The end of the learning curve is nigh, says McGregor
28nm is the first process shrink which doesn’t deliver a cheaper chip, says Scott McGregor, CEO of Broadcom.
According to McGregor, not only is 28nm more expensive than 40nm now, as might be expected at the start of a node, but that, on Broadcom’s current projections, 28nm will never be cheaper than 40nm across the whole lifetime of the node.
“What we’re seeing in the industry is the cost of next-generation nodes rising exponentially,” says McGregor, “what this means is, unless you need the advanced process because of performance reasons or die-size reasons, you’re not going to get a cost benefit from converting to the new node.”
McGregor went on to say that the same would be true at 20nm.
The consequences will be dramatic: companies will stick longer with a node, foundries won’t be so aggressive in their process development and the old learning curve of chips getting cheaper by 30% a year will decelerate.
Apple, reportedly, has stuck with 40nm and 65nm technology rather than risk its products on the availability of 28nm parts.
Broadcom’s strategy on 28nm was to wait on the HPM process which, says McGregor, “tends to have a better figure of merit than the HP or the Poly/SiON processes that some of our competitors are using. So when we do come out with 28nm products, they should have better specs, better power performance and whatnot.”
So, McGregor points out, Broadcom has not suffered initial ramp issues of its competitors and its business has not been limited by access to wafers.
However TSMC is currently pulling out all the stops to get 20nm into production as soon as possible to give its mobile customers a competitive process to Intel’s 22nm finfet process. It was reported last week that TSMC is putting another $700m into bringing forward 20nm deployment.