Predicting the Material Removal Rate in Chemical Mechanical Planarization Process: A Hypergraph Neural Network-Based Approach

Abstract

Material removal rate (MRR) plays a critical role in the operation of chemical mechanical planarization (CMP) process in the semiconductor industry. To date, many physics-based and data-driven approaches have been proposed to predict the MRR. Nevertheless, most of the existing methodologies neglect the potential source of its well-organized and underlying equipment structure containing interaction mechanisms among different components. To address its limitation, this paper proposes a novel hypergraph neural network-based approach for predicting the MRR in CMP. Two main scientific contributions are presented in this work: 1) establishing a generic modeling technique to construct the complex equipment knowledge graph with a hypergraph form base on the comprehensive understanding and analysis of equipment structure and mechanism, and 2) proposing a novel prediction method by combining the Recurrent Neural Network based model and the Hypergraph Neural Network to learn the complex data correlation and high-order representation base on the Spatio-temporal equipment hypergraph. To validate the proposed approach, a case study is conducted based on an open-source dataset. The experimental results prove that the proposed model can capture the hidden data correlation effectively. It is also envisioned that the proposed approach has great potentials to be applied in other similar smart manufacturing scenarios.

Privacy‐preserving gradient boosting tree: Vertical federated learning for collaborative bearing fault diagnosis

“…Data-driven modelling approaches are widely used in many disciplines [6][7][8][9][10]. Therefore, data-driven approaches have been effectively applied in predictive maintenance [11,12], especially fault diagnosis [13,14]. Nonetheless, many realworld manufacturing tasks related data only have limited labels and features, causing underfitting and overfitting, and weakening the model's generalisation capacity [15].…”