The industry’s guiding principle – Moore’s Law – is elastic – starting as doubling the number of components on a chip every year, then moving to every 18 months and then every two years.
So I wasn’t totally surprised to be told by an Intel-ite last week that they don’t know what 22nm is a measurement of on Intel’s 22nm process.
And if the industry leader isn’t providing a scientific justification for the descriptor of its processes, then probably no one else is either.
The process might just as well be called Trixiebelle as 22nm.
If the name of a process is now immaterial – a convenient fiction – then recognising that fact should stop the clamour of claims that one firm’s process is better than another’s.
Another surprise last week was IBM saying EUV probably won’t be ready for 10nm. Might be ready for 7nm. The difference between the 40W power of the laser light source which has been achieved, and the 250W required, is huge. The alternative of multiple patterning will jack up die costs.
Eight years ago IBM vp Bernie Meyerson declared the death of scaling. 130nm was, he said, the last great shrink when density went up, cost went down, performance went up and power went down.
As Meyerson predicted, eight years on new nodes don’t bring better performance or lower power.
If new nodes don’t reduce cost – even though they continue to deliver more density – is this the end of shrinking?