← Back to the feed
Chips & computingResearch findings

Self-Supervised Attention-Based Virtual Metrology for CMP MRR Prediction

Original published 2026.11.17 09:56 KSTAdded to Blackdeck 2026.11.17 11:29 KST

What has changed?

Introduces Global Regression Pretraining (GREP) to learn temporal patterns from unlabeled sensor signals.

Key points

  1. Proposes GREP, a self-supervised attention-based framework for CMP MRR prediction.
  2. Learns temporal representations from unlabeled sensor signals via global proxy-target regression.
  3. Outperforms supervised baseline under limited labeled data, sensor noise, and process drift.

Connect it to your work

A self-supervised framework enables MRR prediction using unlabeled sensor data without prior labeling.

What remains unproven

Performance degrades under sensor noise and temporal process drift conditions.

Originals used for this note

Links open the original publisher in a new tab.

  1. semiconductorsSelf-Supervised Virtual Metrology For CMP MRR Prediction (SUTD, A*STAR, NUS)