LONDON – Management changes recently announced at Globalfoundries Inc. are related to problems with 32-nm chip yield and the slowness of creating a foundry-like operation at Globalfoundries' Dresden facility in Germany, according to analysts at Nomura Equity Research.
Globalfoundries announced June 16 that Doug Grose, CEO and Chia Song Hwee, COO, would be stepping down from their posts and the appointment of Ajit Manocha as interim CEO. Globalfoundries itself cited "demands from customers for more capacity, faster technology development and greater agility," in its press release.
Yield problems at 32-nm were driving processor vendor Advanced Micro Devices Inc. (AMD) into the arms of rival Taiwan Semiconductor Manufacturing Co. Ltd. (Hsinchu, Taiwan) at the same time that Globalfoundries was struggling to reduce its reliance on AMD as lead customer, the analysts indicated in a note issued to clients. The transfer of the manufacturing operation out from AMD was the mechanism that created Globalfoundries in March 2009.
The note was issued mainly for the benefit of clients invested in AMD and concluded that the changes at Globalfoundries would have minimal impact on AMD, partly because AMD recently recut its deal so that it would only pay for good die coming off the 32-nm line.
The authors of the notes said the announced management changes reflect the past execution in changing the Dresden facility from a one-customer, few-product wafer fab into a traditional many-customer, many-product foundry.
"As it relates to AMD, we believe that the majority of the 32-nm issues have already been addressed. AMD is currently shipping the 32-nm Llano A-series APUs for production. And, while 32-nm yields are still not optimal, we do not foresee AMD materially under-shipping demand in the second half of 2011. We also note that AMD recently ratified its pricing arrangement with GF to provide downside protection by allowing AMD to pay only for good dies,” the note said.
@peter.clarke: some heads roll at GF so it is natural every one wants to know the inside scoop.
Daniel Nenni had the guessing game (TSMC's yield at 28nm) for a while on Semiconductor Professionals @LinkedIn. I don't remember the final consensus (correlated?) yield figure but it was no where north of 90%.
According to this other post also by Nenni, it would seem quite difficult to get yields above 80% as one goes to 32/28nm nodes:
In addition to the continuing statistical effects, it appears proximity and stress effects are more pronounced as the nodes are shrunk.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.