Live data from Hacker News

TSMC head says drive to onshore chip supply chain is 'unrealistic'

asia.nikkei.com

121–130 of 221 posts

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#121

Earlier quoted context omitted.

Do people just forget that intel has fabs all over the world with that expertise as well?

Since TSMC reached 7nm and 5nm first (and it will be years before Intel reaches parity with 5nm technology, and TSMC is expected to reach 3nm by that point...), its obvious for us to be focused on what TSMC/Taiwan is doing that we in the USA aren't. TSMC seems to have achieved a sizable technological lead over Intel, that will perpetuate for the next few years or even decade.

It's not clear to me that onshoring would be without utility if it's not on the current node. Surely there are vast improvements to security and defense concerns to be had through onshoring even if it's (conservatively) 3 years behind the cutting edge.

There's also a good possibility that this would be useful in the market as well if it solves some supply chain issues. Car manufacturers won't need 3nm tech for quite some time, if ever. Hobbyist boards like the Pi have been hard to get for months now. Graphics cards that were 2-3 years old were going for huge premiums up until recently as well. All in all it seems as if there would be at least some markets for chip tech that isn't on the bleeding edge.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#122

Earlier quoted context omitted.

The ones where it has not been successful at making their own chips using their own designs? Those fabs?

Well, they are not the cutting edge anymore, but are still a powerhouse.

No, they were a power house. We will know if they still are a powerhouse if a year from now we no longer have these kinds of discussions.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#123

Earlier quoted context omitted.

Since TSMC reached 7nm and 5nm first (and it will be years before Intel reaches parity with 5nm technology, and TSMC is expected to reach 3nm by that point...), its obvious for us to be focused on what TSMC/Taiwan is doing that we in the USA aren't. TSMC seems to have achieved a sizable technological lead over Intel, that will perpetuate for the next few years or even decade.

It's not clear to me that onshoring would be without utility if it's not on the current node. Surely there are vast improvements to security and defense concerns to be had through onshoring even if it's (conservatively) 3 years behind the cutting edge. There's also a good possibility that this would be useful in the market as well if it solves some supply chain issues. Car manufacturers won't need 3nm tech for quite…

USA has plenty of older fabrication labs, and most of our critical technology is probably on 22nm or 40nm nodes (or older).

US Military equipment is pretty famously on old nodes / older technology for reliability and supply chain purposes for example. And its probably entirely domestic (Micron, Intel, GloFo).

The need for higher-end nodes is almost a "luxury" factor. There's a benefit if the latest iPhones were brought onshore and were US-made. This is for economic reasons (less economic disruption if China attacks Taiwan), even if it leads to slightly higher costs.

------

I know that TSMC is pissed off at the "culture clash" in Arizona (one of these "on shore" US fab-labs) however. I can pretty much imagine a "Gung Ho" (1986 movie) situation occurring there. It doesn't look like the Arizona fab-lab will be complete in time, and TSMC is blaming US worker culture.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#124
As Moore's law slows down, fabs will have longer life near the leading edge.

So, for example, 45 nm rolled out in 2007. If in 2005 you were crazy enough to try to bootstrap a 32 nm fab, you might have hoped to have the line completed a year behind the industry if you were lucky. (You have to get it working, get the customers,and get them shipping in volume to get the business healthy.)

So Joe's Tool and Die startup fab rolls out their 32->36 nm process in 2010. And they luck out because the 22->28 happens in 2012 (a 3 yr gap vs the typical 2 yr at that time.)

But 22->28 rolls out and JT&D is playing catch up again.

If the same cycle stretches to 5 years, the technology ramp becomes much more manageable.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#125

Earlier quoted context omitted.

While I do agree that e-waste is a real problem, it’s not like we can just sit with old tech and say “nah, that’s good enough.” We need both more and better fabs, and better waste handling / recycling.

Maybe a certain level of tech IS good enough. A better world for people means education and healthcare, not a 1mm slimmer disposable phone.

A phone from 3 years ago, running software from 4 years ago would be as fast as anyone would reasonably need, but it gets obsoleted via software bloat.

More frameworks, more layers of abstraction and no confining force that forces anyone to write software that isn't a bloated mess.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#126

Earlier quoted context omitted.

Do people just forget that intel has fabs all over the world with that expertise as well?

Since TSMC reached 7nm and 5nm first (and it will be years before Intel reaches parity with 5nm technology, and TSMC is expected to reach 3nm by that point...), its obvious for us to be focused on what TSMC/Taiwan is doing that we in the USA aren't. TSMC seems to have achieved a sizable technological lead over Intel, that will perpetuate for the next few years or even decade.

Bet against the US if you like, but it's a losing bet.

Onshoring will happen. In the US. In China. in the EU. And likely in Russia too.

This has become a national security issue, and at least in the US, we are resolved to make it happen.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#127
> to bring a full supply chain back and try to be fully self-reliant is totally not efficient. ... At the end of the day, that additional capacity could become non-profitable capacity."

It doesn't matter if it isn't profitable, because if cars, trains, and computers go offline because of chip production supply chain issues then a huge portion of the workforce can't participate. If regional war blocks trade from Taiwan we will be hosed within a matter of years.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#128
post #63

Morris Chang (founder of TSMC)'s speech a year ago touched on 3 huge advantages Taiwan uniquely has. Having a whole, well-knit-together country with such intense knowledge & specilaization in making chips is something that would be extremely hard to recreate anywhere else. A community of practitioners is key. https://interconnected.blog/morris-chang-global-semiconducto... The United States had extreme success growing…

Do people just forget that intel has fabs all over the world with that expertise as well?

Morris Chang's point is that TSMC is able to draw upon Taiwan's very best talent, as the country's most attractive employer. Likewise, Samsung is among the most prestigious Korean employers. In contrast, where does Intel rank among US employers?

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#129
post #65
post #22

Earlier quoted context omitted.

greater variety is not a thing that would happen (and incidentally, as an electrical engineer, there's plenty of variety already). The hardware used in semiconductor foundries is all made by the same few companies, which is a major reason everyone gets similar advances at the same time- they're all working with the same tools. The vast majority of variety in product lines comes from who is ordering the chips, not the…

Eventually the tech to create the chips will become more common and people will make their own purpose built chips. Eventually.

That assertion fits worse with this industry than like, any other. The constant flow of freshly obsolete semiconductor machinery literally feeds entire other industries. There is always a huge amount of easily available equipment to make high performance semiconductors because new technology is constantly replacing it with new tooling.

Expertise and facility time are the limiting factors. Low volume purpose built SOCs will never be cheap (outside chiplet-based designs) because the ability to fill out a production line is critical to amortizing the immense startup cost of a design.

It is already possible to have completely custom chips built in lots of <100 on older manufacturing nodes. Runs cost low five figures to low millions of dollars. Some college courses involve having an entire chip made. Anyone can order one on 20-30 year old equipment. There is no demand and it will never get significantly cheaper aside from mass-produced chiplet ICs.

Re: TSMC head says drive to onshore chip supply chain is 'unrealistic'

#130
post #55

Earlier quoted context omitted.

Why isn't it possible to do this industry at less than razor edge?

It takes >$10B and >2 years to build a fab which will be effectively obsolete ~2 years after it opens. If it doesn't open on schedule or is underloaded, the ROI goes negative almost immediately. (I somewhat oversimplified this of course but it's to make the point.)

Obsolete? Automakers etc are huuuungry for old chips.
Post reply on HN