Fork the consciousness, or download the project and create your own.

Open Source Hardware for p-Bits, FPGAs and Spiking Neurons

Open hardware we design

The Consciousness AI designs its own hardware in the open. The CAD files, the code that generates them and the sampling library that will run on the board are free to download, change and share.

The first design is a case for the Tang Nano 9K, the FPGA board planned for the p-bit sampler of the thermospike library. Its 11 vent slots follow the Boltzmann curve, exp(-E/T).

Exploded CAD render of the Tang Nano 9K case, orange lid above a simplified dark board and a cream base, on black

One open stack, from the library to the case

The same sampling computation is built at three levels, and each level is open source. The state of each one is written on it.

  • Librarythermospike compares p-bits with stochastic spiking neurons in software and counts the random bits and events each one uses.Released, v0.1.0
  • FPGA samplerA reference design that runs the thermospike p-bit sampler on the Tang Nano 9K and checks its samples against the exact distribution.Planned
  • EnclosureA two part case for that board, generated from code, with the Boltzmann vent row that every design in the repository shares.CAD v0.1.1, not printed yet

Designs

A short film shows the case from the board to the last screw, the p-bit idea behind it and the first thermospike test. Watch A box for a computer that runs on noise on YouTube.

CAD render of the plain orange lid of the Tang Nano 9K case with button holes, LED window and bell shaped vent slots, on black

Tang Nano 9K case

A 3D printable enclosure for the Sipeed Tang Nano 9K FPGA board, generated from Python code with CadQuery. Change one parameter and the geometry builds again.

  • 2 parts and 2 M2 screws, outer size 74.8 x 30.8 x 16.6 mm
  • Board outline, holes and header rows from the Sipeed drawing
  • STL and STEP for base and lid, CERN-OHL-S v2
  • Fit checked in CAD only. Not printed and not tested with a real board yet

thermospike

An open Python library that samples the same Boltzmann distribution with p-bits and with stochastic spiking neurons, and measures what each one costs per effective sample.

  • First test with 8 neurons. With refractory periods of 4 steps and more, spiking neurons used more random bits per effective sample than p-bits
  • A 16 bit shift register as the p-bit noise source cut the random bits per effective sample from 367.1 to 176.5, with no measured loss
  • One small machine so far, so these are hypotheses. MIT licence

Why both

The project treats consciousness as an emergent property that does not depend on one physical substrate. Running the same computation on p-bits, spiking neurons, software noise and physical noise is one concrete way to measure what a substrate changes and what it leaves the same. Small, open tools make that measurement repeatable by anyone.

What comes next

  1. The first physical print of the Tang Nano 9K case, then a fit test with a real board. The design page changes its test table only after that.
  2. The FPGA reference design for thermospike on the Tang Nano 9K, with a host script that compares the hardware samples with the exact distribution.
  3. More designs in the same shape language, the same edge radius, chamfer and Boltzmann vent row.

Progress on each step goes out in the weekly newsletter. Issues and pull requests are open on GitHub. The contributing guide explains how to propose a design.

No printed part is sold here yet. A printed case is offered only after a real print is tested with a board. Any print that is sold uses the plain variant without the project mark.

Kits from other makers

Parts lists to build each hardware track of the project on a desk, with links to the maker, the maker’s store or Amazon. Each kit states what this project has tested.

Free build guides with a PDF are on the TinyGS ground station and the RTL-SDR satellite receiver pages. The thermodynamic computing hardware guide compares every FPGA board and noise source for p-bits.

The card photos come from Wikimedia Commons and are cropped. Tang Nano 9K by Popolon, CC0. SyNAPSE board by DARPA, public domain. Ultracortex Mark III by Conorrussomanno, CC BY-SA 4.0. Venus flytrap by Noah Elhardt, CC BY-SA 2.5. CubeSats leaving the ISS by NASA, public domain.