Skip to main content

A closer look at IBM’s potential exit from semiconductor manufacturing

After selling its x86 server business to Lenovo, IBM is apparently considering another major shift in strategy and divesting its semiconductor manufacturing foundries as well. If the company does so, it would be the end of an era. IBM hasn’t dominated semiconductor manufacturing in decades, but the technology that’s come out of IBM’s laboratories has been instrumental in the advancement of CMOS design.

IBM built the first computer capable of defeating a human at chess, the first carbon nanotube transistors, copper interconnects, and pioneered the use of silicon-on-insulator (SOI) technology, which was vital to AMD’s resurgence in the early 2000s. The company has also done a great deal of research on cutting-edge semiconductor manufacturing and technology prototyping, including the use of graphene and the ramping up of III-V materials.

IBM doesn’t plan to stop manufacturing its Power-based processors – it simply wants to sell the plants to someone who would operate them, similar to how GlobalFoundries and AMD remain connected via contract despite being independent companies. On the other hand, the less-than-amazing performance of the GF-AMD partnership is proof that this kind of manoeuvre doesn’t always yield the dividends either company might wish for. When AMD spun GF off in 2009, the expectation and hope was for GlobalFoundries to quickly match TSMC’s 28nm ramp-up, capture a significant portion of the high-margin wins at 28nm, and then push forward to cut Intel’s two-year process lead. None of that happened.

IBM has two major foundries – a 300mm production facility (Building 323) in East Fishkill New York and an older production complex in Burlington, Vermont. The East Fishkill facility uses 300mm wafers and is building 22nm processors while the older Burlington facility may be split between 300mm and 200mm production. Reports differ on this last point.

Would a new buyer continue IBM’s legacy?

Much of the cutting-edge work that IBM has historically accomplished was only possible because the company owned its own prototyping and research divisions. A pure-play foundry like TSMC or GlobalFoundries is a business first and foremost – while they’re obviously interested in ramping up next-generation semiconductor technologies, IBM’s R&D department has likely had more leeway than a conventional manufacturing partnership would offer.

There’s also the question of just how different IBM’s needs are compared to the rest of the industry. IBM, Samsung, and GlobalFoundries are all part of the Common Platform alliance, which means it should be easier for one of those two companies to pick up the reins – but it’s also noteworthy that IBM went for a 22nm process node instead of conventional planar silicon at 20nm. It continues to pioneer significant work on SOI and has demonstrated a FinFET-FDSOI design that put a multi-gate transistor on top of a fully depleted SOI substrate which outperformed either standard alone.

GlobalFoundries’ attempts to create a stronger roadmap for AMD products didn’t work well, and we ultimately saw AMD migrate to a similar technology platform as GF’s other customers. It’s not clear what would happen for IBM under similar circumstances, and there are significant issues to be worked out with the state of New York, which cut IBM significant tax breaks ten years ago in return for promises of continued employment and investment in the region.

IBM’s own consideration of the move is a response to the increasing costs of semiconductors and changing market demands. As lithography costs and new foundry nodes continue to consume ever-higher amounts of capital, the return on investment from building the next-generation plants is increasingly low. IBM lost $130 million (£79 million) on its foundry division last year and is expected to lose an equivalent amount this year as the market finds new footing and buying patterns shift.