New backend house company opening, TSMC fab 18 and upgrades at few minor fabbers.
If you take fab 18 and ardentec away, you will see that Taiwan did not reverse the trend, and semi spend keeps going down industry wide for 4th year straight.
Not to mention that semiconductor manufacturing (among other advanced mfg verticals) is such an integral part of Taiwan’s economic strategy at this point that it would be shocking if they weren’t making heavy investments, global trends be damned.
Newer nodes are hellishly expensive to develop, ballooning in cost over previous nodes. Global Foundries and most others are not working on more advanced nodes than 14nm, leaving just TSMC & Samsung with 7nm fabs. Intel has 10nm kinda crappily working, turning out low end chips that are mostly dead silicon (due to process issues).
We're definitely hitting a silicon bottleneck, the physical barriers are very real, and exceedingly costly to attempt to work around. AMD's strategy of using 14nm chiplets for components that don't scale down well (eg: analog interfaces, DRAM, etc) is likely to become much more common.
Global Foundries has a 12nm for most of the new AMD CPUs.[1] I’ve heard the difference between processes at different companies is just marketing talk but I’ll let someone knowledgeable chime in.
Because all this "national security" is really about trade not national security. Trump declared national security on steel, cars and pretty much any important industry that US didn't lead.
Incorrect. While the Trump administration is more enthusiastic about aggressive intervention in commerce in general, the recent sanctions are the culmination of over a decade of bad behavior by Huawei, recognized by many countries:
I think the assertion is that US has lost core competency at things like steel and cars, so by making it harder to bring in mini-vans from Canada or whatever, they're proposing a magic rebirth of those industries in country. Those are the same competencies we'd need if we wanted to build tanks or battleships or bombers for another assault on Normandy. Which we don't, that was important 80 years ago but not now.
The reality of course is probably more like pandering to steel and auto workers.
Communist China intends to conquer Taiwan. If that happens, all Taiwanese companies will be under the de facto control of the Chinese Communist Party. Selling to Huawei most likely aids Communist China in acheiving it's long-term goal of conquering Taiwan.
So if the owners of a Taiwanese company prefer to control the company rather than cede de facto control to the Chinese Communist Party, this seems entirely reasonable.
If I was the Taiwanese leadership, I'd make it very clear we'd prepositioned enough explosives that in a matter of hours all the top fabs and the like could be turned into junk, denying the PRC these crown jewels.
Just speaking for myself, I've got plans for the inevitable disruptions such a war would create. We'd be throwing away a lot less electronic gear for quite some time.
> Mainland China is by far Taiwan's biggest trading partner.
That's true for most neighboring Asian countries. But I'm also guessing that most fabless customers are American. Sure, there are Chinese fabless chipmakers such as HiSilicon (Huawei), Tsinghua Unigroup, but nowhere as big as Apple, Qualcomm, or AMD.
Where else would Apple, Qualcomm, and AMD (don't forget Broadcom too) go for their cutting edge silicon? Samsung couldn't handle that demand. I think they can pretty much do what they want when it comes to maintaining their business with China.
I thought the majority of their production capacity is for memory and displays, which isn't just lying around ready to go for 7nm chip customers. And having plenty of unused capacity lying around seems unlikely to me considering the costs involved. They only got into fabbing other people's stuff seriously a couple of years ago, I know they have business from Global Foundries but I can't think of any others. I don't think there is anything "of course" about suddenly taking on that level of new business. Ramping up however many new plants is a huge deal, these chips are in full production right now.
Capacity aside, swapping vendor is a monumental task. They may not even have the packaging tech that's needed.
Either way, I guess anyone who actually knows can't say.
Samsung has been running custom foundry business since 2005 when it started designing and fabb'ng Apple's mobile AP used in the iPod and iPhones until around ~2013 when Apple finally developed their own custom AP after PA-semi and Intrinsity acquisitions. QCOM and AMD have been with Samsung on and off for years now.
I specifically didn't talk about when they started their foundry business. I specially was talking about 2017 when they started trying to become an serious option for fabless companies.
Yes, and it is still a monumental task, nobody is saying any different now. Apple must have been in a tough spot with their yields but that's what you get when you are first out the door. It's incredible how quickly they start shipping the new tech.
> I specifically didn't talk about when they started their foundry business. I specially was talking about 2017 when they started trying to become an serious option for fabless companies.
Sure, if you believe Samsung LSI got serious about foundry service only since 2017, you know nothing about Samsung's foundry business at all, which is why the brief introduction was necessary.
35 comments
[ 4.1 ms ] story [ 98.2 ms ] threadIf you take fab 18 and ardentec away, you will see that Taiwan did not reverse the trend, and semi spend keeps going down industry wide for 4th year straight.
We're definitely hitting a silicon bottleneck, the physical barriers are very real, and exceedingly costly to attempt to work around. AMD's strategy of using 14nm chiplets for components that don't scale down well (eg: analog interfaces, DRAM, etc) is likely to become much more common.
[1]https://en.wikichip.org/wiki/amd/microarchitectures/zen%2B
It is obviously in their best financial interest to sell to Huawei. Why should they stop?
https://en.wikipedia.org/wiki/Criticism_of_Huawei
As far as technology is concerned US is using its corporations(Apple, Cisco, microsoft etc) for espionage as well so there is nothing new here.
The reality of course is probably more like pandering to steel and auto workers.
I see lots of accusations, but very little evidence
So if the owners of a Taiwanese company prefer to control the company rather than cede de facto control to the Chinese Communist Party, this seems entirely reasonable.
Just speaking for myself, I've got plans for the inevitable disruptions such a war would create. We'd be throwing away a lot less electronic gear for quite some time.
That's true for most neighboring Asian countries. But I'm also guessing that most fabless customers are American. Sure, there are Chinese fabless chipmakers such as HiSilicon (Huawei), Tsinghua Unigroup, but nowhere as big as Apple, Qualcomm, or AMD.
Capacity aside, swapping vendor is a monumental task. They may not even have the packaging tech that's needed.
Either way, I guess anyone who actually knows can't say.
https://semiwiki.com/semiconductor/259664-samsung-foundry-up...
Global Foundries a licensee of Samsung's 14nm and, along with IBM, they were members of Common Alliance.
> swapping vendor is a monumental task. ...
Sure, that's what everyone said until Apple went dual-sourcing with TSMC 16nm and Samsung's 14nm.
Yes, and it is still a monumental task, nobody is saying any different now. Apple must have been in a tough spot with their yields but that's what you get when you are first out the door. It's incredible how quickly they start shipping the new tech.
Sure, if you believe Samsung LSI got serious about foundry service only since 2017, you know nothing about Samsung's foundry business at all, which is why the brief introduction was necessary.
Although, after a quick check, my own access route seems to touch on Hong Kong (EU -> US -> Pacific).