Wafer2Spike: Spiking Neural Network for Wafer Map Pattern Classification

TL;DR

Wafer map pattern classification is critical for IC yield and defect detection. The paper introduces Wafer2Spike, a domain-specific spiking neural network that encodes wafer maps into spike trains and processes them with spike-based convolutional layers, followed by a non-spiking output, trained via spatio-temporal backpropagation with surrogate gradients. On the WM-811k dataset, Wafer2Spike achieves an average accuracy of $98\%$, surpassing state-of-the-art DNN-based approaches and showing robustness on underrepresented patterns, while offering substantial energy efficiency (up to $22\times$ lower energy than DNNs). This work demonstrates the practical viability of SNNs for wafer-map analytics and provides an open-source release to facilitate broader adoption and neuromorphic hardware exploration.

Abstract

In integrated circuit design, the analysis of wafer map patterns is critical to improve yield and detect manufacturing issues. We develop Wafer2Spike, an architecture for wafer map pattern classification using a spiking neural network (SNN), and demonstrate that a well-trained SNN achieves superior performance compared to deep neural network-based solutions. Wafer2Spike achieves an average classification accuracy of 98\% on the WM-811k wafer benchmark dataset. It is also superior to existing approaches for classifying defect patterns that are underrepresented in the original dataset. Wafer2Spike achieves this improved precision with great computational efficiency.

Figures (3)

• Figure 1: Examples of different wafer map patterns present in the WM-811k dataset (best viewed in color).
• Figure 2: The Wafer2Spike architecture comprising the convolutional spike encoding layer, spike-based convolutional layers, a fully-connected spiking layer, and the non-spiking output layer.
• Figure 3: Convolutional spike encoding operation for the first two time steps.