Hacker News Reader

Comment by gautamcgoel | Hacker News Reader

I've asked this question before but didn't get a good answer. Suppose I had $100M to blow. Could I build a fab in the USA that produced chips at, say, 65nm? I understand that a cutting-edge node like 7nm would be out of reach at that price point, but I'm curious what exactly is in reach.

View on HN

Well I reckon for 100M you could fabricate a chip at nearly any process size you want but probably not at a speed at which you could sell them for a profit. For example many universities have the capabilities to fabricate their own chips for research and development.

It also depends on what type of thing you're trying to fabricate. For example in a phone: cpu, ram, power, radio and sensors might want their own unique fabrication. It gets even more crazy when you want to go into things like automotive or industrial power.

I'll over generalize things and say that you can probably make a fab that goes down to 90nm without too much issue. It was the 65nm node when the lithography started to need something different and new chemistries were required for the fabrication process due to leakage becoming and issue that could no longer be ignored.

At least for me it was the jump to 65nm when I remember things getting hard and staying hard. The cell deigns had to change, the layout took twice as long, the models for simulations were always off and tapeing out to different fabs has vastly different results. It seemed like very node after that the documents that arrived from the fab about the new process were at least twice as thick as the last.

SkyWater bought a primarily 200nm, but 65nm capable fab in Minnesota for 30M in 2017.

https://www.eenewsanalog.com/news/ceo-interview-sonderman-sk...

Granted, it's tooled for 200mm wafers (300mm are the standard right now), but it's still significantly cheaper than expected.

Prices are likely to ramp up really quickly when you get to the 65->45nm jump - that's when Intel started using multiple patterning.

And that was an amazing deal given the glut at 200mm being apparent even back then.

The reason you aren't getting an answer is because you're asking about something that isn't mass produced. So every single build is going to be a contracted cost with many variables. Saying "I want a 65nm fab" doesn't mean anything. What type of EUV scanners do you want, how large of wafers do you want to handle, how many wafers at once, size of the clean room, etc. There is no answer, it's whatever you could get someone to do the job for + materials cost. We only have estimates for cutting edge facilities because we know what public companies have recently spent to produce those fabs.

It's like asking "if I had $100m, how large of a secret underground meth production facility could I have built"....who knows, it's bespoke.

Now I'm just wondering what would $3.5B meth lab look like? it will be breaking all records for sure

Umm..., a pharmaceutical company?

There is a list of fabs, including plant cost, here:

https://en.wikipedia.org/wiki/List_of_semiconductor_fabricat...

The few I can see within your range seem to be very specialized (e.g. Murata) or educational (SUNY). Maybe have a closer look at

https://en.wikipedia.org/wiki/SUNY_Poly_CNSE

Some light research shows that it would cost closer to 3-4 billion dollars to build a fab from 2014[1] This makes a good bit of sense since the clean rooms required are massive and require a good bit of work to get right. Also from my experience with auditing PCB manufacturing in the united states. 100M might be able to get you a old TI Fab to make some TTL Parts at.

[1] https://en.wikipedia.org/wiki/Semiconductor_fabrication_plan...

Do you happen to have any resources (checklists?) On what auditing PCB manu. entails ?

Reply on news.ycombinator.com