Neuromorphic Spiking Neural Network Architecture for Feature Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computing paradigms, such as von Neumann architecture, are inefficient in processing large volumes of data due to power and space inefficiencies, and struggle with learning and feature extraction, limiting their ability to handle Big Data effectively.
Innovation Solution
A neuromorphic computing system with spiking neurons and a synaptic competition mechanism that uses spike-based learning to adjust synaptic weights, enabling feature learning and efficient data processing through potentiation and depression functions, inspired by biological principles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If conventional von Neumann architecture is used for data processing, then computing power can be increased by adding more cores, but power consumption and space requirements increase significantly
Solution Approach 1:
The patent replaces conventional CMOS logic-based von Neumann architecture with a neuromorphic spiking neural network architecture that mimics biological neural computation. This substitution fundamentally changes the computing paradigm from sequential instruction execution to event-driven spike propagation, achieving higher computational efficiency with lower power consumption. The spiking neurons and synapses are implemented using hardware circuits that consume significantly less power than traditional multi-core processors for equivalent computational tasks.
Solution Approach 2:
The patent changes the fundamental parameters of computation by using sparse, event-driven spike signals instead of continuous data streams, and by implementing learning through spike-timing-dependent plasticity (STDP) mechanisms. These parameter changes enable the system to achieve comparable or superior computational performance with dramatically reduced power consumption and area requirements compared to conventional architectures.
2Speed
If conventional von Neumann architecture is used for data processing, then computing speed can be increased, but the physical separation of memory and CPU limits throughput
Solution Approach 1:
The patent merges the functions of memory and computation within the same hardware structure by implementing synapses that inherently store weights and perform multiplicative operations during spike propagation. This eliminates the need for separate memory access operations, achieving high throughput without the von Neumann bottleneck. The spiking neural network hardware integrates storage and processing in a unified architecture that naturally handles parallel computations.
3Measurement precision
If template learning is used in spiking neural networks, then exact pattern memories can be stored, but accuracy improvements diminish with increasing number of neurons
Solution Approach 1:
The patent extracts and emphasizes the feature learning capability from conventional artificial neural networks and adapts it to spiking neural networks through the overflow neuron mechanism. Instead of storing complete patterns, the system learns to detect and respond to salient features through competitive synaptic mechanisms and selective neuron activation, achieving high accuracy with fewer neurons by focusing computational resources on discriminative features rather than memorizing entire patterns.
4Adaptability or versatility
If feature learning is implemented in spiking neural networks using conventional methods, then features can be extracted, but complex feedback mechanisms are required
Solution Approach 1:
The patent implements self-organizing feature learning through competitive synaptic mechanisms where synapses automatically compete for input ownership based on spike timing and strength. The overflow neuron mechanism enables automatic feature detection and neuron specialization without external feedback control, allowing the network to autonomously develop feature detectors through local competitive interactions. This eliminates the need for complex global feedback mechanisms while maintaining feature learning capability.
Data Source
AI summary
A neuromorphic architecture for a spiking neural network comprising a plurality of spiking neurons, each with a plurality of synapses and corresponding synaptic weights, the architecture further comprising a synaptic competition mechanism in connection with a spike-based learning mechanism based on spikes perceived behind a synapse, in which architecture synapses of different neurons connected to the same input compete for that input and based on the result of that competition, each neuron of the neural network develops an individual perception of the presented input spikes, the perception used by the learning mechanism to adjust the synaptic weights.