The Chip Design Game at the End of Moore’s Law

I just came across and interesting video from last year’s Hot Chips conference. Dr. Robert Colwell of DARPA discusses how the processor design industry is likely to change after it becomes too difficult to continue scaling transistors to ever-smaller dimensions. This is likely to occur sometime within the next decade, so companies need to be planning for the transition today.

Today Dr Colwell heads up programs looking into promising future technologies, but his talk draws a lot on his many years at Intel as a CPU architect. He even points out that CPU microarchitecture has been pretty ineffective. Since he started designing processors that ran at 1MHz the performance has improved 3,500 times due to improvements in semiconductor performance. Due to changes in how processors are architected (pipelines, branch prediction, caches etc) maybe just 50X.

The big challenge is that CMOS has been such a wonderful technology and we have had a free ride. Until recently every process generation is faster, more area, lower power and lower cost. It is an exponential that has gone from his boyood 6-transistor radio to multi-billion transistor chips today. In future the improvements will be much less from process generation to generation. If we have to live with architectural improvements primarily then progress will be very slow. Of all the technologies that DARPA is looking at, he has about 50. But he reckons that only 2 or 3 are truly promising.

He sees 7nm as being the end of the road and Moore’s Law is over around 2020 or 2022. Worse, Intel actually makes all its real money moving its processors into the next node and if 5nm is not coming or is too expensive then designing in 7nm is less attractive.

He has on one of his slides something that I’ve said for some time: When Moore’s Law ends it will be economics that stops it, not physics. Follow the money.

This has implications for many industries that are fed on semiconductors. For example, almost all improvements in cars for ages have been things like engine control units, navigation and so on. Basically processors. So improvements in cars will get a lot slower.

In case the video above doesn’t work then it is here.

More articles by Paul McLellan…