Neuron Output Level Adjustment in Memristor Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multilayer neural networks using memristors and analog processing circuits face saturation issues with output values, leading to inaccurate recognition, prediction, and motion control due to high output values exceeding the write threshold voltage of memristors, causing conductance changes that hinder weight assignment.
Innovation Solution
A method to adjust the output level of neurons in multilayer neural networks by using memristors and analog processing circuits, where scaling factors are calculated to ensure output values remain below the write threshold voltage, preventing saturation and conductance changes, thereby maintaining accurate recognition and prediction performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If output values are allowed to increase freely in multilayer neural networks, then the network can represent more complex patterns and improve recognition accuracy, but the output values exceed the write threshold voltage of memristors causing saturation and conductance changes that degrade performance
Solution Approach 1:
The patent introduces a scaling factor that dynamically adjusts the output values of neurons to keep them within the permissible range below the write threshold voltage. This parameter change prevents saturation and unwanted conductance changes in memristors while maintaining the network's ability to represent complex patterns through adjusted weight values.
Solution Approach 2:
The patent implements a feedback mechanism where the output values are monitored and scaled down if they approach the write threshold voltage. The scaling factor is calculated based on the maximum output value and the permissible range, creating a closed-loop control system that maintains output stability and prevents harmful effects on memristor conductance.
2Adaptability or versatility
If the output range of neurons is expanded to improve processing capability, then the neural network can handle larger dynamic ranges, but the analog processing circuit cannot properly process values outside its maximum output range, leading to information loss
Solution Approach 1:
The patent applies scaling factors to neuron output values in advance, before the values are processed by the analog processing circuit. This preliminary action ensures that all output values fall within the permissible range of the analog circuit, preventing clipping and information loss while maintaining the full dynamic range capability through adjusted weight representations.
3Speed
If memristors are used to implement synaptic weights for high-speed analog processing, then the neural network achieves fast computation, but the memristor conductance changes due to excessive output values cause inaccurate weight assignment
Solution Approach 1:
The patent compensates for the conductance changes in memristors by introducing scaling factors that adjust the output values. This parameter change ensures that the effective weight assignment remains accurate despite the physical changes in memristor conductance, maintaining both the high-speed analog processing capability and the precision of weight representation.
Data Source
AI summary
A method for adjusting output level of a neuron in a multilayer neural network is provided. The multilayer neural network includes a memristor and an analog processing circuit, causing transmission of the signals between the neurons and the signal processing in the neurons to be performed in an analog region. The method includes an adjustment step that adjusts an output level of the neurons of each of the layers, causing the output value to become lower than a write threshold voltage of the memristor and to fall within a maximum output range set for the analog processing circuit executing the generation of the output value in accordance with the activation function when each of the output values of the neurons of each of the layers becomes highest.