 6+ month lead on the competition.
On the flip side, there are reports of ~50% yields on the process, Llano specifically. The low end of this range is a 45-50% yield on Llanos, as published by a few financial analysts last week. SemiAccurate’s checks say that 32SHP is about 15% below where GloFo wants it to be. This puts the range for Llano at approximately 50-66%, IE not good. Really not good.
Our checks with some GloFo sources say that the process is improving, and it is workable, but not good. AMD is shipping Llanos by the millions a quarter, but they are also getting about half of what they wanted. Both companies are leaving a lot of money one the table, and neither one is happy.
So that brings us to the elephant in the room, is the 32SHP process good, or is it a dog? The answer is unquestionably, “Yes it is”. The process is both. By what customers want, and by GloFo’s balance sheets, it is a dog, period. If you look at it in comparison to the other foundries out there, none of them can do a 32nm HKMG + strain/SiGe chip at all. So by that metric, the process is 50% better than any of the competition, their yield is 0%. If you need a 32nm HKMG+strain chip, you have only one choice, and the yield is what it is.
A little more on the up side, the difference between 28SHP and 32SHP is minor, almost negligible. This means that GloFo has solved many of the hard problems. There isn’t another foundry out there that has solved these problems and are even close to shipping, and every day brings a further slip from almost every foundry out there. Only one company other than GloFo, Intel, can ship sub-40nm logic parts, and they are not a foundry. Intel only competes with GloFo’s customers, not GloFo itself. For now.
In the end, it is a glass half-full or glass half-empty situation? Is the process a dog compared to past processes? Sure. Is it doing things no one else can? Sure. How you spin it depends on your particular point of view, and which side of the ax you want to grind. That said, work continues.S|A
Latest posts by Charlie Demerjian (see all)
- SemiAccurate has Skylake-SP die shots, sizes, and more - Jul 21, 2017
- Skylake-SP has a diverged core - Jul 19, 2017
- Intel’s Purley platform architecture is a step forward - Jul 13, 2017
- Intel launches Purley aka metal Xeons - Jul 11, 2017
- AMD’s Epyc has lots of connections - Jun 26, 2017