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.

This is amazing, thanks a ton for digging this up user. This is proof this can be done, and done cheaply.

Also the patent has been abandoned which is great
patents.google.com/patent/US20120170014A1/en

PDF related is a paper I found that cites the paper mentioned and shows some results from practice runs

Yeah, I guess higher resistance is what would happen if the gate oxide was too wide. But as long as there is at least a little bit of gain, you probably could still have as many logic gates as you can fit on the die, and still drive them at 1 or 2 mhz, wouldn't you? That said, I think an LCD backlight would work to provide an uniform light source. Maybe the stock light source would be enough, but if it isn't you could replace it with UV LEDs.

...like two seconds on startpage . Fifth result for 'homemade mask aligner ic'

I was thinking, could alignment be done with a stationary microscope type thing with a camera? When you first put it down, you snap a pic, then when you go to do the next mask, you basically have the original pic overlaid over the current view, and you line up the registration marks. The camera view wouldn't even have to be perfectly overhead or anything, and long as nothing but the wafer moves between masks.

And then would you swap out the microscope head for an exposure head or something? Might help if you drew a diagram.

The camera doesn't have to be looking straight on. It can be tilted off to the side. Yes, the image will be distorted, but as long as its distorted the same every time it doesn't matter.

Attached: Oekaki.png (500x250, 5.03K)

Sounds like a good idea to me, might need more than one camera though otherwise you'll be prone to overcorrecting it in the opposite direction.

Let's say through some miracle we sucessfully etch a processor. how do we connect all the terminal leads to it? Aren't the ones on mass produced ic's done with gold somehow? How do we go about potting it and stuff, too?

You can just solder a wire onto the terminals:
youtu.be/s1MCi7FliVY?t=1624

Seems like it's just putting some stuff in a mold and letting it dry, so that shouldn't be an issue.

Sorry, not too wide, I meant to say too thick.

So what's the alignment mechanism? They call it a mask aligner, but all I see is a way to expose an already aligned mask? Also, how are masks made, anyway? The point of Sam's system was not needing masks in the first place. Maybe they're just as hard to make as the chips.

Good idea, I see no reason why it wouldn't work, at least for large feature sizes.
Still, I'm not so sure e-beam is that complicated compared to the alternatives, it doesn't require a mask and solves the alignment problem by being a microscope at the same time as a radiation source. The main problem I believe is getting a focused beam. I wonder if there's a way to do that besides making and placing the electrodes mechanically perfectly. Apparently (en.wikipedia.org/wiki/Electron-beam_lithography#Lenses) the main limitation is that the spread of the energy level of the electrons has to be extremely tight so they're all deflected to the same extent by the magnetic or electric fields, which requires that the cathode be made of special materials. Although I suppose beginning with low energy electrons and then progressively accelerating them could also work, since then the energy of the emitted electrons wouldn't matter that much relative to the energy imprinted by the accelerating electrodes. I don't know how the quality of the vaccum affects all of this either. But I know a couple of people have made DIY SEMs, so it can't be that hard (for example hackaday.com/2011/03/23/diy-scanning-electron-microscope/)
The electrons could also be filtered after coming out of the electron gun, which would have to be compensated with longer exposure times or higher voltages.
Maybe it is too complicated, I don't know. But on the other hand making masks every time you want to test some idea would be crazy too, and it's not like a shitty e-beam setup is worse than a shitty optical alignment setup.
Maybe the solution is to use an computer LCD display to generate the patterned light, and then use the mirror to focus it, or just use the original projector idea and just deal with the thicker oxide gates by designing the circuits around the electrical limitations of the mosfets.

Yeah, that's the easy part. Pic related is how he did it, but there's no reason you could just just put epoxy all around it.

Attached: sam-zeloof-chip.jpg (1280x720, 63.78K)

Masks can printed with a decent printer but they'll still have to be scaled down with a lens or something.

allaboutcircuits.com/projects/diy-pcb-photolithography-microfeature-fabrication/

Attached: ClipboardImage.png (2000x2000 265.44 KB, 5.79M)

It will take you years if not decade to designe and build a 32bit computer.
You will NEVER be able to build a computer that can use to shitpost interweb alone.

That's fine, whether because of natural causes or because of (((them))), by the time there aren't any usable pre-made computers, the Internet as we know it will be far gone.
Plus building exotic stuff is fun and the electronics and chemistry stuff translates to other domains too. I'm the Kurt Saxon kinda guy.

8-bit is fine. You can get things done with that, even networking.

Why not use old early 90s hardware? No way they put backdoors on them.

Its cheap. And if you optimize Gentoo on them you could shitpost on interweb and do programming

Attached: consider-the-following.jpg (300x301, 25.01K)

Bump. This is interesting.

The plan is to be prepared for when you have eventual hardware failures with said equipment by having the skills necessary to manufacture or design your own replacements with minimal equipment and resources. Can't exactly keep those original 80s and 90s systems going forever- especially not in the event of a hypothetical disaster/total societal collapse.

Sort of going into mining/refinement though

I don't want a 32bit computer. 8-bits is enough for the important stuff.

Well that's your plan, I just hate israel and china and don't want to be dependent on them.

imagine unironically posting bill nye the pedophile psuedoscience goy and using reddit spacing

Don't slide the thread with inane post.

he's not wrong though

32 bits alone is easy... try pipelining, reasonable sized ram chips and hundreds of mhz devices, then it gets really hard

yeah but how big? there's plenty of demand for the SID isn't there
Well you would want to get as compatible/authentic of a replica as possible, and considering the SID is kind of an analog device you'd do well to copy the circuit pretty much unchanged. but even digital stuff has a lot of undocumented quirks and bugs that would be better replicated by reverse engineering the actual circuit if you want to make a bootleg

Thread theme.

Also, I'm thinking a DLP based setup may be the way to go. Now days you can get true 4K DLP projectors. This would greatly reduce the stairsteps while keeping things pretty simple. All we need is alignment. What kind of machinery do we need for manipulating something on such a small scale? You can get stages from places like Thor Labs, but that shit isn't cheap, and that's for micrometer adjustable stuff.

Attached: 720d2f196c612a0202202ead2bdbee43e5cf369f1ec0cf1cf00d1f354f34604d.webm (800x450, 5.6M)

I was thinking, if the projector solution causes uneven exposure, why not just correct that in software by lowering the brightness in those areas of the image?

Theoretically you could do that but it would have to be adapted to the particular unevenness of the light which would change. Worth a shot though.

I'm routinely impressed by how this board is dedicated to exclusively using tech in the most hardcore possible way.
God speed, you crazy bastards. It better be able to run Gentoo when it's done.

All we need is a camera we can use to get feedback to make a totally even field. It might actually be better to buy parts and make our own custom made UV projector. All we would need is the UV led array, a milky diffuser, the DLP mirror array, and a controller and software for the DLP array.

And some lenses for shrinking everything down, of course.

Yeah, actually, the resolution of the DLP isn't event that big of a deal, as you can just move the die and apply a second exposure on the side, or a third one, or however many are needed, as long as you have the optics to get the picture focused on a small area. It just makes the process more labor intensive.

Are there DLP projectors that already use LED arrays? I'm wondering how much tweaking the optics would need if they all use lamps.

Also, are any of you guys Britbongs? I can build stuff but I don't have a place for doing chemical stuff.

I found a small pocket projector that says it uses and led light source with DLP. Don't know ow much of an array it is. It also only has WVGA resilution.

What wavelengths of UV would acceptable, and what frequency is ideal? I've been looking at UV COB arrays on aliexpress, and wavelength plays a big factor in price. Also what kind of power would be needed? 100w arrays are pretty common, but I've found a 240w panel, as well.

Also, I'm starting to think the homemade DLP setup is the way to go. It would probably be cheaper than trying trying to retrofit UV LEDs onto an exist projector that already costs a grand. We wouldn't be paying for shit we don't need like colorwheels. Here's a quick oekaki of the setup I have in mind.

Attached: Oekaki.png (500x250, 17.08K)

*what wavelength is ideal.
I know frequency and wavelength describe the same property in different ways, but that's not how I wanted to word that.

I believe you want the highest frequency you can get (with EUV being what's used in the industry right now) but pretty much anything would work.

I mean, except for the fact that you'll have trouble using lenses made for visible light when using too high frequency UV, so that could be a problem.

this is the most niggerlicious thread I have ever came across. Fabricating your own chip is pretty hard. To the point where you are a nigger for even thinking about it. Definitely if you want run any modern desktop app on it. Then what you need a graphics? Are you going to fabricate the graphics chip? Then you are going to need to make a compiler for your special instruction set. In the end you might as well just buy a microcontroller made in China that costs 5 dollars and probably doesnt have spyware in it. It will be better then Any niggerlicous CPU you will ever come up with.

Yeah it might be better to use less extreme UV so plain glass can be used. I can't find fused silica lenses in a large enough size for the main large lens. Not at 100w, anyway.

Fuck off, retard. We've already been over this.

well you are retarded for just not buying a 8 bit avr chip for 5 dollars delivered to your door. Instead you are larping about making your own computer with lasers and shit.

Ok. Shortest wavelength COB arrays I have found on aliexpress claim 365nm. Edmund Optics says their planoconvex condenser lenses are good from 350nm-2200. I'm looking for transmittance data on the substrate.

I thought I told you to fuck off?

Maybe they're doing it just for fun, or to put it on their CVs... And also you can make analog ICs that you can't with an FPGA...

Ok, I found the transmittance data. Lenses made from SCHOTT N-BK 7 have 0.971 transmittance at 365nm and 0.977 at 370nm.
This stuff sounds perfectly usable. It might be worth splurging for fused silica on some of the smaller lenses, but for the large condenser lens this should work just fine. It might actually be a lot larger than needed. The specs on these aliexpress COB arrays are kind of vague.

We've gone over this earlier in the thread. We pretty much all have different reasons for wanting to do this.

So what's your reason for wanting to do it?

Because it sounds fun as fuck. I like fucking around and making shit. I hardly ever have the money to make the things I want, though. Also my living arrangements since I graduated college have been shit. My landlords wouldn't appreciate large satellite dishes and whatnot on the roof.

Are you sure you aren't the nigger?

He's a dumb nigger. He's the blackest retard gorilla nigger I have ever seen.

Attached: nobrainniggermonkey.png (990x682, 308.49K)

At this stage, anything in the UV range is good, Like said we might want to stick to the upper end for prototypes in order (mostly for sorting the optics) to avoid having to use UV lenses immediately.
Speaking of these guys seem to have a good range:
edmundoptics.com/c/uv-lenses/784/#28233=28233_s:UGxhbm8tQ29udmV4IExlbnM1&28233=28233_s:RG91YmxlLUNvbnZleCBMZW5z0&29351=29351_s:MjUwIC0gNDI10&28233=28233_s:RG91YmxlLUNvbnZleCBMZW5z0&29351=29351_s:MjUwIC0gNDI10&27661=27661_d:[6.00 TO 500.00]

I feel like you've gone overkill with the lenses, UV lenses are pretty expensive. Pic related is what I'm thinking.

Attached: ClipboardImage.png (1080x1080, 53.06K)

I'll be happy with an 8-bit CPU if my money isn't going to israel and corporate leeches.

Good video on mirror and lens behaviour for any of you that don't into optics:
youtube.com/watch?v=uQE659ICjqQ

Wish I could contribute, but my knowledge of chemistry, optics, semiconductors is NULL.
I'm researching early pocket calculators for fun.
I'm guessing a starting point would be 74 series chips, then MASKROM/EEPROM, recapitulating the progression that lead to CMOS VLSI.

Attached: out.png (512x2176, 188.11K)

Dw fam, you can just ask stuff on the Physics Stackexchange or in this thread. It's also worth watching Zeloof's videos.

I wonder if there are programs to simulate semiconductors and their crystal lattice etc.

That's CMOS, the easier to make kind is NMOS, and the nitride and polysilicon layers aren't strictly required.

Yeah. Edmund has the N-BK7 lenses I was talking about. The 100mm condenser lens is only like 70 bucks. Smaller lenses would be cheaper. Even with four lenses, the total 365nm transmittance would be 0.88. I can live with that. We'd lose way more from the diffuser.

Can already do that. Just go to China and there's plenty of fabs that handle all the work for you, they use big machines. If you're saying homebrew fabs then you'll have to wait.

For who? The passive attitude is gross.
Pic related.

Attached: ClipboardImage.png (500x500, 426.33K)

What are the prices and minimum order sizes?
ehh actually if you're in the US it's pretty straightforward to copy Sam, especially if you're rich and willing to buy a SEM
and maybe if you're willing to step up the budget even more you could get some basic lithography equipment
there's a lot of equipment designed for small production runs done by universities for research purposes

There is fpga to asic shit. Cost less then you expect but more then you are willing to pay. Think of it as buying a small car. Usally if you are making a one off embedded gate logic shit you just us a fpga. FPGA to asic shit cost less per chip but the initial cost is more so you can do the math if you want.

Is the cost for chips with analog circuitry on it higher (per transistor I guess) or the same?

Apparently you can get greyscale DLPs.
Also Zeloof's DLP got a few upgrades, his mitigating the radiance issues with a lens mask and managed 0.5um, he even turned it into a stepper
sam.zeloof.xyz/maskless-photolithography/

I'm starting to lean towards modding a DLP now, mask lithography on a budget would be very limiting and you'd need the stuff to make a mask anyway which is quite similar.

Thoughts on 3D printed photomasks?
sciencedirect.com/science/article/pii/S0003267015004791

Was speaking to a friend and he pointed out that HF is tightly regulated so it's probably worth looking at alternatives also I don't want to die
modutek.com/etching-silicon-wafers-without-hydrofluoric-acid/

Who says we need 32-bit? I can do 99% of my computing on my IMSAI with a 2MHz 8-bit Z80.

Contiki OS has a TCP/IP stack and a browser that will run on a 6502. I've not tested it on 8ch, but I've seen it post to a smaller chan whose name escapes me.
If you could build individual transistors, you could replicate the MOnSter 6502 project. There's also the C74-6502 project, to build a C64 from 7400 series TTL. The logic is out there, it would (hopefully) just be a matter of translating it to a usable form on homebrew components. Speed could be an issue as it is with the MOnSter (it can't replace the Apple ]['s CPU for timing reasons, for example), but there are already systems that are compatible with it, such as the Cactus.
Even if you could only get a few hundred KHz out of it, with a fast, character addressing serial terminal it would still feel more responsive than the 110baud teletypes people put up with in the 70s.

Apparently you can get 20% KOH off Amazon, anything over 16% needs a loicence in Bongistan but the etch time is pretty bad for 20% already so idk

Attached: KOH.JPG.jpg (625x473, 34.44K)

cleanroom.byu.edu/KOH
Looks like you can speed it up with isobroply alcohol

Based on this you can get a 3cm wafer etched in 2 days, which is alright

You may be right, but there was also the clipper chip around then, so the glowinthedarks were at least trying. You can't keep those systems running forever though. I think 60-70s systems based on transistors or simple TTL ICs will outlast 90s stuff purely because most of the parts (or equivalents) are still in production, and the parts that aren't are simple enough to work around. (I recall seeing a 70s-era minicomputer controlled mass spectrometer having a couple of single IC ALUs replaced with a drop-in clone made from handful of 4000 series chips. It was ugly but it worked)

That could work but it wouldn't be fun. I've seen it done and booting alone took nearly 15 minutes on a 486 (can't remember which model). You'd be better off patching the shit out of a 2.4 series kernel. Slackware was usable on a DX2-66 into the early 2000s.

Glass etcher/rust remover. You can always make your own. Or, you could just move to freedomland. Yes there are regulated chemicals, but that's mostly just for ones needed for drug manufacture.

Move to Murrica already. Holy shit. You seriously need a loicense for drain cleaner? I have a friend that gets pure KOH off eBay for making liquid soap. You can buy relatively pure NaOH and various acids at most hardware stores.

I'd rather move literally anywhere else. You don't need one for drain cleaner I'm guessing our drain cleaner just has different stuff in it or a lower concentration.

So I was looking at Zeloof's pictures. It looks like he was using a trinocular microscope. The projector is projecting down the trinocular port where the camera adapter would normally go. He then uses a usb camera adapter in place of one of the eyepieces. Pretty clever.