The fact that IBM needed to agree to pay GlobalFoundries $1.5bn in cash to take on its fabs underlines the problems that the world's second-largest foundry will have in squaring the needs of IBM and its wider customer base.
IBM has clung on to chipmaking long after any of the other server makers disposed or shuttered their operations for one simple reason – to ensure that its server-hardware business has access to technology that is tuned to its needs. For years that meant being at the head of process migration. More recently it has been about ensuring access to technologies such as silicon-on-insulator (SOI) and 3DIC – both of which have been key to recent server designs.
In many ways, absorbing IBM Microelectronics in GlobalFoundries makes sense. The two companies have manufacturing facilities that are located very close to each other. Both have long experience in alternative technologies such as SOI and have cooperated on process development for a number of years because of it – some that dated back to original parent AMD.
GlobalFoundries also gains access to IBM's well-regarded RF CMOS and BiCMOS processes, which will help it win deals for the more specialized sensor node and communications IC designs we can expect in the low-energy space.
As well as the cash, IBM has agreed that GlobalFoundries is to be its exclusive supplier of server processors for the 22nm, 14nm, and 10nm generations. IBM will continue to perform its own fundamental semiconductor research, the company said in a statement. IBM said it will spend $3bn on semiconductor research over the next five years. GlobalFoundries will gain access to that but not directly – it will be through joint collaboration projects at the SUNY Polytechnic Institute in Albany. However, the foundry will pick up thousands of IBM patents that the computer maker has agreed to sign over.
The deal provides GlobalFoundries with some continuity but, at the same time, we would expect that the foundry will have to conform to IBM's often quite specialized needs to support the manufacture of what will be comparatively few wafers. $1.5bn will no doubt soften the impact of that but it does indicate the scale of the issues both companies will have in meeting the needs of a small but, to IBM, a critical area of technology.
What happens after 10nm will be interesting to watch. IBM seems to have flexibility to shop around but the server maker will have limited ability to tweak the process in its own direction if that does not meet the needs of foundries to support mobile communications and compute. But, as the trend in servers is towards these low-power targets, maybe IBM is betting that its needs will become increasingly similar to those of everybody else and that its $3bn will not just benefit its own servers.