Another Step Toward the End of Moore’s Law

Semiconductors is at the core of Trump’s attack on China. I think everybody should know something about the latest development.

https://spectrum.ieee.org/semiconductors/devices/another-step-toward-the-end-of-moores-law

Taiwan Semiconductor Manufacturing Co. (TSMC) and Samsung—announced in April that they’d climbed one more rung on the Moore’s Law ladder. TSMC spoke first, saying its 5-nanometer manufacturing process is now in what’s called “risk production”—the company believes it has finished the process, but initial customers are taking a chance that it will work for their designs. Samsung followed quickly with a similar announcement.

5nm is the most advanced. Only two companies in the world can do it: TSMC and Samsung. But returns are fast diminishing as we progress even closer to the physical limit:

TSMC says its 5-nm process offers a 15 percent speed gain or a 30 percent improvement in power efficiency. Samsung is promising a 10 percent performance improvement or a 20 percent efficiency improvement. Analysts say these figures are in line with expectations. Compared, though, with the sometimes 50 percent improvements of a decade ago, it’s clear that Moore’s Law is not what it used to be.

https://spectrum.ieee.org/image/MzMwNzU1Ng.jpeg

China’s SMIC is 3 generations behind, only know starting to produce at 14nm. TSMC was at that stage around 2013.

1 Like

14nm is not good enough for most semi components?
What is the theoretical limit? 1 nm?

Nobody knows exactly where the limit is. It can’t be smaller than individual atoms for sure. But even before you reach that there are all kinds of weird quantum effects.

14 is enough for many applications. I think 10 is the sweet spot. It’s unlikely anything afterward would be 2x better than 10. Even today DRAM’s are still manufactured between 10 and 20nm with little reasons to go below 10.

1 Like

Moore’s law is actually more a law on economics than technology. To have hopes of recouping the tens of billions in investment a fab has to have the volumes and customers who can benefit from these more advanced nodes. Last year Global Foundry officially dropped out of the race. I suspect intel will also lag further and further behind. Not that it doesn’t have the talents and know how but there may simply be no economic reason to sink in the money.

Most people say China can’t catch up on semis in the next 20 years. I am actually much more optimistic than that. China has so far no reasons to develop its own thing. But now Trump leaves China no choice. China should ban certain areas of semis itself, to use its huge market to create domestic capabilities.

There is an open source architecture called RISC-V developed at Berkeley. I wish China can make more strategic use of Open Source. China can show itself to be a responsible member and westerners and Chinese can work together without ugly politics getting in between. That would also teach that Son-of-a-bitch SoftBank dirtbag a lesson.

1 Like

https://spectrum.ieee.org/view-from-the-valley/semiconductors/processors/china-trade-war-portends-period-of-pain-for-semiconductor-industry

“I think China is going to struggle with memory,” Hutcheson said. “In memory, the costs are in the equipment and the efficiencies of running the fab; only 5 percent of the cost is labor. That is going to be difficult for them, to have to get materials and equipment from around world.”

French disagreed. “If we take a policy of not selling the best stuff to China,” he said, “if they are forced to use their own [technology], they will, even if it’s a little bit worse.”

“Maybe they can be competitive in China where they are protected,” Hutcheson countered, “but they won’t be able to sell outside China.”

That won’t matter, Kim indicated. “If you are the dominant player in China, you are already doing good.”

Exactly. China’s big strength is not that it has the best engineers or the best schools. They don’t. It’s just that its market is the biggest in the world, bigger than US + Europe + Japan altogether. If you increase the price of something to infinity, which is what an export ban does, Econ 101 says they will have extremely strong incentive to work on replacements.

Restrictive immigration policies are also hurting the semiconductor industry, the panelists indicated, particularly in an era in which U.S. students are tending to ignore electrical engineering and other hardware-oriented fields in favor of computer science programs.

“The immigration problem is real,” said Hershenson. “When I did my graduate research, 70 percent of the students in my group were from Iran; for the last couple of years, Iranians can’t even come to the country.”

China needs to open itself more at the time US is shutting its door. Iran has tons of young, smart people.

On IP theft

But what about all that theft of intellectual property, Rodriguez asked the group. Shouldn’t China be punished?

“IP theft is a big emotional issue, and there is legitimacy to the issue,” French said. “But I don’t think China has cornered the market on IP theft. I don’t think they are the best at it or the most prolific.”

“If there were people from 19th century Britain,” he mused, “they would say the same thing about Americans.”

In any case, it’s a short-term problem. When China’s home-grown intellectual property “gets to a significant level—and it will—China will become more about the protection of IP than acquisition,” French said, reminding the audience that Japanese tech companies followed a similar path.

Hutcheson pointed further back in history. “Europeans today are competitive even though we stole all their tech in the 19th century,” he said. “We all love German cars, we buy European products.”

Americans stole IP in the 19th century :scream: Apparently better than Chinese at stealing too :face_with_symbols_over_mouth:

History is written, or rather re-written, by the victors. Every American will tell you they came up with all the tech themselves, from the beginning of time til today.

1 Like