AVS 66th International Symposium & Exhibition
    Plasma Science and Technology Division Monday Sessions
       Session PS2-MoM

Paper PS2-MoM9
Prediction of Etch Rates for New Materials by Machine Learning - Case Study for Physical Sputtering

Monday, October 21, 2019, 11:00 am, Room B130

Session: Plasma Modeling
Presenter: Kazumasa Ikuse, Osaka University, Japan
Authors: K. Ikuse, Osaka University, Japan
H. Kino, National Institute for Materials Science (NIMS), Japan
S. Hamaguchi, Osaka University, Japan
Correspondent: Click to Email
Due to the latest development of new chip designs, various non-conventional materials, such as ferromagnetic metals for magnetoresistive random access memories (MRAMs) and perovskite-type oxides for resistive random access memories (ReRAMs), have been introduced to microelectronics devices and required to be processed together with conventional Si based materials in the chip manufacturing processes. Furthermore, as the device dimensions are approaching the atomic scale, new process technologies, such as atomic-layer deposition (ALD) and atomic-layer etching (ALE), have been introduced to the surface processing, where the interactions between the processed surface and newly introduced gaseous species are not necessarily well understood. Having a variety of choices for surface materials and process conditions increases the complexity of process development because of our insufficient knowledge on surface reactions in the new processes. With a large number of possible choices of process conditions, exhaustive search for process optimization by experiments is prohibitively expensive. One of the possible solutions to this problem is to use machine learning (ML) to predict certain characteristics of the surface reactions such as the etching/deposition rates and surface chemical compositions, based on the existing knowledge of materials and gas-phase molecules involved in the new processes.

As the first step to develop such technologies based on data driven science, we have developed a system to predict the physical sputtering yields of single-element materials under single-species ion bombardment [1,2], based on the experimental sputtering yield data provided in Ref. [3]. Identification of the material/ion properties, which we call “descriptors,” that the sputtering yield strongly depends on is the key for the successful prediction of sputtering yields under unknown conditions. In this study, the selection of descriptors was performed by the sparse modeling with the exhaustive search method [4] and the subgroup relevance method [5].

[1] H. Kino, K. Ikuse, H. C. Dam, and S. Hamaguchi, 2nd International Conference on Data Drive Plasma Science (Marseille, 2019).

[2] K. Ikuse, K. Kino, and S. Hamaguchi, ibid.

[3] Y. Yamamura and H. Tawara, Atomic Data and Nuclear Data Tables 62, 149-253 (1996).

[4] K. Nagata, J. Kitazono, S. Nakajima, S. Eifuku, R. Tamura and M. Okada, IPSJ Online Transactions 8, 25 (2015).

[5] H. C. Dam, V. C. Nguyen, T. L.Pham, A.T. Nguyen, K. Terakura, T. Miyake and H. Kino, J. Phys. Soc. Jpn 87, 113801 (2018).