Extropic

Unofficial concept

X0 probabilistic circuits

Programmable randomness

Extropic’s X0 chip proved four probabilistic circuits in silicon. They turn the thermal noise of ordinary transistors into samples from a probability distribution you program. Tune them below and watch them sample.

Simulation

The samples on this page come from a pseudo-random generator running in your browser, not from an Extropic chip.

01 pbit

A weighted coin

Samples from a Bernoulli distribution: it outputs 0 or 1, and you set how likely the 1 is. The pbit is the building block of Extropic’s chips.

0.70

02 pdit

A weighted die

Samples from a categorical distribution: one of k outcomes, each with its own probability. This one has four outcomes, 0 to 3. Set their relative weights.

1
5
2
4

03 pmode

A programmable Gaussian

Samples a continuous value from a Gaussian (bell-curve) distribution. You program its mean, where the curve is centred, and its variance, how widely the values spread.

0.5
0.60

04 pmog

A mixture of Gaussians

Combines the two circuits above. For every sample, a pdit picks one of several pmodes, and that pmode produces the value. Here a pdit chooses between two.

0.40
−1.6
1.4
0.30

Where they fit

From one circuit to a chip

  1. Circuits

    pbit, pdit, pmode, pmog

    Each pairs a sampled state with the distribution it comes from.

  2. X0 · 2025

    Silicon prototype

    Extropic’s first chip, which proved these circuits could be manufactured.

  3. XTR-0 · 2025

    Desktop platform

    Connects Extropic chips to a conventional processor, so early adopters can run programs on them.

  4. Z1 · taped out 2026

    269,568 pbits

    Each pbit wired to sixteen neighbours, so the chip samples a whole network of related variables in place.

To program them, Extropic publishes open-source libraries: THRML and Torx.