Weighted Event Encoder With Voltage-Based Spike Timing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neuromorphic systems face challenges in efficiently encoding time-series correlations between events, leading to information loss and reduced accuracy in spike neural network (SNN) inference operations.
Innovation Solution
An encoder that generates input spike signals with varying firing periods based on the voltage level of nodes, where the voltage level is adjusted by applied weights and leakage, incorporating a spike generation circuit and discharge mechanism to maintain time-series correlations, thereby enhancing the accuracy of SNN operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If voltage accumulation method is used to generate spike signals, then time-series correlation information is preserved, but circuit complexity increases due to continuous voltage accumulation mechanism
Solution Approach 1:
The encoder is divided into multiple independent nodes, each responsible for accumulating voltage from specific event signals. This segmentation allows parallel processing of different event correlations while maintaining the time-series information through distributed voltage accumulation across multiple nodes.
Solution Approach 2:
Voltage levels serve as an intermediary mechanism between event signals and spike outputs. The continuous voltage accumulation at each node acts as a mediator that integrates temporal information from multiple events, enabling the preservation of time-series correlations without requiring complex direct signal processing circuits.
2Measurement precision
If multiple event signals with different weights are processed, then encoding accuracy improves, but processing time increases due to sequential weight application
Solution Approach 1:
The system employs periodic discharge cycles where accumulated voltages are reset at regular intervals. This periodic action allows the encoder to process multiple weighted events continuously without indefinite voltage buildup, maintaining encoding accuracy while preventing unbounded processing time through structured reset cycles.
Solution Approach 2:
The encoder dynamically adjusts voltage thresholds and discharge timing parameters based on the accumulated voltage levels from weighted events. By changing these parameters adaptively, the system optimizes processing speed while maintaining the precision required for accurate time-series correlation encoding.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The proposed encoder minimizes information loss and improves the accuracy of SNN inference by encoding time-series correlations, resulting in more precise neuromorphic system performance.
Implementation Method 1
a capacitor connected between the first node and a ground voltage
Implementation Method 2
the voltage level of the first node is continuously reduced by charge leakage
Data Source
AI summary
Disclosed is an encoder including event layer outputs first and second event signals, weight layer applies first and second weights to the first and second event signals respectively, and provides the first event signal in which the first weight is applied and the second event signal in which the second weight is applied to first node, and first spike generation circuit generates first input spike signal of which firing period is changed based on voltage level of the first node. The voltage level of the first node is reduced continuously, increases for first voltage corresponding to the first weight in response to the first event signal activated, and increases for second voltage corresponding to the second weight in response to the second event signal activated.