SNN Control Signal Conversion Using State-Machine PWM

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing spiking neural networks (SNNs) are not efficiently utilized for generating control signals due to high switching activity, limiting their energy efficiency and latency benefits in applications like PWM control.

Innovation Solution

Convert SNN spikes into discrete-valued variable length signals using a state machine, such as a flip-flop unit, to generate pulse-width-modulated signals, reducing unnecessary switching and enhancing efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If SNN spikes are directly used to control a load, then the control response is fast with low latency, but the switching activity is high causing high energy consumption

Engineering Contradiction:

Improvecontrol response speedVSAvoidenergy consumption

Core Design Contradiction:

SpeedVSUse of energy by moving object

Solution Approach 1:

A state machine is introduced as an intermediary between the SNN spike generator and the load controller. The state machine receives spikes from the SNN and generates PWM control signals, acting as a mediator that translates neural spikes into efficient control signals. This intermediary reduces the direct switching activity while maintaining the fast response characteristics of SNNs.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The invention changes the parameter representation of control signals from direct spike-based binary switching to PWM duty cycle modulation. By varying the duty cycle parameter of PWM signals instead of using high-frequency binary switching, the system maintains control effectiveness while significantly reducing switching losses and energy consumption.

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If SNNs are used for control tasks, then adaptability and programmability are improved, but the direct use of spike output requires high switching activity

Engineering Contradiction:

Improvecontrol adaptabilityVSAvoidswitching operation complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The state machine serves as an intermediary that simplifies the interface between adaptable SNN controllers and physical loads. It handles the complexity of signal conversion and PWM generation, allowing the SNN to maintain its adaptability while the state machine manages the switching complexity efficiently.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The invention replaces direct mechanical/electrical switching control with a software-based state machine that generates PWM signals. This substitution reduces the complexity of physical switching operations while maintaining the adaptability of the neural network controller.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Loss of energy

If pulse-based communication is used in SNN, then energy efficiency is improved, but the output requires high switching activity for load control

Engineering Contradiction:

Improvecommunication energy lossVSAvoidload control energy consumption

Core Design Contradiction:

Loss of energyVSUse of energy by moving object

Solution Approach 1:

The invention changes the control parameter from spike frequency/direct switching to PWM duty cycle. This parameter transformation allows the system to maintain the energy-efficient pulse-based communication of SNNs while reducing the energy consumption associated with load control switching through efficient PWM modulation.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentEP4693106A1Device and method for generating a control signal

Publication Date: 2026.02.11 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV

Figure 1
Figure 2
Figure 3A~3C

AI summary

A device for providing a control signal based on a task comprises a neuromorphic unit to operate a spiking neural network, SNN, configured for providing an event signal based on the task. The device comprises a state machine configured for providing at least one discrete-valued variable length control signal based on the event signal.