Venue: International Conference on Data-Driven Plasma Science
Year: 2021
Abstract
Creating a stable and repeatable processing environment is a long-standing challenge in the semiconductor industry. Clusters of tools require tuning to match chambers and produce the same process results and run to run control strategies are required to correct for process drifts over time. The capability to perform real-time measurement and control of the state of a processing plasma could bring significant benefits.
Current approaches have focused on trying to measure specific plasma parameters, such as electron density, and develop systems to control that specific parameter[1]. Although these have shown some success they have not been widely adopted in the industry[2], [3]. In this approach, there is always difficulty in determining how to use collected sensor inputs and how to construct a control function for them.
In this work we are presenting our initial progress in using state of the art deep learning in a unique approach to address this problem. We will demonstrate how multiple non-invasive plasma diagnostics can be combined in Convolutional Neural Networks (CNN) to predict the state of the plasma in an industrial etcher. Our initial focus is on industrially relevant etch plasmas and optical plasma diagnostics. We will also present some early validation work using existing plasma probes, that have a good theoretical basis for calculating plasma parameters.
[1] B. Keville, Y. Zhang, C. Gaman, A. M. Holohan, S. Daniels, and M. M. Turner, ‘Real-time control of electron density in a capacitively coupled plasma’, J. Vac. Sci. Technol. A, vol. 31, no. 3, p. 031302, Mar. 2013.
[2] S. Ikuhara, A. Kagoshima, D. Shiraishi, and J. Tanaka, ‘High-accuracy Etching System with Active APC Capability’, 2005.
[3] S. Umeda, K. Nogi, D. Shiraishi, and A. Kagoshima, ‘Advanced Process Control Using Virtual Metrology to Cope with Etcher Condition Change’, 2018 Int. Symp. Semicond. Manuf. ISSM, pp. 1–4, 2018.
Additional information
- Note: This is a talk at ICDDPS rather than a paper.