It was just over a year ago that we toured GlobalFoundries, heard the company talk about its plans for 7nm and its 22FDX and 12FDX process lines, and thought — at long last — that the firm had righted itself. Between then and now, GF announced it would quit 7nm development and redouble its efforts to create a differentiated business around its 22FDX and 12FDX products that use fully-depleted SOI (FD-SOI) and are intended for low-power devices. Now, there are rumors that ATIC — the investment arm of the Abu Dhabi government — simply wants out of the project altogether.
GlobalFoundries has never come anywhere near fulfilling on the promises it made during or after AMD’s spinoff. In its early days, the company promised its alliance with IBM and Samsung would allow it to compete more readily with TSMC. It would bring its new New York plant online, add additional capacity at 40nm to deal with bulk customers that AMD hadn’t previously addressed, and compete aggressively with TSMC for 28nm business. Virtually none of these things happened the way they were supposed to, or within the contemplated timelines.
After canceling its own 14nm, GF found some success as a second source for Samsung. AMD’s Ryzen CPUs have all been built on its 14nm production lines, as is AMD’s Polaris GPU. A year ago, it genuinely seemed as if GF might hae finally, finally turned the corner. It had a 7nm strategy. It had its own process node. It had a product differentiation play, with FD-SOI. But GlobalFoundries has also never been profitable. Despite growth its own revenues, the constant cycle of tool upgrades and capital investment have left ATIC on the hook for billions in losses. This graph from Anandtech shows what we know (GF is not a public company and is not required to make the same types of disclosure).
GlobalFoundries recently announced the sale of its 200mm Fab 3E in Singapore, where it built MEMS and analog/mixed-signal chips. 200mm fab space has actually been at a substantial premium of late, and interest in MEMS and other devices has been rising, thanks to the advent of 5G and IoT. But there are rumors that ATIC wants to unload the company altogether after a decade of losing billions of dollars.
It’s hard to fault the investors. AMD and ATIC may have slapped a brave face on it, but AMD was GF’s only large customer, and they lost AMD’s future business at a time when the firm is enjoying a sales surge. It may have made business sense for the two companies to part — AMD may not have been likely to do enough business to make a 7nm fab line make sense for GF to build, given how expensive new nodes have become. Company CEO Thomas Caulfield has blamed GF’s shift away from cutting-edge manufacturing on the dwindling number of customers for such parts, which would put some teeth behind this point.
AMD will continue to use GF for its 12/14nm needs, but those needs will decrease over time as its GPUs move to 7nm. Buying I/O die for Epyc and Ryzen CPUs from GF isn’t the same as buying the entire chip. Failing to compete for leading-edge nodes automatically takes GF out of the running for future consoles or cutting-edge parts.
Now, none of this matters if GF can spin up enough 22FDX and future support for its 12FDX platform to replace the lost revenue. GlobalFoundries did, at least, announce last year that it had hit $2B in lifetime design revenue for 22FDX. It also amended its plans for its Chinese facility to remove the 180/130nm technologies it originally planned to build there, in favor of additional 22FDX output and an accelerated timeline. But it’s not clear how much of a future 22FDX or 12FDX have. $2B in lifetime design wins for 22FDX, which GF announced last year, isn’t a huge win for a foundry that does more than $4B in yearly business. With no cutting-edge business on 7nm, GF needs to be aggressively ramping customer commits for 12FDX. We don’t know if that’s happened, or what the status of the transition is. Given GlobalFoundries’ repeated decade-long problems with node transitions, it’s fair to ask whether the company can bring this one off successfully, either.
Presumably, GF would be chopped up on the auction block and its businesses sold to specialty foundries that wanted to work in those specific areas. It’s not clear who would buy the firm’s cutting-edge facility in New York or what the plan for that facility would be. The market for pure-play foundries has been shrinking for decades, with fewer and fewer firms playing in every product generation.
- IBM, Samsung Ink Agreement to Build Future Power Chips
- GlobalFoundries Departing the Leading Edge is an Ominous Sign for Foundry Industry
- GlobalFoundries Radically Restructures, Kills 7nm, Spins off ASIC Design Team