“Moore’s Law has become invalid for a while, as the time needed for the transistor density to double is no longer 18-24 months although the density has indeed kept increasing, Morris Chang, chairman of Taiwan Semiconductor Manufacturing Company (TSMC), has said in a recent tech forum hosted by the firm to mark its 30th anniversary,” Josephine Lien and Willis Ke report for Digitimes.
“Chang said that the timeframe set in Moore’s Law is no longer applicable. He said TSMC has kept increasing transistor density, but not at a pace according to the law,” Lien and Ke report. “Based on his firm’s internal planning, Chang said, TSMC will continue to increase transistor densities in the coming eight years, but will see a major challenge in 2025.”
Read more in the full article here.
MacDailyNews Take: The timeframe has slipped to 2.5 years of late.
Some things will change. We won’t have the rate of progress that we’ve had over the last few decades. I think that’s inevitable with any technology; it eventually saturates out. I guess I see Moore’s Law dying here in the next decade or so, but that’s not surprising. — Gordon Moore, March 30, 2015
[Thanks to MacDailyNews Readers “Fred Mertz” and “Dan K.” for the heads up.]
The end of Moore’s Law has long been anticipated. No growth process (increasing or decreasing magnitude) can continue linearly and without bound. We are approaching the atomic scale. Different materials and dielectrics may extend the circuit ministrations process for a bit longer. However, unless we find a way to build circuits at subatomic scales, future improvements in computing will increasingly take the form of increased efficiencies and increased parallelism.
Increased efficiencies includes both logic architecture and software coding. A couple of interesting logic architectures which are gradually finding real worldapplications include neural networks and quantum computing.
Decreasing transistor costs and power requirements could continue the trend of more powerful devices every year even if transistor sizes remained fixed.
From the customers view, Moore’s Law would seem to continue unabated.
Quantum computing in a couple decades could vastly increase the computing power of computing elements, also without any size decreases. Like GPUs, quantum circuits won’t speed up everything a regular CPU does but would still allow astonishingly new applications possible.
Has it been continuing at Intel un recent years? Little outward sign of it.
Chipmakers have seen this coming 15+ years ago. That we have multiple cores. Apps that know how to take advantage of them can still see significant speed-up, even if the chip speed remains constant.
I’ve been saying this for years.
It began with 65nm, which was 3 months late.
45nm was then an additional 6 months late.
32nm was again an addition to that being over 3 months late.
22nm was also an additional 6 months late.
14nm, depending to whom you speak, was 12 to 18 months late,
And 10nm was late as well.
But, beginning with the 28-32nm process, we began to see transistor density fall from the theoretical levels, with only Intel coming close to actual process numbers there. It’s just gotten worse, so that a process size no longer means what we think it means. Intel has said that from now on, ratings should be for transistors per square mm, but naturally, other chip manufacturers have balked.
I don’t know how many people here are old enough, or were interested at the time, but it was back with the 90nm process that we experienced the first major problems with heat from quantum effects (mostly leakage) being much worse than expected. They were able to find ways around it, but the problems don’t actually go away, and are back in a way that won’t be workable before too long. Once that happens, it’s done. There is disagreement at what node that will be, with some experts still thinking that 5nm will not be successful, or that if it is, it will just be a minor improvement, and that 3nm, which TSMC is saying they’re planning on, as impossible.
This is just a few years, folks. After that, it’s just optimizations. If we have a new technology to replace it, it’s not anywhere near ready. Carbon nanotubes are being touted as being it, but so far, just small hand ensembled chips have been seen from IBM and Hp. But we can’t even make carbon nanotubes to order, much less make mass produced circuits yet. And we don’t know when, or even if, that’s going to happen.
Do you have a blog at least? Or have you been “saying this for years” to yourself in a mirror?
I’m on a number of web sites and forums. It’s really easy to find me. I’m also on Discus, where, for some really odd reason, I seem to have accumulated a fair number of followers.
Ok, because you haven’t linked them to your handle which is the most obvious place to do so on this forum.
So then…not a law. More like Moore’s Maxim or Moore’s Axiom which no longer holds true.
No, it never was a “law” as in the “laws of physics.” But that was a popularized term, as it makes it sound more significant. It’s really “Moore’s guide to estimated transistor shrinkage and density for chip manufacturing over the next few years”. That’s my term.
Well let’s just give up and go home, gents, nothing more to do here. It’s all over. Technology has hit a dead end.
It’s actually been, IMHO, 18-24 months since Moore’s Law was valid. Intel’s own processor release timeline has been consistently delayed over at least that period of time.
Meanwhile, quantum computing remains a concept with meagre demonstrable results.
At least RISC processing has come into its own during the same period of time. We’ll see if ARM et al. can out-design Intel and restore Moore’s Law.