Neuromorphic Controller Adjusts Discretization Step Size
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The use of neuromorphic elements in neural networks leads to reduced accuracy due to quantization errors resulting from discrete resistance changes, which affect the performance of product-sum operations, particularly in weight storage and updating functions.
Innovation Solution
A controller for an array of neuromorphic elements that adjusts the discretization step size to improve accuracy by using a dynamic range and offset quantity, allowing for precise control of the neuromorphic element characteristics, thereby reducing errors and enhancing identification performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If a neuromorphic element is used for weight storage and updating in a neural network, then low power consumption and high speed processing are achieved, but identification accuracy deteriorates due to quantization errors from discrete resistance changes
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the discretization step size of the neuromorphic element's resistance based on the magnitude of weight values. When weight values are small, a finer discretization step size is used to maintain precision, while for large weight values, a coarser step size is acceptable. This adaptive parameter adjustment resolves the contradiction by optimizing the trade-off between power consumption and identification accuracy across different operating conditions.
Solution Approach 2:
The patent implements dynamics by making the discretization step size variable rather than fixed. The control unit dynamically selects appropriate step sizes based on the current weight values being stored, allowing the system to adapt its precision requirements in real-time. This dynamic approach enables the neuromorphic system to maintain high identification accuracy when needed while preserving the low power consumption benefits of discrete resistance changes.
2Device complexity
If a fixed discretization step size is used in the neuromorphic element, then device complexity is reduced, but quantization errors increase leading to reduced accuracy in product-sum operations
Solution Approach 1:
The patent applies segmentation by dividing the weight storage range into multiple segments, each with its own optimized discretization step size. The control unit segments the weight value range and selects appropriate step sizes for different segments, allowing high precision where needed while maintaining simplicity in control logic through standardized segmentation rules.
Solution Approach 2:
The patent introduces an intermediary control unit that mediates between the weight values and the neuromorphic element's resistance states. This intermediary layer translates continuous weight values into discrete resistance changes with optimized step sizes, adding precision without directly complicating the neuromorphic element itself. The control unit acts as a buffer that manages the complexity while preserving accuracy.
3Measurement precision
If the resolution of the neuromorphic element characteristic is increased to reduce quantization errors, then accuracy of product-sum operations is improved, but the expressiveness of variables is insufficient and convergence time increases
Solution Approach 1:
The patent applies parameter changes by varying the discretization step size according to the weight value magnitude rather than using a fixed fine resolution throughout. This allows the system to achieve sufficient accuracy for product-sum operations in most cases while avoiding the excessive convergence time that would result from uniformly high resolution across all weight values.
Data Source
AI summary
A controller is a controller of an array including a neuromorphic element that multiplies a weight based on a value of a variable characteristic by a signal, and includes a control unit that controls the characteristic of the neuromorphic element by using a discretization step size obtained so that a predetermined condition for reducing an error or a predetermined condition for improving accuracy is satisfied on the basis of a case where a true value of the weight obtained with a higher accuracy than a resolution of the characteristic of the neuromorphic element is used and a case where a discretization step size which is set for the characteristic of the neuromorphic element is used.