4 ms·

I've been following home-made semiconductors since Jeri Ellsworth made her own NMOS FETs in 2010.[1] When Sam managed to make his own ICs, I was happy for him,

I've been following home-made semiconductors since Jeri Ellsworth made her own NMOS FETs in 2010.[1] When Sam managed to make his own ICs, I was happy for him, and for the hobby. Now he seems to be on the verge of democratizing access to semiconductor production in very low volumes. I've got my own project to try out once I can get my hands on it.[2] Who knows, petaflop chips made in the garage could be right around the corner, in less than 2 decades! [1] https://www.youtube.com/watch?v=w_znRopGtbE https://www.youtube.com/watch?v=w_znRopGtbE [2] https://bitgrid.blogspot.com/ https://bitgrid.blogspot.com/

RetroTechie 3y ago
About [2], BitGrid: If I read correctly, that concept maps really well to LUTs as used in FPGAs (but note it's not the same!). Its author/inventor Mike Warot has the simple but useful idea of making the configuration space of such a device a simple memory array (like RAM or EEPROM cells). Write some bits to a memory location -> cell [x,y] in the device's grid is configured. This would make it trivial to control/interface such a 'compute grid' to traditional cpus. To a cpu it would just look like regular RAM. Anyone know if there's work towards applying that idea to FPGAs? It would basically evaporate the problem of [closed source FPGA vendor toolchains]. Obviously, bitstreams produced by FPGA vendor tools will do a somewhat similar thing. But what/how exactly is opaque. Not to mention encrypted bitstreams, read-protection & what have you. Hence require re-engineering effort for open source tool support. Turning this into a dead simple 1:1 mapping would be a huge step forward.
xw3098 3y ago
You can have this now https://efabless.com/ https://efabless.com/
- mikewarot 3y ago
  I can't... you might be able to afford $10k, but right now, I had to think hard before spending $3 last night. I hope to learn enough Verilog/VHDL/RTL to take advantage of the Free Shuttle program they have[1] eventually. Currently, I'm wondering how to go from expressions to code I can load into a bitgrid. The thought of parsing code, building an AST, converting that to an acyclic directed graph of binary expressions, to actual bits to load the grid with.... overwhelms me, and has me putting it off. 8( I do have a working simulator, written in Pascal[2], that can run fairly quickly for toy problems on my desktop PC. So, there's been some progress. I strongly suspect I can get there, but need an accountability buddy to push me along. I used to be a self starter, but I stayed in an IT Admin job too long that crushed it out of me. [1] https://efabless.com/open_shuttle_program https://efabless.com/open_shuttle_program [2] https://github.com/mikewarot/Bitgrid https://github.com/mikewarot/Bitgrid
  - xw3098 3y ago
    You might try starting here then. It’s a good community as well https://www.tinytapeout.com/ https://www.tinytapeout.com/ You’re not required to buy the tile to take the lessons
    - mikewarot 3y ago
      That looks awesome... thanks!

RetroTechie 3y ago

About [2], BitGrid: If I read correctly, that concept maps really well to LUTs as used in FPGAs (but note it's not the same!). Its author/inventor Mike Warot has the simple but useful idea of making the configuration space of such a device a simple memory array (like RAM or EEPROM cells). Write some bits to a memory location -> cell [x,y] in the device's grid is configured. This would make it trivial to control/interface such a 'compute grid' to traditional cpus. To a cpu it would just look like regular RAM. Anyone know if there's work towards applying that idea to FPGAs? It would basically evaporate the problem of [closed source FPGA vendor toolchains]. Obviously, bitstreams produced by FPGA vendor tools will do a somewhat similar thing. But what/how exactly is opaque. Not to mention encrypted bitstreams, read-protection & what have you. Hence require re-engineering effort for open source tool support. Turning this into a dead simple 1:1 mapping would be a huge step forward.

xw3098 3y ago

You can have this now https://efabless.com/ https://efabless.com/

mikewarot 3y ago
I can't... you might be able to afford $10k, but right now, I had to think hard before spending $3 last night. I hope to learn enough Verilog/VHDL/RTL to take advantage of the Free Shuttle program they have[1] eventually. Currently, I'm wondering how to go from expressions to code I can load into a bitgrid. The thought of parsing code, building an AST, converting that to an acyclic directed graph of binary expressions, to actual bits to load the grid with.... overwhelms me, and has me putting it off. 8( I do have a working simulator, written in Pascal[2], that can run fairly quickly for toy problems on my desktop PC. So, there's been some progress. I strongly suspect I can get there, but need an accountability buddy to push me along. I used to be a self starter, but I stayed in an IT Admin job too long that crushed it out of me. [1] https://efabless.com/open_shuttle_program https://efabless.com/open_shuttle_program [2] https://github.com/mikewarot/Bitgrid https://github.com/mikewarot/Bitgrid
- xw3098 3y ago
  You might try starting here then. It’s a good community as well https://www.tinytapeout.com/ https://www.tinytapeout.com/ You’re not required to buy the tile to take the lessons
  - mikewarot 3y ago
    That looks awesome... thanks!