Few-Shot Test-Time Optimization Without Retraining for Semiconductor Recipe Generation and Beyond

2025-05-21

AI Summary

In industrial settings—such as semiconductor manufacturing—where deployed AI models or hardware systems are immutable or prohibitively expensive to retrain or modify, conventional test-time adaptation methods are inapplicable. Method: This paper proposes Model Feedback Learning (MFL), a parameter-free test-time adaptation framework that optimizes few-shot inputs via lightweight inverse modeling and iterative gradient-based search—without updating model parameters or altering hardware. MFL incorporates a stability-aware constraint mechanism to ensure robustness under high-dimensional industrial control. Contribution/Results: MFL establishes the first parameter-free test-time adaptation paradigm, significantly improving robustness and generalization in safety-critical industrial processes. In plasma etching, it generates target process recipes within only five iterations—outperforming both Bayesian optimization and expert knowledge. Its cross-task generalization is further validated on chemical vapor deposition (CVD) and wire bonding, demonstrating broad applicability across diverse semiconductor fabrication processes.

Technology Category

• Machine Learning: Hardware-aware ML
• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

We introduce Model Feedback Learning (MFL), a novel test-time optimization framework for optimizing inputs to pre-trained AI models or deployed hardware systems without requiring any retraining of the models or modifications to the hardware. In contrast to existing methods that rely on adjusting model parameters, MFL leverages a lightweight reverse model to iteratively search for optimal inputs, enabling efficient adaptation to new objectives under deployment constraints. This framework is particularly advantageous in real-world settings, such as semiconductor manufacturing recipe generation, where modifying deployed systems is often infeasible or cost-prohibitive. We validate MFL on semiconductor plasma etching tasks, where it achieves target recipe generation in just five iterations, significantly outperforming both Bayesian optimization and human experts. Beyond semiconductor applications, MFL also demonstrates strong performance in chemical processes (e.g., chemical vapor deposition) and electronic systems (e.g., wire bonding), highlighting its broad applicability. Additionally, MFL incorporates stability-aware optimization, enhancing robustness to process variations and surpassing conventional supervised learning and random search methods in high-dimensional control settings. By enabling few-shot adaptation, MFL provides a scalable and efficient paradigm for deploying intelligent control in real-world environments.

Problem

• Optimizing inputs to pre-trained models without retraining
• Efficient adaptation to new objectives under deployment constraints
• Enhancing robustness to process variations in high-dimensional control

Innovation

• Model Feedback Learning optimizes inputs without retraining
• Lightweight reverse model enables efficient input adaptation
• Stability-aware optimization enhances robustness in control