Spiking Neural Network Core Partitioning for Noise Resilience

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing automatic signal recognition systems face inefficiencies in power consumption, latency, and noise resilience due to the sequential processing nature of microprocessors and the limitations of digital signal processors, especially in power-constrained devices, and spiking neural networks face distortions from micro-electronic circuit variations and noise in large arrays.

Innovation Solution

A spiking neural network architecture is implemented using a core-based approach with programmable interconnect structures and configurable arrays of spiking neurons and synapses, partitioning large networks into smaller sub-networks on multiple cores to mitigate noise and variability, and employing heterogeneous learning rules and communication protocols for improved resilience and efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional neural networks are used, then high accuracy in clean environments can be achieved, but performance degrades significantly in noisy or adversarial environments

Engineering Contradiction:
Improveperformance in noisy environmentsVSAvoidnetwork structure complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The neural network is segmented into multiple specialized sub-networks (clean network, noisy network, adversarial network), each optimized for specific input conditions. This segmentation allows the system to maintain high reliability across different environments without requiring a single overly complex network to handle all cases simultaneously.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system dynamically selects which sub-network to use based on the characteristics of the input data. A determination module assesses whether input images are clean, noisy, or adversarial, and routes them to the appropriate network. This dynamic adaptation enables reliable performance without the complexity of a static all-purpose network.

Inventive Principle:
Principle #15Dynamics

2Reliability

If multiple specialized networks are used to handle different environments, then reliability improves, but system complexity increases

Engineering Contradiction:
Improveperformance consistency across environmentsVSAvoidnumber of networks
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The determination module serves as a universal controller that manages multiple specialized networks. It automatically identifies input characteristics and routes to the appropriate network, making the overall system multi-functional without requiring manual intervention. This universal control mechanism handles the complexity of coordinating multiple networks while maintaining reliability across different environments.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The determination module acts as an intermediary between the input data and the multiple specialized networks. It assesses input characteristics and mediates the selection process, preventing the need for direct complex interactions between multiple networks. This intermediary simplifies the system architecture while maintaining the benefits of having multiple specialized networks.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If a single robust network is designed to handle all conditions, then versatility improves, but training difficulty and computational cost increase

Engineering Contradiction:
Improvehandling multiple input typesVSAvoidtraining computational cost
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

Instead of training a single network to handle all conditions, the system segments the training task across multiple specialized networks. Each network is trained on specific data types (clean, noisy, or adversarial), reducing the computational burden on each individual training process while maintaining overall versatility through the combination of specialized networks.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3881246B1Resilient neural network
Publication Date: 2026.05.20 INNATERA NANOSYSTEMS BV
  • EP3881246B1 patent drawingFigure 1
  • EP3881246B1 patent drawingFigure 2
  • EP3881246B1 patent drawingFigure 3

AI summary

The present invention discloses a spiking neural network for classifying input signals. The spiking neural network comprises a plurality of spiking neurons, and a plurality of synaptic elements interconnecting the spiking neurons to form the network. Each synaptic element is adapted to receive a synaptic input signal and apply a weight to the synaptic input signal to generate a synaptic output signal, the synaptic elements being configurable to adjust the weight applied by each synaptic element. Furthermore, each of the spiking neurons is adapted to receive one or more of the synaptic output signals from one or more of the synaptic elements, and generate a spatio-temporal spike train output signal in response to the received one or more synaptic output signals. The spiking neural network is partitioned into multiple sub-networks, wherein each sub-network comprises a sub-set of the spiking neurons connected to receive synaptic output signals from a sub-set of the synaptic elements. The sub-network is adapted to generate a sub-network output pattern signal in response to a sub-network input pattern signal applied to the sub-network. Furthermore, each sub-network forms part of one or multiple cores in an array of cores, each core consisting of a programmable network of spiking neurons implemented in hardware or a combination of hardware and software. Communication between cores in the core array is arranged through a programmable interconnect structure.