Neural Network Memory Fault Mitigation via Bit Replacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks face significant accuracy issues in image recognition due to faults or damages in memory storage, particularly in higher bits, leading to low success rates in image recognition tasks.
Innovation Solution
A data processing circuit and fault mitigating method that determine the repair condition based on the difference between the output of computing layers and correct values, replacing fault bits with non-fault bits from corresponding positions in memory to improve recognition success rates across various neural network models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If memory is used to store neural network data, then data storage capacity is provided, but fault/damage in memory blocks causes loss of data completeness and accuracy
Solution Approach 1:
The patent performs fault simulation and identifies fault bits before actual neural network execution. By pre-determining which bits are faulty and establishing repair conditions in advance, the system prepares compensation strategies beforehand, ensuring data accuracy is maintained during runtime without requiring real-time fault detection.
Solution Approach 2:
The patent changes the parameter of bit value by replacing fault bit values with corresponding non-fault bit values based on determined repair conditions. This parameter substitution restores data accuracy by correcting erroneous bit values while preserving the overall data structure and storage capacity.
2Measurement precision
If fault bits are replaced with non-fault bits, then data accuracy is improved, but additional processing steps are required
Solution Approach 1:
The repair conditions mapping between fault bits and non-fault bits is determined in advance through fault simulation, before actual data processing. This preliminary preparation stores the replacement strategy in a lookup table or mapping structure, so that during runtime, the system only needs to perform simple value substitution based on pre-computed repair conditions, minimizing additional processing complexity.
Data Source
AI summary
A data processing circuit and a fault mitigating method are provided. The method is adapted for a memory having at least one fault bit. The memory provides a block for data storage. A difference between an output of a value of a plurality of bits input to at least one computing layer in a neural network and a correct value is determined. The bits are respectively considered the at least one fault bit. A repair condition is determined based on the difference. The repair condition includes a correspondence between a position where the fault bit is located in the block and at least one non-fault bit in the memory. A value of at least one non-fault bit of the memory replaces a value of the fault bit based on the repair condition.