Energy-Efficient Brain-Inspired Self-Attention-Spiking Neural Network Framework for Mix-Type Wafer Defect Recognition

In semiconductor manufacturing, wafer defect recognition plays a critical role. As the technology advances and wafer feature sizes decrease, defect detection has become more challenging, particularly for mix-type defects. Although artificial neural network (ANN) models are currently used for this task, the high-precision floating-point operations required by ANN models put significant pressure on edge computing devices in the industry. Additionally, ANN models have poor cognitive abilities and high data dependencies. Spiking neural networks (SNNs), as an excellent brain-inspired computational model, have the potential to address these problems. In this article, we present a novel spiking-self-attention network (Spiky-SANet), which emulates the transmission of membrane potential signals in the brain using SNNs. This approach significantly reduces energy consumption during the running process, enabling efficient deployment on edge computing devices. Moreover, we introduce a multitime-step self-attention (MTS-SA) module that emulates the interactive behavior between visual neurons and brain neurons, thereby enhancing the model’s cognitive ability. Additionally, we tackle the issue of imbalanced focus on positive and negative samples by employing an asymmetric loss (ASL) function, effectively mitigating the influence of powerful negative samples. Our framework represents the pioneering attempt of SNNs in the field of wafer defect detection, achieving state-of-the-art performance and paving the way for more efficient and energy-saving semiconductor manufacturing processes.