Abstract:Mixed-signal implementations of SNNs offer a promising solution to edge computing applications that require low-power and compact embedded processing systems. However, device mismatch in the analog circuits of these neuromorphic processors poses a significant challenge to the deployment of robust processing in these systems. Here we introduce a novel architectural solution inspired by biological development to address this issue. Specifically we propose to implement architectures that incorporate network motifs found in developed brains through a differentiable re-parameterization of weight matrices based on gene expression patterns and genetic rules. Thanks to the gradient descent optimization compatibility of the method proposed, we can apply the robustness of biological neural development to neuromorphic computing. To validate this approach we benchmark it using the Yin-Yang classification dataset, and compare its performance with that of standard multilayer perceptrons trained with state-of-the-art hardware-aware training method. Our results demonstrate that the proposed method mitigates mismatch-induced noise without requiring precise device mismatch measurements, effectively outperforming alternative hardware-aware techniques proposed in the literature, and providing a more general solution for improving the robustness of SNNs in neuromorphic hardware.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in mixed - signal neuromorphic chips, the performance degradation caused by device mismatch in analog circuits. Specifically, device mismatch will introduce noise, affecting the accuracy of spiking neural networks (SNNs) built on these processors, making them less accurate than traditional artificial neural networks (ANNs) based on digital computing techniques. In order to enable these neuromorphic systems to be deployed in practical applications, a method must be found to reduce the performance loss caused by mismatch while improving the robustness of model training and adaptability to tasks. ### Overview of the Solution To solve the above - mentioned problems, the author proposes a new architecture solution inspired by biological development. This method optimizes the network topology by re - parameterizing the weight matrix and introducing gene expression patterns and genetic rules into SNNs. Specifically: 1. **Introducing Network Motifs**: - The author draws on network motifs formed during brain development in biology. These motifs can enhance the robustness of neural networks to noise. - The probability of network connections is defined by gene expression patterns and genetic rules, so that the generated network has robustness similar to that of biological neural networks. 2. **Re - parameterizing the Weight Matrix**: - Use the formula \( B = H(XO X^T) \) to re - parameterize the weight matrix, where \( X\in\{0, 1\}^{N\times G} \) represents the gene expression pattern, \( O\in[0, 1]^{G\times G} \) represents the genetic rules, and \( H \) is the Heaviside function. - This re - parameterization method ensures that the weight matrix can be optimized by gradient descent, so that the network has better robustness and adaptability. 3. **Experimental Verification**: - The author uses the Yin - Yang classification data set to verify this method and compares it with existing hardware - aware training methods. - The experimental results show that the proposed method is significantly superior to existing methods in classification accuracy and can effectively mitigate the noise caused by mismatch without accurately measuring device mismatch. ### Key Contributions - A new architecture solution is proposed to improve the robustness of neuromorphic systems by introducing biologically - inspired network motifs. - This method does not require accurate measurement of device mismatch, thus simplifying the hardware design and training process. - The experimental results show that this method performs better than existing methods in classification tasks, especially in the presence of device mismatch. In conclusion, by introducing biologically - inspired network motifs, this paper provides a general and effective solution to deal with the device mismatch problem in mixed - signal neuromorphic chips, thereby improving the robustness and performance of neuromorphic computing systems.

Genetic Motifs as a Blueprint for Mismatch-Tolerant Neuromorphic Computing

Reliable Memristor-based Neuromorphic Design Using Variation- and Defect-Aware Training

Robustness of Neuromorphic Computing with RRAM-based Crossbars and Optical Neural Networks.

Exploiting Inherent Error-Resiliency of Neuromorphic Computing to achieve Extreme Energy-Efficiency through Mixed-Signal Neurons

Supervised training of spiking neural networks for robust deployment on mixed-signal neuromorphic processors

NEUTRAMS: Neural Network Transformation and Co-Design under Neuromorphic Hardware Constraints

Hybrid Dual-Complementary Metal-Oxide-semiconductor/memristor Synapse-Based Neural Network with Its Applications in Image Super-Resolution

Neuromorphic Extreme Learning Machines with Bimodal Memristive Synapses

An error-propagation spiking neural network compatible with neuromorphic processors

Design Technology Co-Optimization for Neuromorphic Computing

Data and Power Efficient Intelligence with Neuromorphic Learning Machines

Compiling Spiking Neural Networks to Neuromorphic Hardware

Hardware-friendly Neural Network Architecture for Neuromorphic Computing

End-to-End Implementation of Various Hybrid Neural Networks on a Cross-Paradigm Neuromorphic Chip

Improving energy efficiency and classification accuracy of neuromorphic chips by learning binary synaptic crossbars

Biologically Realistic Computational Primitives of Neocortex Implemented on Neuromorphic Hardware Improve Vision Transformer Performance

Bottom-Up and Top-Down Approaches for the Design of Neuromorphic Processing Systems: Tradeoffs and Synergies Between Natural and Artificial Intelligence

In-Hardware Learning of Multilayer Spiking Neural Networks on a Neuromorphic Processor

Neuromorphic analog circuits for robust on-chip always-on learning in spiking neural networks

A design of magnetic tunnel junctions for the deployment of neuromorphic hardware for edge computing

MorphIC: A 65-nm 738k-Synapse/mm$^2$ Quad-Core Binary-Weight Digital Neuromorphic Processor with Stochastic Spike-Driven Online Learning