What Pat said is 100% correct. The actual manufacturing of EUV ICs is incredible difficult, only eclipsed by the difficulty of figuring out HOW to do it. Doing the fabrication here in the US is great, but the much more difficult and key skill is the 'figuring how to do it'. Modern EUV is unbelievably complicated, and there are a continuous supply of new engineering problems that at first glance often don't even appear to be solvable.
In many cases you have an idea and it take 2-3 weeks to determine if it made any difference, and the feedback loop just gets slower after that. The steps in EUV requires some extreme engineering, and it is that core engineering talent you want to develop. The fabs are complicated, but the recipe is the key.
(Note I worked at Intel in the 90s, funny enough in Pat's group)
> "If you don't have R&D in the US, you will not have semiconductor leadership in the US," Gelsinger said, according to a report this week in the Financial Times.
A suboptimal amount of would be engineers works for spyware companies I guess. And the mind share is with top of the pyramid financial engineering fluff not actually doing stuff.
When I worked with machine engineering there was this gloomy prospect of all these "lifers" who would reach retirement age some day and be replaced by an consultants burning out every 6 months.
The US could figure something out decades ago doesn't mean it can now. I think ship building (even military ship building) is another example that the US could do decades ago with a very large capacity but I'm not sure whether it can reboot to such capacity nowadays.
Yes let's keep up foreign relations where we pay for people to arm themselves against us (china). Or we pay for the defense of people who won't pay to defend themselves (EU).
> "we pay for people to arm themselves against us"
Sounds like you want a completely isolationist trade policy. How has that worked out for the US in the past? The only way to never pay another country to arm themselves against us or any other country is to never give them money, and unless we force them at gunpoint to give us what we want that is not going to happen.
> "we pay for the defense of people who won't pay to defend themselves"
The EU doesn't spend 0% on defense. Many of them are NATO members who are increasing their spending and who have demonstrated that they'll follow us into war after we are attacked (Afghanistan as an example). The US gets far more in soft power/geopolitical benefits out of our European allies and NATO members than we put in.
Regular reminder that the only time article 5 of NATO was invoked, it was USA asking others to help them in war they actually wanted.
And when unwanted war became possble in EU came USA ended up allying itself with Russia while trying to extort the victim country for minerals. It did helped initially, but it did not lasted and USA turned into a bully.
All the while threatening annexation of parts of EU and Canada.
balderdash. Tech designed by the US, sold in the US, but manufactured in china led to utter dominance of manufacturing for china. This would work just fine in reverse. Possession is 9/10ths etc
Seems like a common theme here. Push manufacturing to another country - that country becomes adept, dominates, and branches out. Primary country can't compete, complains, and put up tariffs to try to bring manufacturing back that it chose to give up to begin with.
I think Pat should moderate his comments. He failed and in spectacular fashion. Moreover he was extremely influential in the US government, the CHIPs act is really an Intel bailout. Intel should have been allowed to fail, its fabs spun off and many of its unprofitable ancillary products shut down or sold (Networking anyone?).
Despite that he’s not entirely wrong but he’s also not saying anything that’s of practical use.
True semiconductor R&D won’t return to the US without investment in educating people in semiconductor technologies in US graduate schools. Funding for this type of research has mostly gone into more esoteric materials and processes and hence many people who emerge from these programs are a poor fit for modern industrial semiconductor manufacturing tech. Countries like Singapore and Taiwan do a much better job giving their graduate students a more practical education that better prepares them for industry. In a sense US is too academic and not vocational enough (I see this across the board in HW).
The fact is that advanced nodes being produced in the US is a gigantic step in the right direction.
Decades of commodity HW and SW dominance have gutted the talent base for HW tech development in the US. It’s going to take a decade and smart US government policy (not the Chips act) at least to recover to where we can even begin to have this kind of R&D again in the US.
> Intel should have been allowed to fail, its fabs spun off and many of its unprofitable ancillary products shut down or sold
If Intel had failed, the fabs would have shut down, not been spun off. Intel's fabs are the primary reason Intel has poor financials - continually running on a loss. Intel's Product division makes a profit - always has - and can survive.
The Bailout is all about propping the fabs, not the products. Letting Intel fail really does equate to losing semiconductor capability in the US. There's no other US company that comes even close to Intel's fabs.
The market is harsh to Intel - it's essentially saying "There's no room for number 2"
Yeah, no. Firstly nobody can use Intel’s fabs because they can’t really provide a PDK the way TSMC can. So Intel’s fabs are fundamentally not competitive as it stands because they are almost impossible for anyone but Intel to use. Who cares what their equivalent process is if it can’t be commercialized?
A break up would force them to get their act together as a commercial fab.
The fabs are substantial capital and would have been spun out in the same way IBM spun out their fabs into Global Foundry. Someone with knowledge of how can turn these things profitable.
The bailout props up a failing mismanaged entity and encourages its moribund status quo.
Moreover the US taxpayer shouldn’t foot the bill for Intel’s mismanagement.
The era of x86 dominating compute is clearly over and hence your last point is anachronistic.
I don't know his intention or whether The Register just framed his words to align with a political viewpoint. Perhaps it's nothing more than one industry person taking shots at their rival... but I'm having a hard time reconciling his current statement with his previous statements, specifically about the CHIPS Act and US manufacturing.
"Gelsinger added on the CHIPS Act: "Supply chains move because they're advantaged. So in that sense, I see chips [making], plus some of the economic inducements, as being a formula that I fully agree with looking forward. But we can't go backward on what we put in place for the CHIPS Act. We need to continue to restore manufacturing. We need long-term research and development." (https://www.msn.com/en-us/technology/tech-companies/former-i...)
TSMC was heavily sponsored by the Taiwanese government with the goal of making Taiwan so valuable that other countries would feel required to defend them against China.
It is against their interests to outsource too much or outsource cutting-edge nodes. This is exactly what we see.
The recent Arizona fab is N4 which is just N5 with some refinements. N5 was sampling iPhone chips in 2019 and N4P has been available since 2020-2021. That's a half-decade ago.
N3 started up in 2022, but won't be in the US until 2028 some 6 years later. By 2028, TSMC plans their Taiwanese factories to have produced 2nm, 1.6nm, 1.4nm, and 1nm.
If the US wants cutting edge chip manufacturing, it 100% is NOT coming from TSMC.
There's also a strategic imperative for chip production in the US that isn't cutting edge; lots of stuff depends on basic ICs that were in short supply during COVID and having more production of stuff that doesn't need to be 2nm is a good thing for the US.
There's generally a strong preference for those basic chips to use planar nodes. This limits them to 28nm for TSMC and 22nm for GlobalFoundries (22FDX+ fab is actually in Dresden, but they announced plans to add a second 22FDX+ fab in New York).
I do agree that more fabs in the US is better than fewer though.
Readit News
In many cases you have an idea and it take 2-3 weeks to determine if it made any difference, and the feedback loop just gets slower after that. The steps in EUV requires some extreme engineering, and it is that core engineering talent you want to develop. The fabs are complicated, but the recipe is the key.
(Note I worked at Intel in the 90s, funny enough in Pat's group)
> "If you don't have R&D in the US, you will not have semiconductor leadership in the US," Gelsinger said, according to a report this week in the Financial Times.
When I worked with machine engineering there was this gloomy prospect of all these "lifers" who would reach retirement age some day and be replaced by an consultants burning out every 6 months.
Sounds like you want a completely isolationist trade policy. How has that worked out for the US in the past? The only way to never pay another country to arm themselves against us or any other country is to never give them money, and unless we force them at gunpoint to give us what we want that is not going to happen.
> "we pay for the defense of people who won't pay to defend themselves"
The EU doesn't spend 0% on defense. Many of them are NATO members who are increasing their spending and who have demonstrated that they'll follow us into war after we are attacked (Afghanistan as an example). The US gets far more in soft power/geopolitical benefits out of our European allies and NATO members than we put in.
Learn history :)
And when unwanted war became possble in EU came USA ended up allying itself with Russia while trying to extort the victim country for minerals. It did helped initially, but it did not lasted and USA turned into a bully.
All the while threatening annexation of parts of EU and Canada.
Yes, we may not be able to out-compete TSMC in 2025, but I feel good about our 2045 and 2065 competitiveness.
Despite that he’s not entirely wrong but he’s also not saying anything that’s of practical use.
True semiconductor R&D won’t return to the US without investment in educating people in semiconductor technologies in US graduate schools. Funding for this type of research has mostly gone into more esoteric materials and processes and hence many people who emerge from these programs are a poor fit for modern industrial semiconductor manufacturing tech. Countries like Singapore and Taiwan do a much better job giving their graduate students a more practical education that better prepares them for industry. In a sense US is too academic and not vocational enough (I see this across the board in HW).
The fact is that advanced nodes being produced in the US is a gigantic step in the right direction.
Decades of commodity HW and SW dominance have gutted the talent base for HW tech development in the US. It’s going to take a decade and smart US government policy (not the Chips act) at least to recover to where we can even begin to have this kind of R&D again in the US.
If Intel had failed, the fabs would have shut down, not been spun off. Intel's fabs are the primary reason Intel has poor financials - continually running on a loss. Intel's Product division makes a profit - always has - and can survive.
The Bailout is all about propping the fabs, not the products. Letting Intel fail really does equate to losing semiconductor capability in the US. There's no other US company that comes even close to Intel's fabs.
The market is harsh to Intel - it's essentially saying "There's no room for number 2"
A break up would force them to get their act together as a commercial fab.
The fabs are substantial capital and would have been spun out in the same way IBM spun out their fabs into Global Foundry. Someone with knowledge of how can turn these things profitable.
The bailout props up a failing mismanaged entity and encourages its moribund status quo.
Moreover the US taxpayer shouldn’t foot the bill for Intel’s mismanagement.
The era of x86 dominating compute is clearly over and hence your last point is anachronistic.
"Gelsinger added on the CHIPS Act: "Supply chains move because they're advantaged. So in that sense, I see chips [making], plus some of the economic inducements, as being a formula that I fully agree with looking forward. But we can't go backward on what we put in place for the CHIPS Act. We need to continue to restore manufacturing. We need long-term research and development." (https://www.msn.com/en-us/technology/tech-companies/former-i...)
It is against their interests to outsource too much or outsource cutting-edge nodes. This is exactly what we see.
The recent Arizona fab is N4 which is just N5 with some refinements. N5 was sampling iPhone chips in 2019 and N4P has been available since 2020-2021. That's a half-decade ago.
N3 started up in 2022, but won't be in the US until 2028 some 6 years later. By 2028, TSMC plans their Taiwanese factories to have produced 2nm, 1.6nm, 1.4nm, and 1nm.
If the US wants cutting edge chip manufacturing, it 100% is NOT coming from TSMC.
I do agree that more fabs in the US is better than fewer though.