Neuron Array Circuit for Neuromorphic Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neuromorphic hardware faces inefficiencies in processing and understanding images or sounds for pattern recognition due to structural constraints such as high power consumption and a narrow dynamic range of output, limiting its ability to perform like human recognition.
Innovation Solution
A neuron module circuit device with a processor-configured neuron array that adapts and trains neuron modules to mimic target patterns by updating synaptic weights and operating in various modes, including visible, hidden, and relay modes, to enhance signal transmission and processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If typical neuromorphic processors are used for pattern recognition, then parallel processing capability is improved, but power consumption increases excessively
Solution Approach 1:
The neuromorphic processor is divided into multiple independent neuron modules, each capable of autonomous operation. Each neuron module contains separate functional units (synaptic weight updating unit, spike signal generating unit, signal transmission unit) that can operate independently, enabling parallel processing while reducing overall power consumption through modular architecture
Solution Approach 2:
Different neuron modules can operate in different modes (visible mode, hidden mode, relay mode, block mode) depending on their specific function. This allows power-efficient operation where only necessary modules are fully active, while others operate in lower-power states or are disabled entirely
2Productivity
If typical neuromorphic processors are used for pattern recognition, then parallel processing capability is improved, but dynamic range of output becomes very narrow
Solution Approach 1:
The neuron modules dynamically adjust their operation mode based on processing requirements. The system can switch between visible mode (for output neurons requiring wide dynamic range), hidden mode (for intermediate processing), relay mode (for signal routing), and block mode (for disabling inactive neurons), thereby expanding the effective dynamic range while maintaining parallel processing efficiency
Solution Approach 2:
The synaptic weights and threshold values are dynamically updated during training and operation. The synaptic weight updating unit continuously adjusts connection strengths based on learning algorithms, enabling the system to adapt its output characteristics and expand its dynamic range to match diverse pattern recognition tasks
3Productivity
If neuron modules are trained to mimic target patterns, then pattern recognition efficiency is improved, but computational complexity increases
Solution Approach 1:
The neuron modules perform self-training through autonomous synaptic weight updating based on local spike timing information. Each neuron module independently adjusts its own weights using Hebbian learning rules or similar algorithms, eliminating the need for complex centralized training control and reducing overall computational complexity while achieving effective pattern recognition
Data Source
AI summary
Disclosed are an apparatus and method with neural processing. The operating method includes constructing a neuron array including a plurality of neuron modules, mapping a target pattern to the neuron array, adapting the neuron modules to the target pattern in response to a reception of the target pattern, and training the neuron modules to cause the neuron array to mimic the target pattern.