Thermodynamic SoftMax Gadget Using Engineered Oscillator Potentials

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine learning algorithms using classical computing devices face challenges with increased execution time and energy consumption due to complex statistical probability calculations, and thermodynamic computing systems require information conversion to classical form, leading to potential measurement errors and reduced efficiency.

Innovation Solution

Implementing a SoftMax function using a thermodynamic chip with an analog SoftMax gadget that maintains information in a thermodynamic state, utilizing oscillators configured with an engineered potential to evolve to a one-hot encoded vector, and employing relay oscillators to relay thermodynamic information directly between energy-based models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If statistical probability calculations are performed using classical computing devices, then algorithms can be executed, but execution time and energy consumption increase significantly

Engineering Contradiction:

Improveexecution speedVSAvoidenergy consumption

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent replaces classical digital computing systems with a thermodynamic computing system that uses physical oscillators to perform computations. The SoftMax function is implemented through the natural thermodynamic evolution of oscillators configured with an engineered potential, substituting mechanical/digital calculation processes with physical thermodynamic processes that inherently compute statistical probabilities through their dynamic behavior.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental operating parameters from digital bit states to continuous thermodynamic states of oscillators. By configuring oscillators with specific potentials and allowing them to evolve thermally, the system computes SoftMax probabilities through physical parameter evolution rather than iterative digital calculation, achieving faster execution and lower energy consumption.

Inventive Principle:

Principle #35Parameter changes

2Productivity

If thermodynamic computing is used, then computation speed improves, but information conversion to classical form introduces measurement errors and reduces efficiency

Engineering Contradiction:

Improvecomputation speedVSAvoidinformation accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent introduces relay oscillators as intermediary elements that maintain thermodynamic information between computational stages. These relay oscillators receive thermodynamic states from one set of oscillators and transfer them to subsequent oscillators without requiring conversion to classical digital form, thus preserving measurement precision while enabling continued thermodynamic computation.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent maintains continuous thermodynamic information flow through the computational system. By using relay oscillators to directly transfer thermodynamic states between computational stages, the system eliminates interruptions for measurement and conversion, maintaining the continuity of useful thermodynamic action and preventing information loss or measurement errors.

Inventive Principle:

Principle #20Continuity of useful action

3Adaptability or versatility

If information is converted to classical computing device form, then further processing can be performed, but conversion delays and potential measurement errors occur

Engineering Contradiction:

Improveprocessing capabilityVSAvoidconversion delay

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent replaces the conversion process from thermodynamic to classical form with continued thermodynamic processing. By configuring subsequent oscillators to accept and process thermodynamic states directly, the system eliminates the conversion step entirely, maintaining adaptability for further processing without incurring conversion delays or measurement errors.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

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

This approach accelerates computations and improves energy efficiency by performing SoftMax functions directly in a thermodynamic state, reducing measurement errors and delays associated with information conversion.

Implementation Method 1

the oscillators of the analog SoftMax gadget thermodynamically evolve to have values that encode a one-hot vector

Methodology Applied

Scientific EffectThermal evolution:

Implementation Method 2

employing relay oscillators to relay thermodynamic information directly between energy-based models

Methodology Applied

Scientific EffectThermodynamic information transfer:

Data Source

PatentUS20250284947A1Thermodynamic computing softmax gadget

Publication Date: 2025.09.11 EXTROPIC CORP

AI summary

An analog SoftMax gadget is implemented using one or more thermodynamic chips (neuro-thermodynamic processors). The analog SoftMax gadget takes a thermodynamic input and calculates a result of the SoftMax function thermodynamically according to an engineered potential used for oscillators and oscillator couplings for a set of oscillators that implement the analog SoftMax gadget. The analog SoftMax gadget returns the result of the SoftMax function as a thermodynamic output that may be relayed to other energy-based models of a thermodynamic computer. The input, processing, and output are all performed thermodynamically (e.g., in an analog fashion) without a need to convert the information into a classical representation.