Neuromorphic Weight Transfer for Defective Synaptic Cells
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hardware-based spiking neural networks face performance degradation due to defective synaptic elements during inference processes, with current methods lacking a fundamental solution to address this issue effectively.
Innovation Solution
A weight transfer apparatus and method that builds an artificial neural network learning model considering defective synaptic cells, sets corresponding weights to 0, and transfers the rebuilt model to a neuromorphic device, thereby maintaining performance even with defective elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If synaptic elements are used in neuromorphic devices, then computing performance is improved, but defective cells occur during manufacturing
Solution Approach 1:
The system performs preliminary actions by detecting defective synaptic cells before they affect computing performance. The weight transfer apparatus identifies defective cells and adjusts weights in advance, preventing performance degradation from occurring in the first place.
Solution Approach 2:
The system implements feedback mechanisms where the weight transfer apparatus continuously monitors synaptic element performance, detects defects, and automatically adjusts weights accordingly. This closed-loop feedback ensures that defective cells are compensated for, maintaining reliable computing performance.
2Measurement precision
If existing neural network weights are transferred to hardware SNN, then inference accuracy is maintained, but defective synaptic elements cause performance degradation
Solution Approach 1:
The system converts the harmful effect of defective synaptic cells into a beneficial process by using the weight transfer apparatus to detect and compensate for these defects. The defects are identified and corrected through automated weight adjustment, transforming a liability into an opportunity for system optimization.
Solution Approach 2:
The weight transfer apparatus changes the parameter of weight values corresponding to defective synaptic cells. By modifying these weight parameters (setting them to zero or adjusting them), the system compensates for the physical defects in the hardware, maintaining inference accuracy despite the presence of faulty cells.
3Reliability
If redundant array methods are used to address defective elements, then manufacturing complexity increases, but defect tolerance is improved
Solution Approach 1:
Instead of using complex redundant array structures, the system creates a software-based copy or representation of the synaptic weight matrix. The weight transfer apparatus processes this digital copy to identify and compensate for defective cells, avoiding the need for complex hardware redundancy while achieving defect tolerance.
Solution Approach 2:
The weight transfer apparatus acts as an intermediary between the physical synaptic elements and the computing tasks. It mediates the interaction by detecting defective cells and adjusting weights, thereby protecting the system from the complexity of handling defective hardware directly.
Data Source
AI summary
A weight transfer apparatus for a neuromorphic device includes a memory storing a weight transfer program for the neuromorphic device, and a processor configured to execute the weight transfer program. The weight transfer program builds an artificial neural network learning model, transfers a weight of the built artificial neural network learning model to the neuromorphic device, determines whether a synaptic cell included in the neuromorphic device to which the weight is transferred is defective, rebuilds the artificial neural network learning model after the artificial neural network learning model sets a weight corresponding to a defective synaptic cell to 0, and transfers a weight of the rebuilt artificial neural network learning model to the neuromorphic device.