Thermodynamic and probabilistic computing: implementation evidence

← Return to the implementation matrix

These systems use controllable physical noise or relaxation to sample from distributions and solve probabilistic subproblems. Noise is part of the operation rather than merely an error to suppress.

Stage Mark Summary
Reference THRML software environment
Physical Reported stochastic CMOS silicon
Integrated Prototype sampling platforms
Scaled No broad scale demonstrated
Access Open software and restricted hardware access
Operational No recurring external deployment located

Reference — demonstrated

Claim
The matrix credits thermodynamic and probabilistic computing at the Reference stage: THRML software environment.
Evidence
THRML provides a JAX library for thermodynamic hypergraphical models. It lets researchers define, train, and test probabilistic computations on conventional machines before mapping them to stochastic hardware.
Criticism
Executable semantics do not establish purpose-built hardware, integration, scale, external access, recurring use, or comparative advantage.
Sources
THRML repository

Physical — demonstrated

Claim
The matrix credits thermodynamic and probabilistic computing at the Physical stage: reported stochastic CMOS silicon.
Evidence
Extropic reports X0 stochastic silicon, while Normal Computing reports a CN101 ASIC tape-out. The full mark relies on the disclosed existence of physical silicon, not on vendor performance projections.
Criticism
A physical realization does not by itself establish system integration, efficient scaling, external access, recurring use, or comparative advantage.
Sources
Extropic hardware; Normal Computing ASICs

Integrated — demonstrated

Claim
The matrix credits thermodynamic and probabilistic computing at the Integrated stage: prototype sampling platforms.
Evidence
Extropic describes XTR-0 as a prototype platform around its thermodynamic sampling hardware. A platform that joins stochastic compute elements with control and readout crosses the integration threshold, although independent characterization remains sparse.
Criticism
A coherent system does not by itself establish efficient scaling, external access, recurring use, or comparative advantage.
Sources
Extropic hardware

Access — limited

Claim
The matrix credits thermodynamic and probabilistic computing only partially at the Access stage: open software and restricted hardware access.
Evidence
The reference software is public, and hardware access is described through early or partner programs rather than an unrestricted public system. The cell is therefore partial.
Criticism
The mark is limited on the current public record: open software and restricted hardware access. Access does not establish broad availability, recurring use, workload generality, or comparative advantage.
Sources
Extropic hardware; THRML repository

Stages not credited

No public evidence yet shows large, composable arrays running externally selected workloads, or a recurring production deployment with independently reproduced advantage.

Sources