Spiking Neuron Firing Timing via Monotonic Time Segments

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The calculation load of the firing timing of a spiking neuron in spiking neural networks is high, which poses a challenge in efficient information processing.

Innovation Solution

A neural network device and method that narrow down the time segments for calculating firing timings of spiking neurons by representing membrane potential as a monotonic function of time and comparing it with a threshold, reducing the calculation load through efficient identification of time segments containing the firing timing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the membrane potential is calculated continuously to determine firing timing, then the accuracy of firing detection is improved, but the calculation load increases

Engineering Contradiction:

Improvefiring timing accuracyVSAvoidcalculation load

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent divides the continuous time period into discrete time segments based on spike arrival times. Instead of calculating membrane potential continuously, the system only evaluates potential at specific time points (when spikes arrive). This segmentation reduces the calculation load while maintaining accurate firing detection, as the membrane potential is monotonic within each segment and the threshold crossing can be determined by comparing potential at segment boundaries.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary sorting of spike arrival times and pre-determines the time segments before calculating membrane potential. By organizing the spike arrival times in advance and establishing the time segments beforehand, the system avoids redundant calculations and only computes membrane potential where necessary, reducing overall computational complexity while maintaining precision.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If the threshold for membrane potential is lowered to detect firing more sensitively, then the detection sensitivity is improved, but the stability of the neuron decreases

Engineering Contradiction:

Improvefiring detection sensitivityVSAvoidneuron stability

Core Design Contradiction:

Measurement precisionVSStability of the object's composition

Solution Approach 1:

The patent dynamically adjusts the threshold based on the membrane potential's monotonic behavior within each time segment. Instead of using a fixed low threshold that causes instability, the system adapts the threshold evaluation to the dynamic characteristics of the membrane potential at different times. The threshold is effectively adjusted according to the local dynamics of potential change, maintaining sensitivity while preserving stability.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent changes the parameter of threshold evaluation from a static value to a dynamic comparison based on monotonic function properties. By utilizing the monotonic characteristic of membrane potential within time segments, the system changes how the threshold is applied - comparing potential at different time points within segments rather than using a fixed threshold value, thereby maintaining both sensitivity and stability.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12572784B2Neural network device, firing timing calculation method, and recording medium

Publication Date: 2026.03.10 NEC CORP

AI summary

A neural network device narrows down, from among respective time segments from when a spike is received to when a next spike is received, a candidate of a time segment including a firing timing of a spiking neuron, a membrane potential of the spiking neuron during a period from when a spike is received to when a next spike is received being represented by a monotonic function of time, a firing condition of the spiking neuron being represented by a comparison between the membrane potential and a threshold.