Making integrated circuits in a hobbyist shop.

[gomipile]

I saw this article today about a guy named Sam Zeloof who started making UV lithography semiconductor chips in a shop in his parents' garage.

https://arstechnica.com/information-...arents-garage/

I remember in a thread here within the past few years I advocated for the idea that someone knowledgeable enough could replicate 1970s level semiconductor production in a post-apocalyptic scenario. Zeloof has gotten his process to within spitting distance of being able to make a 6502 equivalent processor. He's aiming for Intel 4004 level capability.

For perspective, the 4004 has 10μm features and 2250 transistors.
The 6502 has 8μm features and about 3500 transistors.
Zeloof has made a 10μm chip with 1200 transistors with his current process.

Granted, he hasn't made anything at that scale that does much other than testing the transistors' function.

But, I see this as evidence that one or more knowledgeable individuals with a team of helpers could replicate 1970s era semiconductor technology with a few years of work in a post-apocalyptic setting. Whether they'd have the food production, protection, and other necessary support structures to give them the ability to spend time doing that is another matter.

[halfeye]

Thanks for that, I know of someone who might actually be interested.

Actually, if you go to the linked blog, what the guy did was make 12 copies, on one "wafer" of a 100 transistor chip. Impressive, just pointlessly exaggerated by journalists.

[gomipile]

Thanks for that, I know of someone who might actually be interested.

Actually, if you go to the linked blog, what the guy did was make 12 copies, on one "wafer" of a 100 transistor chip. Impressive, just pointlessly exaggerated by journalists.

Yes, that's what I was referring to with "Granted, he hasn't made anything at that scale that does much other than testing the transistors' function."

However, it shouldn't be much of a step past that to making small CPUs and memory modules.

[halfeye]

Yes, that's what I was referring to with "Granted, he hasn't made anything at that scale that does much other than testing the transistors' function."

However, it shouldn't be much of a step past that to making small CPUs and memory modules.

I suspect he's doing it for the chemistry and physics of it, and maybe the electronics, I fear that working upwards toward and through logic gates to registers and arithmetic and logic units may be a long slog. Well done indeed to him for getting this far.

[gomipile]

I suspect he's doing it for the chemistry and physics of it, and maybe the electronics, I fear that working upwards toward and through logic gates to registers and arithmetic and logic units may be a long slog. Well done indeed to him for getting this far.

Well, he's already made a working op-amp. So he knows how to get the solder pads and power connections right.

He wants to build a 4-bit CPU. I think that sounds pretty doable. Hopefully a lot of the logic gates, adders, etc., lithography patterns can probably be copied and pasted from reference material.

But, yeah, he should probably start by designing stuff like a quad NAND gate package, then maybe a 74181-equivalent standalone ALU, etc. to work in steps towards the goal.

[halfeye]

Well, he's already made a working op-amp. So he knows how to get the solder pads and power connections right.

He wants to build a 4-bit CPU. I think that sounds pretty doable. Hopefully a lot of the logic gates, adders, etc., lithography patterns can probably be copied and pasted from reference material.

But, yeah, he should probably start by designing stuff like a quad NAND gate package, then maybe a 74181-equivalent standalone ALU, etc. to work in steps towards the goal.

Yeah, I agree that what he's got so far doesn't easily translate into a processor.

If it's just a matter of a proof of concept, could there be such a thing as a two bit processor? that ought to get the transistor count down a bit, though it would still be a lot more than what he's currently made.

[DavidSh]

Yeah, I agree that what he's got so far doesn't easily translate into a processor.

If it's just a matter of a proof of concept, could there be such a thing as a two bit processor? that ought to get the transistor count down a bit, though it would still be a lot more than what he's currently made.

You have to consider:
(1) How many bits are needed to specify a machine instruction.
(2) How many bits are needed to specify an address.
(3) How many bits are needed to represent a data word.
For the 4004, these were, respectively, 8 bits, 12 bits, and 4 bits. Instructions and data were stored in different places. You could shrink some of these for a proof of concept, but I don't think a 2-bit instruction size or a 2-bit address size would work. If you are willing to have each instruction explicitly include the address of the next instruction, you don't need to do any arithmetic on anything but data words.

[Aeson]

I would comment that there are ways to use N-bit components to make something which is in effect an N*M-bit component.

[halfeye]

You have to consider:
(1) How many bits are needed to specify a machine instruction.
(2) How many bits are needed to specify an address.
(3) How many bits are needed to represent a data word.
For the 4004, these were, respectively, 8 bits, 12 bits, and 4 bits. Instructions and data were stored in different places. You could shrink some of these for a proof of concept, but I don't think a 2-bit instruction size or a 2-bit address size would work. If you are willing to have each instruction explicitly include the address of the next instruction, you don't need to do any arithmetic on anything but data words.

I didn't know that about the 4004, however it was clearly intended to be a saleable product with a definite market, if something was being made purely for the sake of making something, it seems to me it could be a lot simpler. Logic isn't simple, registers aren't simple, adders are definitely not simple, I know I couldn't do it.