Analog Neural Network Compensation for Memristor Weight Variations

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Analog neural network systems face errors due to variations in weights stored in memristive elements and errors introduced by digital to analog and analog to digital conversion processes, leading to inaccuracies in output signals.

Innovation Solution

Incorporating compensation circuitry that applies a compensating term to input data signals based on measured errors in weight values, using digital circuitry to adjust for gain and offset errors, and employing a crossbar array with programmable compensation units to correct for errors in memristive elements, allowing for accurate dot product calculations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If analog neural network circuitry is used to perform weight multiplication operations, then computational speed and energy efficiency are improved, but manufacturing precision deteriorates due to variations in weight values stored in memristive elements

Engineering Contradiction:

Improvecomputational speedVSAvoidweight value accuracy

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The system measures the actual weight values stored in memristive elements and uses this feedback information to generate compensating terms that correct for deviations from ideal weight values, thereby maintaining computational accuracy despite manufacturing variations

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system changes the parameter of input data by applying compensating terms that adjust the input values based on measured weight errors, transforming the input data to account for manufacturing imprecision in the analog weights

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If digital to analog and analog to digital conversion circuitry is used to interface digital systems with analog neural network circuitry, then adaptability is improved, but measurement precision deteriorates due to conversion errors

Engineering Contradiction:

Improveinterface compatibilityVSAvoidsignal conversion accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system measures conversion errors introduced by DAC and ADC circuitry and uses this feedback to generate compensating terms that correct for these errors, maintaining signal fidelity through the digital-analog interface

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system performs preliminary measurement of conversion errors during a calibration phase and stores compensating terms that are applied to subsequent input data, preventing conversion errors from affecting operational accuracy

Inventive Principle:

Principle #10Preliminary action

3Reliability

If compensation circuitry is added to correct for errors in analog neural network circuitry, then reliability is improved, but device complexity increases

Engineering Contradiction:

Improveoutput accuracyVSAvoidcircuit architecture

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system introduces compensation circuitry as an intermediary component that sits between the digital interface and the analog neural network circuitry, measuring errors and applying corrections without fundamentally redesigning the core analog computation architecture

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11748608B2Analog neural network systems

Publication Date: 2023.09.05 CIRRUS LOGIC INC

AI summary

The present disclosure relates to a neural network system comprising: a data input configured to receive an input data signal and analog neural network circuitry having an input coupled with the data input. The analog neural network circuitry is operative to apply a weight to a signal received at its input to generate a weighted output signal. The neural network system further comprises compensation circuitry configured to apply a compensating term to the input data signal to compensate for error in the analog neural network circuitry.