3 ms·

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 semicon

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 foundries where they are made. Adding more foundries will not introduce any more different chips, just chips that have subtly different performance or reliability, which is already a nightmare engineers deal with continuously. China sells the cheapest chips, and companies only buy the cheapest chips. Making more foundries in the EU and Europe will not change that fact. China will have the huge majority of demand. If anything, a reduction in demand will give more power to counterfeiters and raise the floor price you have to pay to avoid them. It won't really impact prices at all (counterfeit chips are almost nonexistent for anything costing more than a couple cents) but it will make life significantly more irritating for people like me.

kennywinker 5y ago
When I read greater variety I was thinking of a greater variety of functions not a greater variety of fabrication techniques. Like if you’re shopping for a regulator there are dozens of manufacturers with dozens of options. If you’re shopping for something strange… say a 16x16 switch matrix? I think there’s a single chip? Excess capacity because of a proliferation of fabs could lead to a lot of cool chips being made for a low cost. Which in turn enables new technology in that “these are cheap, what can i make with them?” way.
MangoCoffee 5y ago
>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. it depend on the craft smith's skill. everyone have access to Python but some is able to craft a robust app while others app is buggy and unusable. TSMC consistently have the best yields and quality chips
- yellow_postit 5y ago
  Acquired did a podcast deep dive in TSMC history which supports your argument — they’ve arrive at those yields from just a massive set of iterations. I had a lot more respect to the art + science aspect of chip manufacturing the more I learn about fabs, having originally thought it was a basic recipe easily replicated. [1] https://www.acquired.fm/episodes/tsmc https://www.acquired.fm/episodes/tsmc
R0b0t1 5y ago
Eventually the tech to create the chips will become more common and people will make their own purpose built chips. Eventually.
- hwillis 5y ago
  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.
  - R0b0t1 5y ago
    You could say the same thing about something less niche like spectrum analyzers 20-30 years ago. They used to take up equipment cabinets, and you needed purpose built versions for specific analyses.

kennywinker 5y ago

When I read greater variety I was thinking of a greater variety of functions not a greater variety of fabrication techniques. Like if you’re shopping for a regulator there are dozens of manufacturers with dozens of options. If you’re shopping for something strange… say a 16x16 switch matrix? I think there’s a single chip? Excess capacity because of a proliferation of fabs could lead to a lot of cool chips being made for a low cost. Which in turn enables new technology in that “these are cheap, what can i make with them?” way.

MangoCoffee 5y ago

>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. it depend on the craft smith's skill. everyone have access to Python but some is able to craft a robust app while others app is buggy and unusable. TSMC consistently have the best yields and quality chips

yellow_postit 5y ago
Acquired did a podcast deep dive in TSMC history which supports your argument — they’ve arrive at those yields from just a massive set of iterations. I had a lot more respect to the art + science aspect of chip manufacturing the more I learn about fabs, having originally thought it was a basic recipe easily replicated. [1] https://www.acquired.fm/episodes/tsmc https://www.acquired.fm/episodes/tsmc

R0b0t1 5y ago

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

hwillis 5y ago
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.
- R0b0t1 5y ago
  You could say the same thing about something less niche like spectrum analyzers 20-30 years ago. They used to take up equipment cabinets, and you needed purpose built versions for specific analyses.