The news that IBM's CEO Ginni Rometty is looking to sell off IBM's chip unit raises issues around IBM's Power processor architecture and numerous other chip companies that have stood on the shoulders of this giant's semiconductor R&D. These include most of the industry's chipmakers with the exception of Intel Corp. and Taiwan Semiconductor Manufacturing Co. Ltd.
The likes of Globalfoundries Inc., United Microelectronics Corp., SMIC, STMicroelectronics – and even the mighty Samsung – have all been involved in manufacturing process collaborations with IBM and IBM's exit is likely to at least lower morale there even if it does not have a direct impact on their immediate sales. With IBM not around it does imply that their cost of R&D in the medium- to long-term is going to go up. Unless one or more of them also withdraw from the market.
The move. reported in the Financial Times, is indicative of the lack of profitability in vertically integrated IC production, which has fostered the fabless-foundry business model. But it also begs the question who could pick up IBM's manufacturing in New York State or the burden of its R&D?
Rometty's move is not a surprise. She will be taking IBM down the route taken before by the likes of Siemens and Philips in Europe and by Hitachi, NEC and Mitsubishi in Japan and the likes of General Electric in the United States. In fact what is more surprising is that the move has not been made earlier. IBM's heavy spending on fundamental silicon R&D and technology has been an anomaly for more than a decade.
It is part of a long-term move by IBM away from hardware and towards software and consultancy that was begun many years ago under previous CEO Lou Gerstner. It fits in with IBM's sales of its PC business to Lenovo in 2005 and with the planned sale of its x86-server business, also to Lenovo, announced last month.
Even if IBM cannot find a buyer for the chip division and only succeeds in selling the manufacturing plants, the intention is clear. IBM can no longer afford to be the semiconductor industry's R&D lab-rat and cheerleader.
Over more than 30 years IBM has been a leading – if not the leading – presenter of learned papers at the International Electron Devices Meeting (IEDM) and the International Solid-State Circuts Conference (ISSCC) stalwart, but it is the software that makes IBM its money. IBM has also been a leading filer of hardware patents but it looks like that will all be coming to an end.
A really short list
So who could buy IBM's fabs. Exclusing dark horses such as Chinese OEMs on the up and up, Apple and Qualcomm, the list of likely potential buyers is short for reasons mentioned above.
Globalfoundries is top of the list because it is involved with IBM on process development and committed to manufacturing in upstate New York. Also, because it is backed by the Abu Dhabi government through its Advanced Technology Investment Co. (ATIC) investment vehicle, it has the money which many others do not. In other words we have here motive and opportunity. It was reported that Globalfoundries is planning to spend up to $10 billion over the next few years to build a wafer fab and bring up 20nm and 14nm manufacturing capacity. It would be relatively easy to divert a small portion of that $10 billion to include the acquisition of IBM plant.
Certainly foundry market leader TSMC has the money but is an unlikely buyer. It has the opportunity but I see litte motiovation. TSMC was being courted by New York during 2012 and 2013 to put down a wafer fab and roots in the Empire State through something called Project Azalea. TSMC chairman Morris Chang admitted that U.S. locations for a wafer fab were being considered back in December 2012 but eventually pulled back raising the old argument that TSMC does best when its facilities are concentrated together, as they are in a couple of locations in Taiwan.
But what if Globalfoundries chooses not to buy and TSMC doesn't want to? Samsung is a possibility. It has foundry manufacturing capacity in Austin, Texas. That is mainly to satisfy customer Apple, a relationship that is likely to diminish in the short term but nonetheless expanding foundry manufacturing in the United States to support long term goals in a possibility for the consumer giant.
The political solution would be for Intel to buy out IBM's chip business and thereby keep the intellectual property under U.S. domestic ownership. Intel is still the number one chip company in the world and could afford to make such a deal. But would it want to? Intel's own position, developed as a vendor of expensive processors for PCs, somewhat mirrors that of IBM. Intel's profitability has declined along with the PC market, which has gone down under the onslaught form smartphones and tablet computers.
In January Intel said it would leave the shell of Fab 42, in Chandler, Arizona, empty because of that slow down in the PC industry. The shell was completed in 2013 and was expected to start making chips using Intel's leading-edge 14nm FinFET process on 300-mm diameter wafers.
No more the industry lab-rat
The news of the departure of IBM from chip manufacturing will also cast a pall of uncertainty over STMicroelectronics' fully-depleted silicon on insulator (FDSOI) manufacturing process, which it is touting for business as an alternative to planar bulk CMOS and FinFET CMOS.
Much of the research into the performance of transistors made on SOI wafers was done by IBM and it is companies associated with the IBM research community that have shown interest in the process. ST has championed FDSOI and turned it into a manufacturing process. ST has signed up Globalfoundries as a licensed second-source for manufacturing. However, it is possible that ST might come to market with another foundry.
Jean-Marc Chery, executive vice president and general manager of embedded processing solutions at ST, reminded Electronics 360 recently that the licensing agreement for FDSOI with Globalfoundries is non-exclusive raising the possibility that SMIC or UMC might become an FDSOI foundry. However, if IBM's chip unit changes hands; or is broken up; if there is uncertainty about the IBM patent portfolio; it is not clear that this group of companies have the self-confidence to drive FDSOI to market.
Overall this development is probably good for Intel and TSMC, which make up the FinFET brigade. They can stand by and take the benefit from the repercussions of this seismic shift. For most other chip manufacturers this is disruptive news – unless they can make a bid for IBM chip business and turn it into an opportunity.
Related links and articles: