Cool - it's amazing how much easier FPGA development has gotten in recent years thanks to efforts like YoSys and these sorts of frontend projects.
RISC-V has also been maturing over the same timeframe, which is bringing mainstream open-source efforts deeper into the stack of general-purpose computers.
We still don't have an open-source VLSI fab, but costs are always falling and Magic has been open-source for decades, so maybe it's only a matter of time until samizdat-style CPUs become possible :)
it would be great to see it end up more like PCB fabrication, where anyone can pick up open source tools and then send files off for fabrication in a few days/weeks at moderate cost... I'd pay a few hundred just for the novelty of getting a chip of my own design.
I do think the trends in the industry may allow this to happen in the next decade or so.
That exists. If a larger process is ok you can get 10 custom packaged ICs for less than $5K. It doesn’t meet your budget but it’s about the cost of a beater car. Depending on the car. ;)
Something like a 3D printer that could produce integrated circuit dies from blank wafers; "VLSI" means "Very Large-Scale Integration". But we're still a long way from that being feasible.
Even if you had a magic box that could turn wafers into dies, you'd still need to be able to source the wafers, ensure an extremely clean environment, screen for imperfections/failures, and package the dies into chips which are robust enough to be handled, soldered, etc.
Some people like Jeri Ellsworth have managed to fabricate individual transistors in the garage, which is extremely impressive, but it's still a long way from there to a packaged integrated circuit.
Cool paper. Thanks for sharing. I’ve worked with OTFT before and this paragraph from the paper pretty much sums up my experience too.
“The operating frequency of the base-line design for OTFTs is approximately 200 Hz while for silicon it is 800 MHz. The optimized design frequency is approximately 40 Hz and 1.36 GHz respectively for organics and silicon. The modest per- formance of OTFTs can be a decent match for applications that need modest computing such as sensors, RFIDs, toys, remote controllers, etc., as described in Section 2. Many of these embedded processor use cases employ embedded processors in the kHz range [31, 35].”
I guess I was wondering what make a fab open source. I.e. say you did have a 3D chip printer, but you didn’t know how it worked and you couldn’t look inside, would that be open source? If that printer counts as open source then I would argue fabs like MOSIS are the equivalent.
You can get all the tech files for their older stuff, with no NDA, from FreePDK. It works fine with VLSI layout tools like magic, which one can git clone the source of.
That said, there isn’t really a lot of mystery in the larger technology sizes like 0.18um. Plenty of universities will tell you exactly how they do it if you look into it.
The big problem with DIY fab ideas is chemical disposal. This will have to be addressed first. Parts of Mountain View are still a superfund site because of older fabs. HF is no joke.
This is exciting. You can do VHDL synthesis with the commercial version of yosys now (uses the Verific commercial parser). Will be great to have an open source solution.
GHDL is written in Ada, if you have ever wanted to see that in the wild. It looks like this is in C++, which will probably make it easier to get contributions
RISC-V has also been maturing over the same timeframe, which is bringing mainstream open-source efforts deeper into the stack of general-purpose computers.
We still don't have an open-source VLSI fab, but costs are always falling and Magic has been open-source for decades, so maybe it's only a matter of time until samizdat-style CPUs become possible :)