Continuous Wavelet Transform and Siamese Network-Based Anomaly Detection in Multi-variate Semiconductor Process Time Series
Bappaditya Dey, Daniel Sorensen, Minjin Hwang, Sandip Halder · 2025
Semiconductor manufacturing is an extremely complex and precision-driven process, characterized by thousands of interdependent parameters collected across diverse tools, process steps, and time scales. multi-variate time-series analysis (MTSA) has emerged as a critical methodology for enabling real-time monitoring, fault detection, and predictive maintenance in such environments. However, applying MTSA for anomaly prediction in semiconductor fabrication presents several critical challenges. These include the high dimensionality of sensor data, severe class imbalance due to the rarity of true faults, the presence of noisy and missing measurements, and the non-stationary behavior of production systems driven by dynamic recipe adjustments, tool aging, and maintenance activities. Furthermore, the complex interdependencies between process variables and the delayed emergence of faults across downstream stages significantly complicate both anomaly detection and root-cause-analysis. This paper presents a novel and generic approach for anomaly detection in multi-variate time-series data using machine learning, with a primary focus on semiconductor manufacturing processes. The proposed methodology consists of three main steps, as: a) converting multi-variate time-series (MTS) data into imagebased representations using the Continuous Wavelet Transform (CWT), b) developing a multi-class image classifier by finetuning a pretrained VGG-16 architecture on custom CWT image datasets, and c) constructing a Siamese network composed of two identical sub-networks, each utilizing the fine-tuned VGG-16 as a backbone with shared weights. The network takes pairs of CWT images as input-one serving as a reference or anchor (representing a known-good or non-anomalous process/tool trace), and the other as a query (representing an unknown or potentially anomalous trace). The model then analyzes and compares the embeddings of both inputs to determine whether they belong to the same class at a given time step. Our proposed approach demonstrates high accuracy in identifying anomalies on a real FAB process time-series dataset, offering a promising solution for offline anomaly detection in process and tool trace data. Moreover, the approach is flexible and can be applied in both supervised and semi-supervised settings.