[adrian_b]

You could make in a garage some decent analog integrated circuits, e.g. audio amplifiers or operational amplifiers or even radio-frequency circuits for not too high frequency ranges.

However you cannot make useful digital circuits. For digital circuits, the best that you can do is to be content to only design them and buy an FPGA for implementing them, instead of attempting to manufacture a custom IC.

With the kind of digital circuits that you could make in a garage, the most complex thing that you could do would be something like a very big table or wall digital clock, made not with a single IC like today, but with a few dozen ICs.

Anything more complex than that would need far too many ICs.

[tliltocatl]

Not true. You are confusing "digital" with "microprocessor". You wouldn't be able to do any single-chip microprocessor, of course, but something like 74181 is very doable at this scale, and building a 1970s-era computer out of a few dozen of these is something enthusiasts still do. The main problem isn't logic, it's memory - memory needs density (thin film magnetics anyone?).

Then, of course, if by "useful" you mean "commercially viable", it is indeed not going to be competitive against either TSMC or your local 500nm foundry ever.

[adrian_b]

A CPU made with ALUs like 74181 would take alone a PCB of ATX or eATX size densely populated with integrated circuits and consuming much more power than an entire computer consumes today, while being slower than a tiny microcontroller with a cost of less than a dollar, which also includes enough memory for a practical application.

I call such a CPU as not useful.

It can be a very useful experience to design such a CPU, but you can simulate the design in a logic simulator and you gain nothing by building it.

As a valuable computer building experience, it is more useful to use much older components than digital integrated circuits, where you can see nothing without special instruments, e.g. you can build interesting computer blocks, like adders, registers, counters etc., made with electromechanical relays or with neon glow lamps, where you can see with your eyes how they function.

[goku12]

What are the factors you expect to limit the integration scale in a garage fab?

[brennanpeterson]

Variance, data rate/cost, and lithography.

You can do lithography small but slow and expensive. But small means you need a stack, which is even more expensive. At small sizes, defectivity/variation are really difficult.

So if you want a paradigmatic shift, you need low cost patterning, and the best way I can see is to use clever chemistry and a much different design style.

[goku12]

Don't you think that a lot more improvement in variability and integration can be achieved with better optics? (for the photolithography, of course. I don't remember what they used for plasma etching and ion implantation.) I don't believe that they have explored a lot on that front yet.

> So if you want a paradigmatic shift, you need low cost patterning, and the best way I can see is to use clever chemistry and a much different design style.

Is that a speculation, or do you have a more concrete idea about what needs improvement and how? I'm especially curious about the 'much different design style' part. Could you elaborate that?

[FarmerPotato]

I heard of one intriguing alternative to photo lithography. Microfluidic channels in a plate (injection molded). I saw a couple research papers in 2021.