Graph Attention-Based Virtual Metrology for Film Deposition Processes in Semiconductor Manufacturing

In advanced semiconductor manufacturing, operating at nanometer and angstrom scales, precise process control necessitates accurate and timely metrology. However, conventional physical metrology faces limitations concerning measurement latency, cost, and sampling constraints, hindering its scalability in high-volume production environments. Virtual metrology (VM) has emerged as a viable alternative, predicting wafer-level characteristics directly from equipment sensor data. Despite recent advancements, many existing VM models are predominantly correlation-driven, often failing to capture structured dependencies among heterogeneous process variables and offering limited interpretability. This research presents a novel graph attention-based VM framework tailored for film deposition processes. The proposed methodology effectively integrates temporal feature learning with structured parameter-layer dependency modeling. It conceptualizes each step-parameter

pair as a distinct node, from which temporal embeddings are extracted using convolutional feature encoders applied to high-frequency equipment traces. Subsequently, a parameter-to-layer graph attention mechanism is utilized to model directional dependencies, enabling each film layer to aggregate pertinent process information. The framework's performance is rigorously evaluated using industrial deposition data collected from production wafers, demonstrating its capability to accurately predict film thickness from multivariate sensor signals. Experimental results consistently show improved predictive performance compared to conventional baseline models, alongside providing enhanced analytical insights into complex process variable relationships, thereby advancing precise and timely process control.

Message: Hi FE, I need help with “Graph Attention-Based Virtual Metrology for Film Deposition Processes in Semiconductor Manufacturing” in “Chemical Engineering”