Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Second IC (zeloof.xyz)
406 points by picture on Aug 13, 2021 | hide | past | favorite | 79 comments


> Now we know that it’s possible to make really good transistors with impure chemicals, no cleanroom, and homemade equipment.

The good news here is that 2 of those 3 issues are pretty easily solvable. With some work, and a ton of HEPA air purifier machines, you could probably turn an area the size of a small shed or 1 car garage into a class 10000 cleanrom pretty easily. It would be simplest if this area were embedded in a larger area, like a 2 car garage, but, you could probably squeeze it into just a plain old 1 car garage.

Chemicals, you can buy online from scientific supply houses. Some, like acetone, you can buy from less specialized sources in high purity. Water and alcohol, you can either buy at high purity pretty easily, or buy lower grade stuff and purify it.

Of course, at the point where you're going for higher purity stuff like this, you might want to switch from acetone as a solvent to DMSO. DMSO itself is safe to handle and doesn't evaporate like acetone will.

> Silane

Yeah, nasty stuff. Toxic as hell, ignites spontaneously when exposed to air. I used to work near a rather large storage tank that contained the stuff. Needless to say, I am glad there never was an accident. :-) Very wise to avoid this stuff.

---

https://www.americancleanrooms.com/class-10000-clean-room/

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

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


His achievements are great, but the only reason why he could this without a cleanroom and with less pure chemicals is that he succeeded to find on eBay a batch of silicon wafers with the MOS gate oxide already grown on them.

Growing the MOS gate oxide without extreme cleanness has no chance of succeeding.

The difficulty of this process has caused a delay of many decades between the time when the MOS transistor was first imagined (in 1928 by Julius Edgar Lilienfeld, then also in 1935 by Oskar Heil) and the time when working MOS transistors were made for the first time, around 1960.


Want to stress how extreme cleanliness have to be. Free halogen ions from the air are enough to mess it up.

That's why early fabs had very unusual placement requirements: not close to chicken farms, agriculture, water, or sewage treatment plants.

Intel famously got seasonal yield losses which they traced to agricultural activities, and seasonal chicken coup cleaning.


Are they able to clean the air well enough now that it’s not an issue?


It’s not really feasible to separate noble gases out of the atmosphere in a building HVAC system.


Couldn't you just use positive pressure of O2 and nitrogen? Push all other gasses out? Would cost a lot but it is doable.


More recent tools do work that way. The whole "tool line" is flushed with process atmosphere.


Probably the buildings are so big nowadays that it's geometrically impossible to have farms nearby the actual plant.


How much m^3 do you need to grow these structures? Couldn't it be done in a very confined space?


This is how it's done now. Tool lines are flushed with chemically regenerated atmosphere (basically getting LOX + LN2 bottles from air liquide, and mixing them,) or argon, or nitrogen only.

Older, and cheaper fabs still rely on natural atmosphere, and other, less sure tricks to keep halogens out.


Are these unfinished wafers commonly used/sold in the industry?


A class 10000 clean bench might be more practical. A lot easier to keep your cleanroom clean when the humans are outside of it, just sticking gloved arms in from time to time.

Be careful about purifying alcohol. Some backward countries have religiously-based legal prohibitions on this. My own backward country tightly controls acetone as a "precursor" to "stupefacients".

Nice thing about silane is that you don't have to worry about accidentally breathing it. There's a cheap and simple process for manufacturing it on demand from magnesium silicide, so you don't have to store it either.


> With some work, and a ton of HEPA air purifier machines, you could probably turn an area the size of a small shed or 1 car garage into a class 10000 cleanrom pretty easily.

I wonder if it would be possible to automate the entire process with robotics and reduce the effective "clean" space even further.

There was this post a while back about a company that used thin cassettes to keep the wafers clean when moving between process steps.


Is DMSO really a replacement for acetone in most applications?


In some applications. It is dangerous if it dissolves things because now they can get under your skin easily. The problem is that it is not easily removed (requires much more energy than acetone) and it oxidizes. It is also more viscous and freezes at pretty high temperature (around 20°C).


Related threads:

Homemade IC (2018) - https://news.ycombinator.com/item?id=21239145 - Oct 2019 (40 comments)

Home Chip Fab - https://news.ycombinator.com/item?id=20657398 - Aug 2019 (131 comments)

A home-made lithographically-fabricated integrated circuit - https://news.ycombinator.com/item?id=16955549 - April 2018 (116 comments)


I want Sam to turn this into a startup somehow. There has got to be something better than the semiconductor industry we have today, which has ridiculous barriers to entry.


Turning old processes into a startup is what Melexis did 30+ years ago: they acquired obsolete fabs in Eastern(?) Europe and repurposed them to make sensors, which don’t need state of the art processes.

The 180nm open source eFabless shuttle runs that are sponsored by Google are done by X-fab, which is owned by Melexis’ founder as well.

I don’t see how this garage stuff (impressive as it is!) can be turned into a startup. It’s not as if there is some kind of breakthrough here?


> It’s not as if there is some kind of breakthrough here

The [potential] breakthrough is in accessibility. Like GP said, this is a non-cleanroom process. Package it up into something reproducible and sell the entire 10um node for 6 figures and you'll have a few takers (unis, high networth enthusiasts). Get it down to 5 figures and you can break into some of the hobbyist market. Iterate/improve the node from there.

The path from here to unicorn startup is not really easy to imagine, but getting from here to cottage industry has some sliver of a chance if Sam wanted to go that route.


Making demos is easy, making a product is hard.

Universities could totally do what he is doing. But that’s still far removed from commercializing this tech (unless the product is “a mini fab for universities for educational purposes only.”)

For a real product, you’d need to meet all kinds of reliability requirements: life time, voltage range, temperature range, parameter control etc.

(Once again: what he’s doing is awesome. It’s just important to keep some perspective.)


Would need a change in the environment. There's no economic reason to do it now, when we've got global supply chains and established businesses with decades of IP. There'd be a strong reason to do it if those supply chains collapsed and we needed to distribute semiconductor manufacturing across a number of separate low-profile sites, if, for example, we entered war, civil disorder, or a migration crisis.


I wonder if it could support specialty audio chip manufacturing.


A company called Alfa in Latvia does this, they use outdated semiconductor fabs (probably from Soviet union era) to make reproductions of vintage synth chips from the 80s, CEM33xx etc.


Exit via Cirrus Logic.


It's not strictly semiconductor-related, but if there's one thing I would like to see, it's low cost, made-to-order integrated analog computers. The low power consumption of analog devices would offset the larger feature size, and hopefully the process could be made cheap/flexible enough so that needing a different IC design for each use case would be a non-issue.


Larger feature size would help the power consumption in this case as well - larger distance between transistor features means less leakage which means less static power consumption. Dynamic power consumption would of course increase but that shouldn't be much of an issue with analog chips.


You can get your chip fabbed at the Skywater foundry for 10K (can't remember how many dies that buys you), the open source toolchain does support analog if you know what you are doing.


I was thinking about something like that too. How hard would it be to have a semiconductor-manufacturing-as-a-service with lower inertia like those PCB manufacturing services? The hobbyist affordable board houses these days generally have extremely good processes and quality that even beat some "industrial" board houses, as they're thoroughly refined by fabricating hobbyist boards with sometimes really awful DFM considerations


There are companies working on scaling e-beam lithography, which might be along the lines you're looking for [1] [2]

[1] https://en.wikipedia.org/wiki/Electron-beam_lithography

[2] http://www.multibeamcorp.com/


You have to find the capability / price point sweet spot and you could probably create a big market, that kind of thing has been happening more and more lately... but it takes a lot of domain knowledge and luck.

A good example of this is how Arduino broke open the hobbyist electronics market which went from circuit programmers and prototype boards that would run in hundreds of dollars on the low end to $20 or something to get in at the ground level.


Given the relative affordability of Skywater's service nowadays, I imagine we will eventually see custom silicon projects on Kickstarter and the like.


Masks are expensive


Which is probably why Sam's process uses a DLP projector install of masks.


A mask at this scale would just be a few hundred dollars I think (chrome on glass)?


The cheapest Taiwanese mask shops who work for discrete device manufacturers will bill you 10x at least.


I imagine this will end up like 3D printing though.

A 3D printer is fantastic to prototype your own designs. Or to build custom parts in low volume.

But it will never replace traditional manufacturing at scale. And even CAD is a massive barrier to entry.

And 3D printing is far more useful for everyday things but we still haven’t figured out a model to make the technology accessible to the general public.

This is a very impressive project though. I’m envious.


But the model for it to be accessible to the general public is readily available, it's just not evenly distributed.

I can go to several local fed-ex offices here in Vancouver and they will print on demand a file that I provide them, for a very reasonable cost (given that they eat the cost of misprints and supervision of the process).

There are a number of fairly successful print on demand businesses that are niche specialized as well, heroforge is a great example.


I can go to several local fed-ex offices here in Vancouver and they will print on demand a file that I provide them, for a very reasonable cost (given that they eat the cost of misprints and supervision of the process).

That sounds more of what I am thinking of. Australia has a population of similar magnitude, but I've never heard of such a thing here.

I disagree that Heroforge is a good example of 3D printing being available to the general public. Firstly it's literally a niche product. Shapeways is more general and I've ordered from them before, but international shipping is very expensive. I can order an Ikea replacement part for $10 and then pay $27 for shipping.

Most of my designs and prints are often for very mundane things. Missing pieces or broken parts that would save you 10s or hundreds of dollars. But most people think of 3D printing as a high tech manufacturing process. Or they think of the hobby as something for making toys. Most don't fathom that you can use the technology to fix your headphones or replace a broken part.

And this is a technology that has direct application to people lives right now. Custom ICs are very cool and even more niche.


Well 3D printing is super great for fast, functional prototyping. So if that's all it is, it'd be great!


We have that problem of trusting our compilers, and the guys that are trying to build a completely clean bootstrap of gcc (?). The ability to manufacture some known-safe hardware might help that project.


> There has got to be something better than the semiconductor industry we have today, which has ridiculous barriers to entry.

I am curious about your exact meaning. When you say barriers to entry, it sounds like you're implying the "semiconductior industry we have today" is actively putting up barriers to prevent you from entering? Is that what you are trying to say? If so, please elaborate so that we can understand what these barriers you are facing are.

Or is it just that you're acknowledging that it actually costs many billions to build up a commercially viable fab?


I notice no response and yet downvotes. Parent claims there are "ridiculous barriers to entry" implying patent thickets by the "semiconductor industry we have today" or something along those lines. I can't help but notice no further explanation was provided to substantiate parent's claims.


This bothers me that these days average engineer can't afford to buy or rent reasonable sized space to do experiments like this. My dream is to build my own IC but I am sure doing that in my living room would likely turn it into something uninhabitable. I've been looking at renting some space in an industrial zone, but that's out of reach money wise even in a remote area.


Does anyone have a rough sense what the cost of Sam’s equipment is, and how someone can emulate this on a budget with second hand purchases?

Also roughly, would anyone know the cost per batch of chips? I’d love to run thing as a learning exercise for middle or high school students and wonder how costly it would be.


His lab would cost a fortune if you were buying the equipment new from the manufacturer. But he has been able to source a lot of decommissioned semiconductor fab equipment through eBay and donations etc. Like his wire bonding machine. He has spent years of effort setting it all up, so it's not something for a one off project.

But replicating his simpler transistor experiments should be doable on a smaller budget. Also check out Jeri Ellsworth's DIY transistor project. Some nasty chemicals and special equipment is necessary but nothing too much for a hobbyist.


I don’t know the minimum amount equipment needed to make these chips, but his garage is stuffed with old but still expensive equipment (if you need to buy it on eBay.) Things like a wire bonder, an electron microscope, a vacuum chamber etc.

I think that most of these were donated to him though.


"Especially my amazing parents, who not only support and encourage me in any way they can but also give me a space to work in and put up with the electricity costs… Thank you!"


He's to the point now where he could make a small ASIC, or CPU, or maybe a high power FET (by paralleling a lot of them)

I wish him all the best of luck.


Would love to know the ancillary stuff like waste disposal. Playing with this stuff without making a new superfund site is arguably just as difficult as the core work.


I recently visited the Berkeley Nanolab. As I recall, the main waste products are either acids that are neutralized before being returned to the city sewer system, or exotic gases that are vented in dilute enough quantities to be safe.


To be clear, my comment isn't a callout; I unironically believe that it's explicitly being addressed. I'd just find a writeup around the subject interesting.


the campus runs its own sewer since 1867. Feeds into East Bay MUD.


I believe Sam avoids using anything that's super awful to deal with. There's well defined procedures for taking care of most chemicals that involve neutralizing them sufficiently to be flushed down the drain


Even the specifics there of what subset of materials gets you to making chips without disposal issues would be super interesting to me. It's a complex space to navigate, and shrouded a bit in mystery and trade secrets.


The "HF" doesn't stand for "high five!"?


HF is no big deal, haven't you watched breaking bad? Piece of cake, just use plastic containers. What could go wrong?


Its actually easy to dispose of, just neutralize with calcium carbonate (forms safe insoluble calcium fluoride) and dispose of down the drain or in the trash. Just don't get it in on your skin in the process. There are other options to neutralize as well.


I was surprised to find that 1% dilute HF is both adequate to purpose, and safe to handle.


Safe is relative. I'm such a genius that I managed to injure myself doing a reaction with a single reagent, namely, making amorphous plastic sulfur. I accidentally caught the sulfur on fire, you see. If you ever find yourself in this situation, resist the temptation to try to blow it out; you will regret inhaling near a sulfur fire.


Low concentration HF is frequently uses in artistic glass etching. With reasonable precautions it’s not more dangerous than cleaning your drain with caustic soda. Just don’t be stupid.


What matters is that if you splash some on your hand, you have time to rinse it off before the flesh melts off your bones.


Tl;dr Don’t listen to amateur forum advice, get informed at professional level.

Long: Please, In general I suggest for anyone dealing with chemicals to read the safety sheets to the very least and ask for professional advice.

Very often you will find second order effects at play (teratogenic, cancerogenic, allergogenic, er cetera), as well as dangerous surprises that can await you in reactions.

HF is a very nasty chemical that not only does funny stuff to your bones but can kill you making your breathing stop. Rinsing off your hand does not help in this case, you will need to get a specialised treatment in hospital. We lost two students at the University of Marburg in Germany to HF exposure.


Yet, I doubt that it was 1% HF solution, as may be purchased in any hardware store, that they were exposed to.


I wonder how far it would be possible to go with an open source / open hardware to have open process for IC development and production.

What I mean is given relatively old process but new tooling and software, could we expect to make usable system that is completely transparent from security point of view? And resistant to future attempts at preventing people from having access to trustworthy hardware?


I'm thinking about trying to take this on myself, after watching and reading everything from Sam and Huygens Optics, and going through Chris Mack's course online.

I think the problems for offering any kind of hardware solution for this is that everything involved is either "pretty easy" or "pretty darn hard".

A decent UV laser, a workable optics bench, and 3/4 of the process chemicals are cheap and readily available. You can even build a "clean enough" fume hood. The processes are decently documented already in an open way.

Super high precision 3x axis closed loop micrometer or piezo stages, a wire bonder, and good quality pre-layered wafers are so far beyond small distributor or hobbyist scale that they're relegated to elusive Ebay scores. The other 1/4 of the process chemicals require getting a big distributor like Sigma to work with you, and some of them really require chemistry training to work with safely, although it's great to see some of the workarounds. I believe Jeroen at Huygens Optics is a chemist by training. He's comfortable working with things like HF that you really don't want to promote to amateurs, but are essential to get past a few thousand transistors.

Getting a complete process beyond the 70s era stuff at an amateur level is probably impossible right now. The absolutely insane requirements for silicon growing/cleaning alone are just more than one person can fit in their head, or their garage.

I'd compare it to large liquid fueled rocket engines, which are another common nerd dream project. Plenty of people understand the principles, and a few college teams or the equivalent have built working ones, but there's probably just too much to know and understand to do it solo and with no serious budget.


To make the economics work, you'd probably want to pool funding & run many wafers. Perhaps you could develop the world's first MaaS venture (Manufacturing As A Service)?


This is all going to depend on exactly how much $$$ you have to spend. The major step-wise jumps in cost can usually be attributed to higher resolution photo techniques and the precision tools required.

Most of the other processes can theoretically be carried out at garage-scale, as the precision is more about chemistry than dimension or scale.


People buy 3d printers. I'm sure there would be people interested in building their own silicon chips, even if it would be a lot more involved.

I can imagine a little niche market for tools and services for that kind of production. I design electronics but I order my PCBs from jlcpcb because I can get much better product with much less hassle.


Very cool and impressive.

Here is another person who had a fab at home at some point : http://www.microfab.co.uk

They were building not microchips but STJ sensors.


Sam is one of my favorite content creators, all his stuff is amazing.


It's a pretty clever process as some of the more critical steps (gate oxide, poly deposition) are done elsewhere. This allows for very stable transistors without a lot of Vth shift. I also like his use of resits for the ILD.

Maybe it would be possible to replace some of the more agreessive acids used for poly and Al etch with KOH and lowly concentrated HF.

Also, one could also use alloy doping with aluminium to create the junctions and skip the need for a high temperature diffusion. But that would probably not result in a 10µm process and present yield issues...


Really cool stuff, I was wondering if anyone was doing diy photolithography.



It would be so cool if microprocessors went the Linux route and the best in class were created from open source projects.

Well one can dream.


It's time for some big software companies like Google to commoditize their complement. So who knows ...


Isn't that what FPGAs are good for?


FPGAs are nowhere near as fast as the equivalent ASIC, not to mention the toolchains are an utter disaster. There’s a very narrow window in which they’re useful, which certainly doesn’t include anything meant to be best in class.


Super cool. These kind of things are needed to keep things from being perceived as being “impossible”


This is such amazing stuff!




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: