IBM has agreed to pay Globalfoundries $1.5 billion to take its chip business off its hands but is retaining the responsibility and cost of advanced semiconductor research. In a conference call to discuss the proposed sale John Kelly, senior vice president of research at IBM, and Sanjay Jha, CEO at Globalfoundries, presented the arrangement as the most natural thing in the world.
Kelly even spoke of semiconductor R&D making its way from IBM sites around the world to Albany, New York, where it would be implemented by Globalfoundries and then be passed back to IBM in the form of chips to be used in its systems. It's a complicated route, albeit one that ties IBM and Globalfoundries together for the next ten years. That is the time that Globalfoundries is expected to execute the manufacturing plan that had been mapped out by IBM.
But this arrangement is unusual and needs some explanation or is the current plan merely a stop-gap that will gradually see Globalfoundries taking on more responsibility for sub-10nm research?
My tenet is that semiconductor R&D is conducted most naturally and efficiently by the chip manufacturer. That is how it is done at Intel, TSMC, Samsung and other places. The chip manufacturer has the direct feedback of what works and what doesn't in the previous generation and the chance to amortize the cost of that R&D in chips sold.
IBM carrying extra weight
IBM, not known for being the most agile of companies, is presenting itself as a systems and software company. But by holding on to advanced semiconductor R&D it is loading itself with billions of dollars of extra cost making it less competitive than its peers. You don't see Cisco, Google, Hewlett Packard, Microsoft or Oracle paying billions of dollars to be a semiconductor think-tank for the rest of the world.
Now, there is much discussion about the necessity for ecosystems to emerge to support the extreme complexity of leading-edge semiconductor design and manufacturing. And on that basis it could be argued that all this stuff needs to get done somewhere and keeping the research with IBM and letting Globalfoundries achieve greater economies of scale in the manufacturing is the most efficient place to go to from here. It could be argued that trying to disentangle the semiconductor research and people from the groups at Zurich, Almaden, Yorktown Heights and Albany would be difficult and ultimately would lead to the breaking up of the network and the destruction of value.
But nonetheless portraying the proposed arrangement as sensible is ignoring an elephant in the room.
It also remains unclear how much chip design is going to stay with a newly fabless IBM and how much will transfer to Globalfoundries. If the latter were the case that would cloud Globalfoundries' role as a pure-play foundry.
In terms of an explanation there are possibilities beyond the purely commercial. One is that amount of money provided by New York state to IBM over recent years may mean that changes of greater significance might invoke attempts to claw back funds. A second is that the United States government considers IBM's advanced semiconductor research to be of strategic significance and has let it be known that control must not be transferred to company owned by an Abu Dhabi sovereign wealth fund. However, strike one against that argument is the fact that IBM is prepared to transfer 10,000 semiconductor patents to Globalfoundries as part of the deal.
All in all it is an enigma that hints at unsustainability and it is one that both IBM and Globalfoundries have chosen not to address, so far.
Related links and articles: