Computex is a major trade show where companies of all sorts show off their upcoming products and discuss manufacturing trends. It's a great place to get a feel for what's going to be hot in the second half of the year, but it also lends itself to a type of exuberant prediction. In this case, Simon Segars, head of ARM's processor division, told reporters that we could see 20nm devices shipping as soon as the end of 2013.
"The whole industry is focused on moving to the next generation as soon as it's economically viable and technologically achievable," Segars told the group. According to CIO, analyst Dan Nystedt, of TriOrient Investments, chimed in with his view that TSMC's 28nm troubles don't indicate that the 20nm ramp will be difficult. Nystedt characterizes the 28nm difficulties as a "capacity shortage" rather than technology issue.
Attempting to split the two is rather disingenuous at best. One of the primary goals of a foundry -- any foundry -- is to accurately forecast and ramp production of a new node. This requires a lot of careful planning and coordination with one's customers. After its 40nm problems and with GlobalFoundries to contend with at 28nm, TSMC has tremendous incentive to get it right. When a number of your top-tier customers are all citing limited availability as the reason they aren't shipping 28nm parts, capacity is not the fundamental issue.
The real problem is that moving to lower process nodes is hard and only getting harder. There's no easy fix and no "turning the corner." ARM may be feeling pressured by Intel's Medfield debut and the company's public roadmaps that show a 22nm Medfield successor
dropping in 2013, codenamed Silvermont. To be bluntly honest, ARM has some reason for concern -- Intel's track record, when it comes to delivering new nodes on schedule and in quantity, is second to none.
Rather than betting on TSMC's
rather sketchy history when it comes to delivering new nodes in volume, ARM would be better served by emphasizing the upcoming 64-bit ARMv8 architecture, the improved performance of Cortex-A15, and the Windows on ARM tablets and systems that'll ship later this year. The company's Mali GPU architecture and its alliance with HP in the server market are other positive talking points that don't involve discussions of going head-to-head with Intel in pushing new technology nodes.
The chances of 20nm volume that early are miniscule, but that's only a problem if ARM makes it one. From Raspberry Pi to WoA, there's a lot of interesting, ground-breaking activity in the ARM ecosystem, and plenty of projects to discuss.