Get email delivery of the Cadence blog featured here
"The number of people predicting the end of Moore's Law doubles every two years." It's a good joke that I had forgotten until Friday when I read the Economist, the magazine that insists on calling itself a newspaper. I have been a subscriber since before I came to the US, so that is a long time ago.This week's cover story, the subject of the first leading article (editorial) and the subject of the entire technology quarterly is The Future of Computng. There are some minor errors, such as 14nm transistors not being 14nm, but mostly it is well written. For something as mainstream as the Economist, it is pretty technical.
The technology quarterly opens:
After 50 glorious years, Moore's Law, which states that computer power doubles every two years at the same cost, is running out of steam. Tim Cross asks what might replace it.
The focus is on Intel since the articles are as much about the stalling out of increases in computer power as they are about semiconductor manufacturing. But all the usual suspects, such as Handel Jones of IBS, Chris Mack, lithoguy of nowhere-in-particular, and Linley Gewenapp of The Linley Group, chime in.
One person mentioned several times was Greg Yeric of ARM. He gave a masterful keynote in December at IEDM on just the same topic. See Moore's Law at 50: Are We Planning for Retirement? Greg's keynote was much more technical, of course. After all, it was IEDM and he was on a mission to educate everyone on how things look from the system end of things. One point he emphasized was that designers care a lot about variation and not the typical performance of devices, but device engineers seem to care about the center-point and make variation an afterthought.
There are even sections on some of the weirder stuff out there, like Quantum Computing (subject of another keynote at IEDM, as it happens) and the pipeline of research devices. But as a couple of people pointed out at IEDM when asked what 5nm would look like, it takes 10 years for new ideas like FinFETs to go from lab to HVM. So 5nm will look pretty much like 14nm with FinFET-like devices, probably with additional III-V materials and copper interconnect, maybe with some additional weird metals to improve resistance. In fact, the interconnect may be a bigger issue than the devices.
It is clear that the computer industry is going to become more like the airplane industry. It's not that no progress will be made. We can certainly design weird planes for the military or other specialized purposes. But in the spot where things are economical for the main market, the basic performance of a Boeing 777 or a 380 Airbus is not much different from the 40-year-old 747. The electronics are improved, the fuel efficiency is improved, but there isn't really an opportunity to improve price/performance by 1000X over the next 20 years as we did over the last, ensuring that we all now walk around with a supercomputer in our pockets known as smartphones. There is nothing currently on the horizon promising exponential growth for another 50 years. We have never seen anything like semiconductors before.
If you are a subscriber to The Economist, no doubt you have already at least seen the cover. I think you need to be a subscriber to read it online (or on your smartphone). Click on the cover photo at the start of this piece. But I recommend buying a copy at the newsstand. If you work in this industry, as I assume you do if you're reading this, it is both a good summary and also is what educated people who don't even know what a transistor is will know about the topic. Your friends who work in banking might suddenly ask you what a FinFET is.