Sony May Pull Out of Cell CPU Production for PS3

One of Sony's senior executives, Executive Deputy President Yutaka Nakagawa, is quoted by several US and Asian press sources as saying that, in the interest of finding cost savings, his company is considering exiting the production end of the Cell processor business. Cell processors are produced by an alliance between Sony, IBM, and Toshiba.

Nakagawa was apparently cornered by reporters after he blamed Cell processor production costs for a 5% decline in quarterly profits for Sony's semiconductor division. Cell processors are currently fabricated using 90 nm lithography, and while plans have been in the works to move the chip to 65 nm, IBM is gearing to retool many of its operations for 45 nm, as part of its other partnership with x86 CPU producer AMD. IBM will have one of only two chip foundries capable of producing silicon-on-insulator (SOI) components such as Cell at the 45 nm level (the other is owned by Chartered Semiconductor), and IBM has already signaled its intention to use that foundry for 45 nm production as soon as next year.

Apparently, when asked whether Sony is prepared to shoulder some of the burden of one, if not two, fabrication transitions over the next twelve months, Nakagawa indicated no.

Since the beginning of the PlayStation 3 project, Sony has intended the console to be the showcase for the Cell processor, an innovative design that already utilizes many of the pipelining and network-on-a-chip approaches that AMD and Intel, respectively, are just now investigating.

Semiconductor Fabtech this morning predicts that Nakagawa's comments are a clue that Sony is considering adopting what's called the fabless model, referring to the approach taken by semiconductor design companies such as nVidia, Texas Instruments, and Broadcom. These companies no longer produce any of their own ASIC designs, instead contracting with other producers who may have already developed the right tools for the job.

This certainly saves the designer tens of millions in up-front production costs. But analysts in recent years believe the only good that may come from adoption of the modern fabless model may be in deferring costs until later, when the designer picks up the tab for someone else's work.

The world's largest producer of ASIC processors for firms that outsource production is Taiwan Semiconductor Manufacturing Corporation. But they may not be available for something as intricate and as susceptible to production errors as Cell, especially since it's an SOI design (as are x86 CPUs) and TSMC does not currently have any SOI customers, according to Semiconductor Fabtech.

So if Sony's going to be doing any outsourcing, IBM is the most likely candidate. IBM was already planning on shouldering some of the costs for Cell, and reaping some of the profits as well. If Sony decides to outsource to IBM, that company will be forced to take on more of the up-front burden. But it also has to come up with a business model that allows it to recoup those costs from Sony later. Normally, recoups come with a surplus, though that's probably not what Sony has in mind.

On the other hand, with few other customers for the Cell processor besides Sony (Mercury Computer Systems, a producer of servers and application-specific systems for hospitals, perhaps being the most prominent), IBM could object to this idea. Raising the costs for Cell might raise them in turn for Cell's few faithful clients, some of whom may re-examine their reasons for not going with Opteron or Xeon. In which case, Sony may find itself in even deeper trouble, and one of the most technologically impressive CPU designs in history may be endangered.