PEOPLE WATCHING TSMC’S process development might have noticed a radical change in messaging a few months ago. The foundry not only changed its high-K/metal gate (HKMG) strategy, but it also pulled in the timetables at the same time.
Insiders had been telling us that the TSMC 32nm process was in deep trouble, and 22nm wasn’t looking good either. Then one day, poof, things got better. This quite remarkable turnaround is due to one thing, Intel.
Remember when we broke the news that Intel was licensing the Atom core? This was confirmed at IDF a few months later, but what wasn’t said was how it would be pulled off. Intel has two secrets that it considers crown jewels – process technology and its CPU cores. Intel has never given process tech out to anyone, and if you think it licenses cores, just ask Microsoft how far it got with integrating the XBox chips. Since the 486 days, core licensing has been forbidden at Intel.
The problem this time is that ARM, and to a lesser degree, Global Foundries have pushed Intel hard on performance per watt, and are killing it on the custom silicon front. Canmore and Sodaville just don’t cut it against things like OMAP and the 73 other variants in the same space. Intel can’t match ARM’s breadth of partners and the accelerators they add.
So, Intel did what it had to and opened the core to partners. Now people can add anything they want to Atom, and get x86 plus widget as well as the traditional ARM plus widget. The problem is where do they fab it? If Intel makes the chips, it has to give up more of its design tools and process tech than it is likely comfortable with doing.
In the crown jewel stakes, process tech is more precious than cores, which is why Intel opened the core up. TSMC making the parts is an extension of this. No process leakage, and only some core leakage. Since Intel is likely only giving its partners the interfaces to its black box cores, there isn’t even much of that.
Getting back to TSMC though, what happens if the chips can’t be made because Intel’s fab partner isn’t up to speed? All of the people who bought into the Intel Atom++ scheme suddenly get burned, badly. Catch-22 time, make the chips in-house and give up more secrets, or drop the program.
Intel tap danced through this minefield in a very interesting way. It gave some process tech to TSMC in order for it to fab the Atom++ parts. That was the only thing it could do, and the lesser of two evils. That said, it didn’t give the foundry everything, and certainly not the best parts of the process, just enough to get the job done.
So, TSMC now has a process that allows the Atom variants to be made. This remarkable turn around from TSMC wasn’t exactly home grown. It had a helping hand in the background. In any case, it is more than enough to do what needs to be done. TSMC has it’s process, Intel has it’s partner, and ARM has competition.S|A
Update: Intel is officially denying this story.
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