DIY Semiconductor Fabrication

In the meantime learn HDL and program FPGAs. You can make all sorts of custom processors.

Any idea as to how a diffuser might stack up against a lens array?

>Anyway, Thor Labs has a 10mmx10mm lens array forthorlabs.com/newgrouppage9.cfm?objectgroup_id=2861 428ish bucks. Will that work?
Looks like it might, it's pretty expensive though so I'll try it without the lens array first, I'll note these guys down though.

Yeah, that was the one step on which Sam provided no info.
Well it could be useful for building discrete transistors for things like radios, and for a vague enough definition of "point contact" it could be used to build power transistors.
Meh, I'm not sure if the impurities would matter that much at the scales and frequencies an amateur would be working at. And if you just want to get a die done you can always just send it to China, the main point of making your own IMO is if shit goes down for years and there's a disruption in the supply chain and trucking companies and factories stop operating and so on. So it would be ideal if everything could be sourced locally, although I guess you could store a lot of these.
If your motivation is CPU backdoors you can always use an FPGA, it would be very hard to put a backdoor on those.
What's the cost anyway of having a large feature size ASIC built on China? Why hasn't anyone for example commissioned a batch of MOS 6581 (sid chip)? Is it because nobody can be bothered to reverse engineer the circuitry or because it's actually expensive?

Why would that happen with mirrors as opposed to lenses?
Oh, I didn't know he did. Links?
The good thing about e-beam is I believe you can solve the alignment problem without complicated mechanical arrangements (which would be pretty hard to be machined by an amateur) by sensing the current through a conductive test pattern (like a few crosses on the corner or something) on the die that's connected to ground, or like the SEMs do it, by measuring reflected electrons. And the other advantage is that you don't need complicated optical arrangements and you aren't limited to any particular resolution unlike DLPs. Even if your DAC isn't very accurate you can do things like PWM through a capacitor.
The main problem is having and setting up the turbopumps etc. The secondary problem is beam focus. The high voltages I don't think are really a problem, you can always reduce the cathode current and wait more time for the stuff to be exposed, so you don't really need a giant transformer and so on.
If you're going with the optical route I had the idea that you could do the fine steering by passing current of the metal rods supporting it, regulating their temperature and thus their length. You would still need some way to sense the offsets though.

Prior to switching to E-beam lithography he jerry rigged a projector with a lens, that's basically what he used.
youtube.com/watch?v=XVoldtNpIzI

They do, wafers are expensive.

We all have our own motivations.

Because chip fabs usually only take bulk orders or contracts

There's very little use reverse engineering these old designs, even at larger than modern scales the knowledge we've gained since will allow us to create radically different designs that can make better use of our resources.

I think mirrors might work, I was speaking about lenses because that's what I'd seen contemporary designs using.

youtube.com/watch?v=SB94rQtKlKI
I don't even think we should bother exploring this though since it's far to expensive and fiddly.

You're describing a stepper here, which is great for smaller feature sizes but we don't want to worry about that, not yet at least.

Yeah, I know he used a projector, I mean he didn't provide the info on how to align the different exposures.
Isn't that the main problem with his setup? Otherwise if you're willing to buy the wafers, the chemicals and the projector, single exposures are pretty much a solved problem, and the digital light processing device provides more resolution than the precision with which you can align the exposures without the stepper, doesn't it?

Not at all. I just know it's how things like light meters do spatial integration. The whole point is to create an even field of light instead of having a gaussian distribution.

You get free snacks with your order. They're cool about small orders. We used them in college to get parts for a custom Shack-Hartmann wavefront sensor.

Well it depends what perspective you take. If your aim is to decrease feature size then yes, but if your aim is to provide a reliable, easily reproducible process for the average autist (which is what I'm doing) then no.

No, single exposures are very unreliable with the projector setup, mostly because the irradiance is non-uniform which is why we were discussing fly eye lenses and LED arrays, because these make the intensity uniform, which prevents overexposure/underexposure of certain areas.

I'm not quite sure what you're saying here, for the DLP to work you still need to align the wafer, it's just done in one shot. With a stepper each part of the wafer is done progressively.

Yeah I think I wasn't actually describing a stepper, according to Wikipedia the stepper is the thing used for exposing multiple dies on a single wafer by repeatedly exposing the same mask multiple times. No, that's not what I was describing. What I was describing is called a "mask aligner".
What I meant was aligning the different masks for the construction of a single die. IIRC for MOSFETs the bare minimum is two masks, first the holes for etching the oxide layer for doping, and then the holes for etching the metal layer where there isn't supposed to be metal.
So my point is he didn't describe how he did that one shot alignment.
Unless you can see the drain, source and gates on each transistor with your bare eyes you're going to need at least an optical microscope with some way of holding the die while finely moving the mask, and for a whole CPU even that is probably not gonna work because the feature size is sub-optical.
Do you have a link to where he described what problems the uneven exposure caused? I mean what do those defects look like with respect to the electrical characteristics of the devices?

This one looks like they're using an array and a diffuser
nanowerk.com/spotlight/spotid=23714.php

They can be, but they can also be used to expose a single mask piece by piece in order to get a smaller feature size.

That's because he just eyeballed it. The mask just has to have light passing through it and the substrates he uses are bigger than the mask.

I think he talks about it in the DLP video I linked before.

It depends on what kind of substrate you use, but it'll either cause too much resistance or a short.