Non-Linear Current Attenuation in Neural Networks for Synapse Drift

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing neural networks face challenges with resistance drift and dispersion in multi-level phase change memory cells used for synapses, leading to implementation issues.

Innovation Solution

Incorporating a non-linear current attenuator between neurons and synapses to compensate for resistance deviations in multi-level phase change memory cells.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If multi-level phase change memory cells are used for synapses to increase storage capacity, then the quantity of information stored is improved, but resistance drift and dispersion occur leading to reduced reliability

Engineering Contradiction:

Improvestorage capacityVSAvoidresistance stability

Core Design Contradiction:

Quantity of substanceVSReliability

Solution Approach 1:

A non-linear current attenuator is introduced as an intermediary component between the synapse (multi-level phase change memory cell) and the neuron. This attenuator compensates for resistance drift and dispersion in the memory cell by applying non-linear current attenuation, thereby maintaining reliable neural network operation despite variations in synapse resistance

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes the electrical parameters (current attenuation characteristics) dynamically to compensate for resistance drift. The non-linear current attenuator adjusts its attenuation factor based on the resistance state of the phase change memory cell, effectively counteracting resistance variations and maintaining stable neural network functionality

Inventive Principle:

Principle #35Parameter changes

2Reliability

If non-linear current attenuator is added to compensate resistance drift, then reliability is improved, but device complexity increases

Engineering Contradiction:

Improveresistance compensationVSAvoidcircuit structure

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The non-linear current attenuator serves as a dedicated intermediary component that handles the complex task of resistance compensation. By isolating this complexity in a single module between the synapse and neuron, the rest of the neural network can operate with simpler, more reliable components

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The non-linear current attenuator automatically adjusts its attenuation characteristics based on the resistance state of the phase change memory cell without requiring external control or calibration. This self-adjusting capability reduces the need for additional control circuitry and simplifies the overall system architecture

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS20250292075A1Non-linear current attenuation in neural networks

Publication Date: 2025.09.18 STMICROELECTRONICS FRANCE

AI summary

A neural network circuit having at least one first neuron coupled to a second neuron via at least one synapse. A non-linear current attenuator receives a first current from the at least one synapse and provides a second current to the second neuron.