Probabilistic CMOS Circuits for Fast Low-Energy Graph Sampling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current integrated circuits (ICs) face challenges in efficiently generating random walks on graphs and sampling from certain distributions, particularly in machine learning applications, due to limitations in energy consumption and time efficiency.

Innovation Solution

Configurable CMOS-based circuits are designed with probabilistic modules and metastable circuits to generate random walks on graphs and sample from Softmax distributions, utilizing probabilistic circuit modules and metastable states to control transition probabilities and voltages, enabling efficient generation of samples from arbitrary graphs and distributions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional integrated circuits are used to generate random walks and sample from distributions, then basic computational functions can be performed, but energy consumption is high and time efficiency is low

Engineering Contradiction:
Improvetime efficiencyVSAvoidenergy consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent replaces conventional deterministic digital logic circuits with probabilistic circuits that utilize metastable states and thermodynamic fluctuations. This substitution allows the circuit to naturally perform random sampling and random walk generation through physical processes rather than computational algorithms, significantly improving time efficiency while reducing energy consumption.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the operating parameters of CMOS circuits by biasing them in the metastable region where thermal fluctuations dominate. By adjusting bias voltages and operating points to exploit thermodynamic effects, the circuit transitions from deterministic switching to probabilistic behavior, enabling efficient sampling from complex distributions without high energy consumption.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If probabilistic circuit modules are implemented to generate random walks, then sampling efficiency improves, but circuit complexity increases

Engineering Contradiction:
Improvesampling efficiencyVSAvoidcircuit complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent designs universal probabilistic circuit modules that can perform multiple functions including random walk generation, sampling from Softmax distributions, and implementing Markov chains. These multi-functional modules reduce overall system complexity by eliminating the need for separate dedicated circuits for each probabilistic operation.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent segments complex probabilistic computations into modular circuit blocks, each handling specific probabilistic operations. These modular units can be independently designed, tested, and configured, making the overall complex system manageable while maintaining high sampling efficiency through specialized hardware acceleration.

Inventive Principle:
Principle #1Segmentation

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

The proposed circuits enable faster and more energy-efficient generation of random walks and sampling from Softmax distributions, enhancing applications in machine learning tasks such as model selection, Bayesian inference, discrete optimization, neural network architecture search, and attention mechanisms.

Implementation Method 1

metastable circuits to generate random walks on graphs and sample from Softmax distributions, utilizing probabilistic circuit modules and metastable states to control transition probabilities and voltages

Methodology Applied
Scientific EffectMetastability: Metastability

Data Source

PatentEP4718728A1Configuring circuits for generating samples based on predetermined functions
Publication Date: 2026.04.01 EXTROPIC CORP
  • EP4718728A1 patent drawingFigure 1
  • EP4718728A1 patent drawingFigure 2A~2C
  • EP4718728A1 patent drawingFigure 2D~2G

AI summary

A method comprises: arranging a configurable circuit module (102) comprising a plurality of input nodes (104A-104N), a plurality of output nodes (106A-106N), and a plurality of probabilistic circuit modules (112A-112N) defining a different respective mapping from each input node to a different respective output node, wherein each probabilistic circuit module of the plurality of probabilistic circuit modules comprises a first input (114B-114N) configured to receive a bias voltage; receiving a function; determining a respective bias voltage for each probabilistic circuit module of the plurality of probabilistic circuit modules based at least in part on the function; providing a respective voltage to each input node of the plurality of input nodes; and generating, by the configurable circuit module, a sample from a probability distribution at the plurality of output nodes, based at least on the respective bias voltages and the respective voltages provided to each input node.