Remember the good old days when blindly following Moore's Law was the blueprint to success in the semiconductor industry? These days, migrating to the next node is much more complicated -- and expensive. Today's chip companies actually have to think.
Soothsayers have been predicting its death since the ink was still drying on Gordon Moore's 1965 paper, aptly titled "The Future of Integrated Electronics." Yet, here it is more than 50 years later and Moore's Law is still the governing principle of the semiconductor industry. Sort of.
The prediction that the number of transistors on a chip will double every 18 months is still taken largely as an article of faith, even though there is some debate about how long the industry can afford to keep up the pace. Virtually everyone agrees that keeping the pace with Moore's Law is getting more difficult. And expensive.
A new paper by Syed Alam and Greg Douglass of Accenture Strategy examines the issues surrounding maintaining compliance with Moore's Law and poses the question: Should we be blindly following it?
"It used to be that when you reduced the die size and moved on to the leading edge node you were getting the benefit of economies of scale. Now, we we get into sub-10nm, the cost is very high," Alam, the lead director of Accenture's semiconductor practice, said in a recent interview with EE Times.
"Our point of view is this: If you are going from 65nm to 45nm, the cost increase was marginally high. But now if you are going from 16nm to 10nm, the cost is significantly high," he added. "So now you have to evaluate your costs even more critically."
This is essentially the thesis of the Accenture paper, titled "Moore... or Less?" The time has long since passed when migrating a design to a more advanced node automatically paid dividends. Chip companies today need to think long and hard about when — or even if — to invest in more advanced chip design and production.
According to the paper, the cost of designing a chip alone has grown by a whopping 1,300 percent from the 65nm node to the 10nm node. The good old days, when following the industry playbook and staying on pace with Moore's Law were the surest path to success, are gone. It's more complicated today, requiring chip companies to actually think.
NEXT PAGE: Knowing When to Scale