Spiking Neural Network Circuit With Selective Spike Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As the number of neurons and synapses in spiking neural networks increases, so does power consumption due to the increased number of synapses receiving spike signals, leading to inefficiencies in signal transfer and processing.
Innovation Solution
A spiking neural network circuit with a hierarchical transfer structure of spike signals, utilizing a tree structure with driving buffers and OR gates to selectively transfer input spike signals to synapse zones based on synapse weights, reducing unnecessary signal transmission and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the number of synapses receiving spike signals is increased to enhance neural network capability, then the processing power and neural network performance are improved, but the power consumption increases significantly
Solution Approach 1:
The patent divides the synapse population into multiple groups or zones, each receiving spike signals selectively through dedicated buffer circuits. This segmentation allows the system to activate only the necessary synapse groups for a given computation, reducing overall power consumption while maintaining the capability to process complex neural network operations across multiple segments.
Solution Approach 2:
The patent implements dynamic control of spike signal transmission through buffer circuits that can be selectively activated or deactivated. This dynamic switching mechanism allows the system to adaptively enable only the synapses required for current computations, preventing unnecessary power consumption from static or inactive connections while preserving the full neural network capacity when needed.
2Device complexity
If spike signals are transferred to all synapses regardless of weight, then the signal transfer is simplified, but unnecessary signal transmission increases power consumption
Solution Approach 1:
The patent introduces buffer circuits as intermediary components between the spike signal source and the synapse zones. These buffers act as mediators that conditionally forward signals based on weight information, adding a layer of control that reduces unnecessary signal transmission to zero-weight or low-priority synapses while maintaining the overall simplicity of the signal transfer architecture.
Solution Approach 2:
The patent applies different signal transfer characteristics to different synapse zones based on their weight values. High-weight synapses receive full spike signals through activated buffers, while low-weight or zero-weight synapses receive attenuated or blocked signals. This local differentiation optimizes power consumption by matching signal transmission intensity to the actual importance of each connection.
Data Source
AI summary
Disclosed is a spiking neural network circuit, which includes an axon circuit that generates an input spike signal, a first synapse zone and a second synapse zone each including one or more synapses, wherein each of the synapses is configured to perform an operation based on the input spike signal and each weight, and a neuron circuit that generates an output spike signal based on operation results of the synapses. The input spike signal is transferred to the first synapse zone and the second synapse zone through a tree structure, and each of branch nodes of the tree structure includes a driving buffer.