Wafer-Level Etch Spatial Profiling for Process Monitoring from Time-Series with Time-LLM
Summary: arXiv:2603.23576v1 Announce Type: cross
Abstract
Understanding wafer-level spatial variations from in-situ process signals is essential for advanced plasma etching process monitoring. While most data-driven approaches focus on scalar indicators such as average etch rate, actual process quality is determined by complex two-dimensional spatial distributions across the wafer. This paper presents a spatial regression model that predicts wafer-level etch depth distributions directly from multichannel in-situ process time series.
Introduction
The semiconductor manufacturing industry has witnessed significant advancements in etching technologies, particularly in plasma etching processes. As the demand for miniaturization and higher performance increases, so does the need for accurate monitoring of etching processes to ensure quality and yield. This article discusses a novel approach using a Time-LLM-based spatial regression model to monitor wafer-level etch depth distributions effectively.
Methodology
In this research, we propose a Time-LLM-based spatial regression model that extends LLM (Large Language Model) reprogramming from conventional time-series forecasting to wafer-level spatial estimation. The innovative approach involves redesigning both the input embedding and output projection to address the unique challenges presented by etch depth profiling.
Key Features of the Time-LLM Model
- Input Embedding Redesign: The model incorporates multichannel in-situ process time series data, allowing it to analyze various parameters simultaneously.
- Output Projection Enhancement: By focusing on spatial distributions rather than scalar values, the model provides a more comprehensive understanding of etch depth variations.
- Data-Limited Capabilities: The model demonstrates stable performance even with limited data, making it a practical solution for real-world applications.
Results
Using the BOSCH plasma-etching dataset, we evaluated the performance of the Time-LLM-based model. The results indicated that the model effectively predicted wafer-level etch depth distributions, showcasing its potential for real-time monitoring and quality assurance in semiconductor manufacturing.
Discussion
The implications of this research extend beyond the immediate application of the Time-LLM model. By achieving accurate spatial profiling, manufacturers can enhance process control, reduce defects, and ultimately improve yield rates. Moreover, the ability to operate effectively under data-limited conditions is crucial in practical scenarios where data collection can be constrained.
Conclusion
In conclusion, the Time-LLM-based spatial regression model presents a significant advancement in wafer-level etch monitoring. By redefining traditional approaches to accommodate spatial distributions, this model stands to revolutionize the way semiconductor manufacturers monitor etching processes. Future work will focus on further refining the model and exploring its applicability across different etching technologies.