GlobalFoundries 2014: a Year of Change

GlobalFoundries at the end of 2014 is a very different company from what it was a the beginning of the year.

At the start of 2014, GF was a company with:

• a CEO in Ajit Manocha who was reputed to be just a safe pair of hands while the company found a new CEO
• several 200mm fabs in Singapore (the old Chartered fabs) running mature processes, and one small 300mm one
• a 300mm fab (fab 1) in Dresden, Germany not capable of running leading edge processes
• a fab (fab 8) under construction in Malta, NY for 20nm, 14nm and beyond scheduled to begin volume production late in the year

They were very late to 28nm, essentially conceding the entire market to TSMC during the first couple of years when the bulk of the money is made. Since everyone agrees that 28nm will be a long-lived process, late is a lot better than never, and the process now seems to be shipping in volume. However, rumors were that their 14nm process development was not going well and was, like 28nm, likely to be late assuming the process eventually yielded acceptably.

Then 2014 started and everything changed.

In January GF appointed a new CEO, Sanjay Jha. His background was at Qualcomm (where he rose to be COO) and Motorola in the mobile business, which since it is the largest semiconductor market ever seen could turn out to be an asset.

Fab 7 in Singapore was upgraded by merging it with the neighboring fab and upgrading everything to 300mm. This is still used for running non-leading-edge processes such as BCD and analog. However, with high-capacity 300mm the economics are much improved, especially for process steps that operate across the whole wafer at once (as opposed to lithography patterning which proceeds a die at a time). Other foundries have also been upgrading their non-leading edge fabs and processes since there is gold in them thar hills.

Next in April, GF announced that they were licensing Samsung’s 14nm FinFET process and would be a true second source in the sense that companies (in particular one whose name is a fruit) could go to either or both of companies for production.

In October at ARM Techcon I attended a panel session where Kelvin Loh of Samsung and Shubhankar Basu of GF presented and the technology transfer seemed to be on-track. Since then there have been rumors about Samsung slipping ramp-to-volume for 14nm but, if anything, that should just make the transfer easier giving it an extra few months to run test wafers.

Then in September and even more significant deal was announced. GF will acquire IBM’s semiconductor division for a sum of…-$1.5B. That’s right, IBM will pay GF billions of dollars to take it off their hands. GF reckons that they can run the business profitably even though IBM could not since foundry is their business and they can run other product in the fabs in a way that IBM was not flexible enough to do. Plus they have IBM as a captive customer for a minimum of 10 years.

The IBM deal comes with two fabs. The old non-leading edge 200mm fab in Burlington, VT running BiCMOS, RF and all sorts of other esoteric stuff; and the leading-edge 300mm one in East Fishkill, NY. It also comes with a lot of people. With IBM having a huge layoff there are also rumors that a lot of extra people got stuffed into the major deals just before the layoff, namely the sale of the low-end server division to Lenovo and the semiconductor division to GF. It remains to see if the economics work.

So at the end of 2014 GlobalFoundries had:

• a new CEO, who (according to people I’ve talked to) has done a good job of making the company much more focused
• upgraded fab 7 in Singapore to 300mm with a capacity of 50,000 wafers per month (wpm)
• the old AMD fab 1 in Dresden with a capacity of 80,000 wpm
• fab 8 in upstate New York with a peak capacity of 60,000 wpm
• a 14nm process from Samsung
• IBM’s semiconductor business and a huge capacity in R&D (of course technically the deal hasn’t closed yet so this statement is really jumping the gun)
• SOI capacity, although I don’t really know if this is of any interested beyond IBM’s server business which is dependent on it
• a 300mm fab in East Fishkill although with a capacity of only 14,000 wpm
• a 200mm fab in Essex Junction VT with a wide portfolio of specialized processes (capacity of about 40,000 200mm wpm)

Quite a transformation in 12 months!