Enhancing Hardware Neural Networks with Self-Healing Perceptron Design

Neural Network (NN) is an essential element in the success of many AI applications. Its widespread application across multiple fields highlights the ongoing research problem of ensuring and protecting its performance against potential faults. This paper explores the concept of self-healing in NNs, empowering systems to detect and recover from faults. The focus is on a novel self-healing approach for hardware neural networks, employing a shared mechanism and a spare layer to address faulty perceptron nodes. The fault detection method centers on Stuck-at-fault detection, crucial for identifying and subsequently recovering faults within the network. The proposed self-healing perceptron operates in two modes: healing and regular, facilitating fault recovery while minimizing area overhead. The area overhead stands at 17% with a 4-layer configuration, showcasing a decreasing trend as more hidden layers are added. Interestingly, this overhead remains unaffected by the number of neurons within each layer. The proposed method is implemented using VHDL and the simulation obtained using Xilinx Virtex-7 FPGA showcases promising results, demonstrating reduced area overhead with increased network complexity. Reliability analysis illustrates the proposed method's effectiveness in ensuring seamless functionality over time compared to traditional approaches.