Spiking Neural Network Neuron Model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of constructing spiking neural networks using analog circuits makes it challenging to simplify the configuration of each neuron model.
Innovation Solution
A computing device with a spiking neural network that includes an accumulation phase for adding currents and a decoding phase for converting voltage to pulse timing, where the current flowing into or out of a neuron during the accumulation phase depends on its membrane potential.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If analog circuits are used to implement spiking neural networks with weighted sum operations and activation functions, then computational efficiency is enhanced, but the configuration complexity of each neuron model increases
Solution Approach 1:
The neuron model configuration is segmented into two distinct phases: an accumulation phase for computing weighted sums of inputs, and a decoding phase for applying activation functions and generating spikes. This segmentation allows each phase to be implemented with simpler, dedicated circuitry rather than requiring complex general-purpose analog computation circuits, thereby reducing overall configuration complexity while maintaining computational efficiency.
Solution Approach 2:
The activation function computation is extracted from the continuous analog computation process and implemented as a separate decoding phase operation. By taking out the activation function application from the main accumulation process and handling it in a distinct phase with specialized circuits, the neuron model configuration becomes simpler while still achieving the necessary computational functionality.
2Adaptability or versatility
If the scale of the neural network increases to handle more complex training subjects, then the network's processing capability improves, but the configuration complexity of spiking neural networks implemented with analog circuits increases
Solution Approach 1:
By segmenting the neuron operation into accumulation and decoding phases, the patent enables scalable network architecture where each neuron follows the same simplified two-phase pattern. This uniform segmentation across all neurons allows complex networks to be built by replicating the same basic building block, improving adaptability without proportionally increasing configuration complexity.
Solution Approach 2:
The patent introduces dynamic phase switching within each neuron, alternating between accumulation and decoding phases. This dynamic operation allows the same simplified circuit configuration to handle complex processing tasks by changing its operational state over time, thereby improving processing capability without requiring proportionally more complex static circuit configurations.
Data Source
AI summary
A computing device includes a spiking neural network that includes an accumulation phase that adds currents and a decoding phase that converts a voltage resulting from the addition to a voltage pulse timing, the spiking neural network comprising a current adding portion wherein the current that flows into or out of an own neuron in the accumulation phase depends on the membrane potential of that neuron.