'Despite Chip Shortage, Chip Innovation Is Booming' (nytimes.com) 33
The New York Times reports on surprising silver linings of the global chip shortage:
Even as a chip shortage is causing trouble for all sorts of industries, the semiconductor field is entering a surprising new era of creativity, from industry giants to innovative start-ups seeing a spike in funding from venture capitalists that traditionally avoided chip makers. Taiwan Semiconductor Manufacturing Company and Samsung Electronics, for example, have managed the increasingly difficult feat of packing more transistors on each slice of silicon. IBM on Thursday announced another leap in miniaturization, a sign of continued U.S. prowess in the technology race. Perhaps most striking, what was a trickle of new chip companies is now approaching a flood.
Equity investors for years viewed semiconductor companies as too costly to set up, but in 2020 plowed more than $12 billion into 407 chip-related companies, according to CB Insights. Though a tiny fraction of all venture capital investments, that was more than double what the industry received in 2019 and eight times the total for 2016. Synopsys, the biggest supplier of software that engineers use to design chip, is tracking more than 200 start-ups designing chips for artificial intelligence, the ultrahot technology powering everything from smart speakers to self-driving cars. Cerebras, a start-up that sells massive artificial-intelligence processors that span an entire silicon wafer, for example, has attracted more than $475 million. Groq, a start-up whose chief executive previously helped design an artificial-intelligence chip for Google, has raised $367 million.
"It's a bloody miracle," said Jim Keller, a veteran chip designer whose resume includes stints at Apple, Tesla and Intel and who now works at the A.I. chip start-up Tenstorrent. "Ten years ago you couldn't do a hardware start-up...."
More companies are concluding that software running on standard Intel-style microprocessors is not the best solution for all problems. For that reason, companies like Cisco Systems and Hewlett Packard Enterprise have long designed specialty chips for products such as networking gear. Giants like Apple, Amazon and Google more recently have gotten into the act. Google's YouTube unit recently disclosed its first internally developed chip to speed video encoding.
And Volkswagen even said last week that it would develop its own processor to manage autonomous driving.
Equity investors for years viewed semiconductor companies as too costly to set up, but in 2020 plowed more than $12 billion into 407 chip-related companies, according to CB Insights. Though a tiny fraction of all venture capital investments, that was more than double what the industry received in 2019 and eight times the total for 2016. Synopsys, the biggest supplier of software that engineers use to design chip, is tracking more than 200 start-ups designing chips for artificial intelligence, the ultrahot technology powering everything from smart speakers to self-driving cars. Cerebras, a start-up that sells massive artificial-intelligence processors that span an entire silicon wafer, for example, has attracted more than $475 million. Groq, a start-up whose chief executive previously helped design an artificial-intelligence chip for Google, has raised $367 million.
"It's a bloody miracle," said Jim Keller, a veteran chip designer whose resume includes stints at Apple, Tesla and Intel and who now works at the A.I. chip start-up Tenstorrent. "Ten years ago you couldn't do a hardware start-up...."
More companies are concluding that software running on standard Intel-style microprocessors is not the best solution for all problems. For that reason, companies like Cisco Systems and Hewlett Packard Enterprise have long designed specialty chips for products such as networking gear. Giants like Apple, Amazon and Google more recently have gotten into the act. Google's YouTube unit recently disclosed its first internally developed chip to speed video encoding.
And Volkswagen even said last week that it would develop its own processor to manage autonomous driving.
Summary. (Score:2)
In other words hard problems require unique solutions.
Re: (Score:2)
It's called creating substitutions [juliansimon.com], when resources get short, which happens at all points on the supply chain, from raw materials to alternative manufacturing to creating brand new workalikes to entire new products that do similar things.
If oil gets short, other energy sources, other ways to create it, to create new plastics, to create products out of other than plastics, entirely new engines not based on combustion.
Here they pus even tinier transistors, new fabs, other custom chips so you don't need off th
Dumb article (Score:3)
Re: (Score:2)
Oh it is. [globalfoundries.com] In the US no less.
Re: (Score:2)
But Jay tells me they'll make more. [youtu.be]
Re: (Score:3)
Software is Easy. Material Science is hard. (Score:5, Interesting)
The most underreported story in tech for the last couple of years is the failure of Intel to get to a new process node. They just can't get the manufacturing to work. The only company that can do it is TSMC, and Samsung, at low yields. This is not a software problem. This is a physics, chemistry and material science problem. We have talented physicists getting billions to look for dark matter on the other side of the universe, or do string theory or whatever, and they should be working at semiconductor companies getting these new process technologies going. I doubt there are a lot of people in Korea or China who are spending their whole careers on things going on at the other end of the galaxy.
We have been in a golden age where software can improve every business and many easy gains have been made here. What's happening now though is that we're running up against the hard problems that don't happen inside a computer and you can't just throw software at it.
Re: (Score:1)
Apparently speculative execution [arstechnica.com] seems to be a hard problem.
Re: (Score:3, Interesting)
> We have been in a golden age where software can improve every business and many easy gains have been made here.
We have also been in an age where software quality/efficiency has been freeriding on hardware improvements. I don't expect this age to stop yet, as I'm sure some very cool lithography/packaging hacks will be implemented. But we clearly are beginning to hit physical limits. At some point us software guys will have to start being clever too. Do a lot with almost nothing.
Heck I have a rasbperry P
Re: Software is Easy. Material Science is hard. (Score:5, Informative)
Much of our physics, without which you cannot even make a transistor, comes from astronomy. Newton, for example, developed calculus and the laws of motion/gravity trying to explain Galileo observations of Jupiter. Einstein developed the general theory of relativity trying to explain the precession of the planet Mercury. Semiconductor device physics utilize the Laplace transform and also Poisson equations .. Who were Laplace and Poisson? Astronomers calculating gravitational interactions. Just about everything fundamental to semiconductor device materials physics comes from astronomy and particle physics. Other high tech fields lean on astronomy too. When Illumina was building their DNA sequencer, they hired expertise from astrophotography because they needed to photograph tiny flashes of light as produced by enzymes acting on DNA. If we halt studies of astrophysics we will hit a wall.
Re: (Score:1)
Intel has failed to focus on the right things in the past decade. TSMC and Samsung stayed focused solely on the technology, i.e. getting the EUV work. Intel was at that time fighting hard to become a more diverse company (under Krzanich). By getting as many exploitable H1Bs into the US as possible, while and at the same time, doing as much outsourcing as possible into East Asian countries where the cost of labour is much lower and so is the quality of work. Infighting within the company, between manufacturi
Re: (Score:1)
Intel also bet the farm on x86 being the primary CPU of the world. Problem is that ARM is eating Intel's lunch, be it embedded ARM CPUs, ARM on the desktop with Apple, or ARM in the server room, as with AWS. It is ironic (and sad) to see Intel farming out work to TSMC, when they should be making their own fabs and at least be competitive on the nanometer levels.
Re: (Score:3)
You have things backwards, if not completely wrong. AWS wouldn't be expanding near-100% with Graviton2 had Intel:
a). kept prices lower and
b). produced something significantly better for the server room in the last four years
Fact is that Intel's still selling Cascade Lake-SP to its customers which is in-and-of-itself little more than a tweaked Skylake-SP. IceLake-SP isn't shipping in high volume (despite what Gelsinger wants you to believe) and Sapphire Rapids is nowhere near market-ready.
It has nothing to
Re: (Score:2)
"doing as much outsourcing as possible into East Asian countries where the cost of labour is much lower and so is the quality of work."
TSMC and Samsung are entire companies based on East Asian labor, yet even you concede that the quality of their work isn't too bad.
Re: (Score:1)
Re: (Score:2)
Intel's R&D and fabbing aren't in Malaysia. They have had packaging facilities there for decades at no detriment to the company.
Re: (Score:2)
We have talented physicists getting billions to look for dark matter on the other side of the universe, or do string theory or whatever, and they should be working at semiconductor companies getting these new process technologies going.
I highly doubt that "billions" are going to pay the salaries of astronomers and string theorists. There really aren't all that many of them, and there are a lot more people who would like that career than there are positions available, so salaries aren't going to be huge.
There's big money being spent on some of their experiments, but most of that ends up going into the down-to-earth hardware that implements the experiments. One side effect of that is the need to develop new cutting edge computers and electr
Re: (Score:1)
Don't forget the good maths people (I understand they are even getting others who are in the hard sciences) to become quants in the finance industry.
Just to get the extra 0.01% profit.
I think there are more people being sucked into finance then into astronomy.
And at least astronomy leads to discoveries and is a science.
Re: (Score:1)
We have talented physicists getting billions to look for dark matter on the other side of the universe, or do string theory or whatever, and they should be working at semiconductor companies getting these new process technologies going. ...
In the same sense you could ask an cancer doctor why he did not made the COVID vaccine
Re: (Score:3)
We have talented physicists getting billions to look for dark matter on the other side of the universe, or do string theory or whatever, and they should be working at semiconductor companies getting these new process technologies going. In the same sense you could ask an cancer doctor why he did not made the COVID vaccine ...
LOL. BioNTech has been working for years on mRNA vaccines to cure cancer. They repurposed their cancer-fighting technology to make a COVID vaccine.
Re: (Score:2)
That is why I made the pun ...
However a company is not a person.
A guy working in cancer is not switching to vaccines over night.
BioNTech has been working for years on mRNA vaccines to cure cancer. ... oops. Probably we never know :P Bottom line the research is how to deliver mRNA based "vaccines" or medicals. The rest is just DNA printing.
And why this is on their web site. In interviews in Germany they said: BioNtech was founded to research Corona Viruses
Re: (Score:2)
That is why I made the pun ...
Ah, cool. Didn't know if you were aware and said it on purpose, or just stumbled into that particular comparison.
A guy working in cancer is not switching to vaccines over night.
Agreed.
Re: (Score:2)
Ha, are you really actually claiming poor old capitalist corporations can't compete because all the big bucks are in theoretical physics?
Why is this a surprise? (Score:2)
Intel is IMHO borked unless they can get their widget making abilities up to snuff, fast. Sending their designs to TSMC puts them in direct competition with ARM, RISC-V, and a handful of other hungry competitors.
Despite? No. Quite the opposite. (Score:1)
If you can't have as many chips, then you want better chips that can do the job of more chips.
No doubt some actually anticipated these shortages, if not their extent and duration.
Two questions (Score:1)
Does anyone know where there's a list of the kinds of chips and shortage (markup) levels? Also, can some kinds of used chips be used, or are they not compatible with newer items?
Re: (Score:1)
As far as reuse, I generally would consider it for say toys, and not automobiles.
it is a demand issue, not a shortage (Score:1)