One answer may be that 5nm was pretty much the end of miniaturization anyway, and TSMC would be stuck there
with a dwindling moat until competitors caught up.
We run into wavelength-of-light issues around 400 nanometers. There's really no practical way to go beyond that. I can provide tons of citations from top engineers stating we're done here. You might have some processes with e-beam lithography, but it's just too expensive and slow for anything outside super-specialized uses.
Fortunately, that's gotten us to the 200MHz Pentium Pro, with 5 million transistors. Seems like that's plenty good enough for what most people would want to do with computers.
I think it's more accurate to say that die shrinkage is over. Moore's Law still has a fighting chance, we just have to wait and see what the clever engineers at AMD, Intel, Qualcom, et al. do about it.
Of course, with 5 million transistors of pure power, I also see no need for further improvements.
TSMC does not build the actual tool that builds the processors, which is the photolithography device. That technology is owned by ASML, the Dutch company, which obeys any commands given to it by their masters, the Americans.
As an analogy, TSMC is like a really good artist, that can use someone else’s tools, but they can’t make the actual tool itself. It’s like a graphic artist that can make very creative drawings using Photoshop, but he’d have no idea how Adobe created Photoshop, since he’s just a user of the product.
And on the flip side, the programmer who created Photoshop, created a very good tool, but he has no artistic skills in his training, to create such magnificent pictures. In this case, it’s that way for ASML. They can create the machines, but they need TSMC to achieve the 5nm production.
What TSMC and others have said, is that the secret to their success, is in the yield. Or how many wafers they can produce, with the minimum amount of defects. This is where TSMC needs a lot of cheaper engineers, in order to make perfect their production process.
This might be the key reason why Intel, and any other western semiconductor manufacturers cannot currently compete against TSMC or Samsung. The cost of a western engineer is likely 5x more expensive, than a Taiwanese TSMC engineer. This is quite possibly a numbers game. You just need a lot of engineers, in order to hit your yield, in order to achieve profitability.
What China needs to do is: (1) create an ASML competitor to build the photolithography device and get their independent intellectual property rights, and (2) employ the massive amounts of semiconductor engineers, like TSMC, in order to achieve the yield.
If they fail to achieve #1, then the United States can easily block their success, by preventing ASML from selling them any photolithography devices.
However, the key difference here, between China and the rest of the world, is that this is now considered an existential crisis. They must now achieve photolithography mastery and dominance, or else, they will die. They must create an ASML competitor.
This means that, China will financially support and backstop whatever domestic Chinese organization, in order to achieve this political goal.
And they are not starting from step 1, so they at least have some understanding of the science behind it. So when they do achieve mastery of the photolithography devices, then TSMC, Samsung, Intel, and ASML, will have some very interesting competition on their hands. At the very least, they will likely lose all future business in China. And they have the American government to thank for that.
> It’s definitely not surprising to see the competition close in on Intel, but it’s also not the end of an era.
TSMC's 7nm process is mature and superior to anything that Intel can currently produce in volume, and their 5nm process has nearly double the transistor density of 7nm (and over 4x the density of Intel's mature 14nm process). Samsung isn't far behind.
That Intel's CEO is telling investors that they may be manufacturing cutting edge products elsewhere is a clear signal that they're getting ready to pivot. This is on top of the earlier steps that Intel took to make their processor architectures more portable between manufacturing processes.
Semiconductor manufacturing, especially at the leading edge, requires massive economies of scale to be sustainable. While Intel didn't explicitly say they were bowing out, the steps that Intel's leadership is taking clearly point to them recognizing that they have no viable path forward as a leading-edge manufacturer.
Within the next 3 years, barring a government bailout, or TSMC/Samsung's 5nm processes failing in a similar fashion (which at least for TSMC, seems unlikely), I expect Intel to spin off its manufacturing business into a separate company and go fabless.
Next they will run to the government with a China scare story to get lots of funding all while continuing to pay lots of executive bonuses. Time honoured tactic of invoking too big to fail (see Boeing)
The 7nm delay shoudnt surprise anyone who's been following their 10nm debacle.
Potentially farming out some products to third part fabs is a little surprising, but seems pragmatic. It could even benefit them to compete with AMD, NVIDIA, and Apple for the limited number of leading edge node wafers coming out of TSMC.
> Swan said on Friday that Intel’s products are still the best, despite the manufacturing delays.
LOL WUT?
AMD’s chips have twice the performance for half the price. The only people still buying Intel are name-brand fanboys or institutional buyers that need platform consistency. Anyone who cares about performance is already looking hard at AMD.
Intel has painted itself into a very dark corner by resting on its laurels. Somehow I’m thinking that it’s going to be the BlackBerry of the semiconductor world.
Swan has a finance background, and isn't a semiconductor expert like Dr Lisa Su. I'm not surprised he knows absolutely nothing about the technology he's selling.
Intel has a massive stake in the software ecosystem that lets them embed certain features that are enhanced or even only possible on their hardware.
For example, there are shops that built their entire security stack in a way that’s totally dependent on Intel and would have an immense cost to re-architect it, much more than just buying 2x as many intel chips because they’re half as fast.
Bob Swan, a non-technical bean-counter, is famous for running a social club rather than a serious Board of Directors at Intel. It is a big threat to national security if Intel goes totally fab-less, or hits some other performance benchmark of severe management failure. Money alone can't fix this, only replacing Bob with a real technical leader will do, preferably sooner rather than later.
23 comments
[ 4.6 ms ] story [ 59.7 ms ] threadAlso, China has just poached over 3k TSMC engineers with govt. funding (>3x salaries!), so we will see if TSMC is disrupted by China’s chosen winner.
One answer may be that 5nm was pretty much the end of miniaturization anyway, and TSMC would be stuck there with a dwindling moat until competitors caught up.
https://www.zdnet.com/article/thousands-of-taiwanese-chip-ex...
There is 3nm in 2022, 2nm GAA in 2024 and 1.4nm in 2026.
We run into wavelength-of-light issues around 400 nanometers. There's really no practical way to go beyond that. I can provide tons of citations from top engineers stating we're done here. You might have some processes with e-beam lithography, but it's just too expensive and slow for anything outside super-specialized uses.
Fortunately, that's gotten us to the 200MHz Pentium Pro, with 5 million transistors. Seems like that's plenty good enough for what most people would want to do with computers.
Of course, with 5 million transistors of pure power, I also see no need for further improvements.
As an analogy, TSMC is like a really good artist, that can use someone else’s tools, but they can’t make the actual tool itself. It’s like a graphic artist that can make very creative drawings using Photoshop, but he’d have no idea how Adobe created Photoshop, since he’s just a user of the product.
And on the flip side, the programmer who created Photoshop, created a very good tool, but he has no artistic skills in his training, to create such magnificent pictures. In this case, it’s that way for ASML. They can create the machines, but they need TSMC to achieve the 5nm production.
What TSMC and others have said, is that the secret to their success, is in the yield. Or how many wafers they can produce, with the minimum amount of defects. This is where TSMC needs a lot of cheaper engineers, in order to make perfect their production process.
This might be the key reason why Intel, and any other western semiconductor manufacturers cannot currently compete against TSMC or Samsung. The cost of a western engineer is likely 5x more expensive, than a Taiwanese TSMC engineer. This is quite possibly a numbers game. You just need a lot of engineers, in order to hit your yield, in order to achieve profitability.
What China needs to do is: (1) create an ASML competitor to build the photolithography device and get their independent intellectual property rights, and (2) employ the massive amounts of semiconductor engineers, like TSMC, in order to achieve the yield.
If they fail to achieve #1, then the United States can easily block their success, by preventing ASML from selling them any photolithography devices.
However, the key difference here, between China and the rest of the world, is that this is now considered an existential crisis. They must now achieve photolithography mastery and dominance, or else, they will die. They must create an ASML competitor.
This means that, China will financially support and backstop whatever domestic Chinese organization, in order to achieve this political goal.
And they are not starting from step 1, so they at least have some understanding of the science behind it. So when they do achieve mastery of the photolithography devices, then TSMC, Samsung, Intel, and ASML, will have some very interesting competition on their hands. At the very least, they will likely lose all future business in China. And they have the American government to thank for that.
It’s definitely not surprising to see the competition close in on Intel, but it’s also not the end of an era.
TSMC's 7nm process is mature and superior to anything that Intel can currently produce in volume, and their 5nm process has nearly double the transistor density of 7nm (and over 4x the density of Intel's mature 14nm process). Samsung isn't far behind.
That Intel's CEO is telling investors that they may be manufacturing cutting edge products elsewhere is a clear signal that they're getting ready to pivot. This is on top of the earlier steps that Intel took to make their processor architectures more portable between manufacturing processes.
Semiconductor manufacturing, especially at the leading edge, requires massive economies of scale to be sustainable. While Intel didn't explicitly say they were bowing out, the steps that Intel's leadership is taking clearly point to them recognizing that they have no viable path forward as a leading-edge manufacturer.
Within the next 3 years, barring a government bailout, or TSMC/Samsung's 5nm processes failing in a similar fashion (which at least for TSMC, seems unlikely), I expect Intel to spin off its manufacturing business into a separate company and go fabless.
Potentially farming out some products to third part fabs is a little surprising, but seems pragmatic. It could even benefit them to compete with AMD, NVIDIA, and Apple for the limited number of leading edge node wafers coming out of TSMC.
The only thing that might help Intel would be the CIA or NSA engaging in some more espionage.
LOL WUT?
AMD’s chips have twice the performance for half the price. The only people still buying Intel are name-brand fanboys or institutional buyers that need platform consistency. Anyone who cares about performance is already looking hard at AMD.
Intel has painted itself into a very dark corner by resting on its laurels. Somehow I’m thinking that it’s going to be the BlackBerry of the semiconductor world.
Searching it looks like amd has 5% server market share and 18% desktop.
(personally I think intel chips are better)
The Intel-beating AMD chips are quite new, so them having less market share is to be expected.
For example, there are shops that built their entire security stack in a way that’s totally dependent on Intel and would have an immense cost to re-architect it, much more than just buying 2x as many intel chips because they’re half as fast.