4 ms·

I think a lot of folks underestimate how fabulously complicated it is to fab a chip. Yes, you can diy it in your garage, here's a wonderful implementation. http

by jfoutz 5y ago

I think a lot of folks underestimate how fabulously complicated it is to fab a chip. Yes, you can diy it in your garage, here's a wonderful implementation. http://sam.zeloof.xyz/first-ic/ http://sam.zeloof.xyz/first-ic/ But take a look at the specialized tools they used. Heck, think about proper handling of https://en.wikipedia.org/wiki/Hydrofluoric_acid https://en.wikipedia.org/wiki/Hydrofluoric_acid and this is just a simple amp. imagine trying to make something with lots of gates? Making a single chip is _hard_. And that's just for a one off. doing that at scale requires an amazing amount of process control. And that diy chip is, according to the author "5µm (1975 tech. level)" Anyway, yes, you could make a chip in your garage, but it'd take a lot of time to ramp up, and you'd only have 1.

amelius 5y ago
If making a chip is so hard, then why do the FAANG companies own this world, not e.g. TSMC or Samsung?
- hans-moleman 5y ago
  I think this year has proven that TSMC does have a very dominant position in the industry.
- pstrateman 5y ago
  I think your comment is being misunderstood because of the ambiguity of "this world".
- jfoutz 5y ago
  I'm not quite sure I follow. I guess the same reason AMF isn't really much of a thing anymore? I don't think making a pinsetter would be a big deal now with ready access to parts and tools, but I'd imagine it was quite a feat in the 40's. I think I need more description around "owning the world". Sure, those leisure focused companies are pocketing big dollars, good for them! But they don't really do much, they're only meaningful given the backdrop of the larger economy.
- baybal2 5y ago
  TSMC, and Samsung are both very glad at FAANG trying to "own the world" — more business for them! I don't see Morris Chang running to open an advertising shop, webhosting company, or mine bitcoins, thought the later was much speculated.
- dcolkitt 5y ago
  One reason I'd say is because there's an at best weak relationship between "hard" and "well-paid". Coca Cola has a larger market cap than Pfizer. mRNA vaccines are definitely a lot harder to make than sugar water. At an individual level sales people are often paid more than the engineers in the org, despite the latter doing much more complex work and having specialized training.
newsclues 5y ago
We are also dealing with a global logistics network that may rival microchips in complexity. Food, and fuel for the humans and planes, trains, trucks, ships and bicycles that are essential to get things places to keep the global economy functioning.
- bob1029 5y ago
  I worked in a semiconductor facility as a systems engineer for a few years. I would have a hard time believing that global logistics management is fundamentally more complex than what happens in just one of the factories responsible for producing a modern HPC chip. The material handling systems are a sight to behold. Most may not be aware, but automation rates of 95% and beyond are feasible in these facilities. Many times, a lot can be moved through the entire manufacturing process without a single human touching or even looking at it. The amount of code and engineering around the material handling system is a mind boggling expanse of complexity. When million dollar product is flying around at 30+mph overhead on robots, you tend to take your time and do it right. Add on top of this the same time domain concerns you have in global logistics, but with far more acute consequences. It's probably ~OK if a cargo ship is 15 minutes late to port. If lots miss a special process timeout by the same duration, you are potentially looking at millions of dollars in scrap. These intervals can be as short as 30 minutes. Imagine trying to schedule high priority lots through special tools, intermixing with unrelated (but also urgent) manufacturing nodes, while also allowing for preventative and break-fix maintenance on these same tools. By far, the most complex parts of these semiconductor manufacturing operations are the business rules for the overall manufacturing environment, followed closely by the IT infrastructure required to tie together tens of thousands of hyper-complex technology systems. You better believe this is the one time it 100% makes sense to use pub/sub messaging. Oh and don't forget about all the HF acid, ultrapure water, EUV lasers, hyper-scale cleanrooms, exotic power distribution systems, et. al.
jiveturkey 5y ago
the author of that post and creator of the chip was a high schooler at the time. (now in college). it’s a lot of detailed work. it was hard for the creators of the process. for us builders, the steps are there and reproducible. maybe you’re using hard in a different way than i am taking it. gate count isn’t really the hard part either, to my best understanding. just like software dev, hobby level and professional level are far apart in terms of what is the hard part. so i think i’m taking offense at your example as being exemplary of the “hard part”
- jfoutz 5y ago
  Ah, no offense intended. There were a lot of calls to just build more fabs. I personally, and I think it's a common belief, that for a lot of software I could build a simple version, alone, over the weekend. When stackoverflow was new, replicating stack overflow was practically a meme. I was trying to highlight, if you want to make a chip, you can, but it'll take a long time to build up the knowledge and skills. So maybe, you hire people with that knowledge and those skills. Of course, a lot of speed comes from not having to coordinate with lots of people. Small teams don't have as many communication channels, solo work has no communication channels. There's probably a group of people and a set of supplies that could get a fab built surprisingly quickly. I don't think automakers are suffering from 5nm chips, they can get away with older processes. Good luck finding all those resources funding, coordinating, and building the process. Perhaps not hard like discovering new process, but a _lot_ of work. It's not going to be a weekend project. This is what I meant by, It's going to be hard.

by jfoutz 5y ago

amelius 5y ago

If making a chip is so hard, then why do the FAANG companies own this world, not e.g. TSMC or Samsung?

hans-moleman 5y ago
I think this year has proven that TSMC does have a very dominant position in the industry.
pstrateman 5y ago
I think your comment is being misunderstood because of the ambiguity of "this world".
jfoutz 5y ago
I'm not quite sure I follow. I guess the same reason AMF isn't really much of a thing anymore? I don't think making a pinsetter would be a big deal now with ready access to parts and tools, but I'd imagine it was quite a feat in the 40's. I think I need more description around "owning the world". Sure, those leisure focused companies are pocketing big dollars, good for them! But they don't really do much, they're only meaningful given the backdrop of the larger economy.
baybal2 5y ago
TSMC, and Samsung are both very glad at FAANG trying to "own the world" — more business for them! I don't see Morris Chang running to open an advertising shop, webhosting company, or mine bitcoins, thought the later was much speculated.
dcolkitt 5y ago
One reason I'd say is because there's an at best weak relationship between "hard" and "well-paid". Coca Cola has a larger market cap than Pfizer. mRNA vaccines are definitely a lot harder to make than sugar water. At an individual level sales people are often paid more than the engineers in the org, despite the latter doing much more complex work and having specialized training.

newsclues 5y ago

We are also dealing with a global logistics network that may rival microchips in complexity. Food, and fuel for the humans and planes, trains, trucks, ships and bicycles that are essential to get things places to keep the global economy functioning.

bob1029 5y ago
I worked in a semiconductor facility as a systems engineer for a few years. I would have a hard time believing that global logistics management is fundamentally more complex than what happens in just one of the factories responsible for producing a modern HPC chip. The material handling systems are a sight to behold. Most may not be aware, but automation rates of 95% and beyond are feasible in these facilities. Many times, a lot can be moved through the entire manufacturing process without a single human touching or even looking at it. The amount of code and engineering around the material handling system is a mind boggling expanse of complexity. When million dollar product is flying around at 30+mph overhead on robots, you tend to take your time and do it right. Add on top of this the same time domain concerns you have in global logistics, but with far more acute consequences. It's probably ~OK if a cargo ship is 15 minutes late to port. If lots miss a special process timeout by the same duration, you are potentially looking at millions of dollars in scrap. These intervals can be as short as 30 minutes. Imagine trying to schedule high priority lots through special tools, intermixing with unrelated (but also urgent) manufacturing nodes, while also allowing for preventative and break-fix maintenance on these same tools. By far, the most complex parts of these semiconductor manufacturing operations are the business rules for the overall manufacturing environment, followed closely by the IT infrastructure required to tie together tens of thousands of hyper-complex technology systems. You better believe this is the one time it 100% makes sense to use pub/sub messaging. Oh and don't forget about all the HF acid, ultrapure water, EUV lasers, hyper-scale cleanrooms, exotic power distribution systems, et. al.

jiveturkey 5y ago

the author of that post and creator of the chip was a high schooler at the time. (now in college). it’s a lot of detailed work. it was hard for the creators of the process. for us builders, the steps are there and reproducible. maybe you’re using hard in a different way than i am taking it. gate count isn’t really the hard part either, to my best understanding. just like software dev, hobby level and professional level are far apart in terms of what is the hard part. so i think i’m taking offense at your example as being exemplary of the “hard part”

jfoutz 5y ago
Ah, no offense intended. There were a lot of calls to just build more fabs. I personally, and I think it's a common belief, that for a lot of software I could build a simple version, alone, over the weekend. When stackoverflow was new, replicating stack overflow was practically a meme. I was trying to highlight, if you want to make a chip, you can, but it'll take a long time to build up the knowledge and skills. So maybe, you hire people with that knowledge and those skills. Of course, a lot of speed comes from not having to coordinate with lots of people. Small teams don't have as many communication channels, solo work has no communication channels. There's probably a group of people and a set of supplies that could get a fab built surprisingly quickly. I don't think automakers are suffering from 5nm chips, they can get away with older processes. Good luck finding all those resources funding, coordinating, and building the process. Perhaps not hard like discovering new process, but a _lot_ of work. It's not going to be a weekend project. This is what I meant by, It's going to be hard.