GPU libraries speed performance analysis for RCWA simulation matrix operations

Jingxiao Xu; Martin D. B. Charlton

doi:10.1117/12.2650112

10 March 2023 GPU libraries speed performance analysis for RCWA simulation matrix operations

Jingxiao Xu, Martin D. B. Charlton

Proceedings Volume 12415, Physics and Simulation of Optoelectronic Devices XXXI; 124150O (2023) https://doi.org/10.1117/12.2650112
Event: SPIE OPTO, 2023, San Francisco, California, United States

Conference Poster

Abstract

Rigorous Coupled Wave Analysis (RCWA) method is highly efficient for the simulation of diffraction efficiency and field distribution patterns in periodic structures and textured optoelectronic devices. GPU has been increasingly used in complex scientific problems such as climate simulation and the latest Covid-19 spread model. In this paper, we break down the RCWA simulation problem to key computational steps (eigensystem solution, matrix inversion/multiplication) and investigate speed performance provided by optimized linear algebra GPU libraries in comparison to multithreaded Intel MKL CPU library running on IRIDIS 5 supercomputer (1 NVIDIA v100 GPU and 40 Intel Xeon Gold 6138 cores CPU). Our work shows that GPU outperforms CPU significantly for all required steps. Eigensystem solution becomes 60% faster, Matrix inversion improves with size achieving 8x faster for large matrixes. Most significantly, matrix multiplication becomes 40x faster for small and 5x faster for large matrix sizes.

Citation Download Citation

Jingxiao Xu and Martin D. B. Charlton "GPU libraries speed performance analysis for RCWA simulation matrix operations", Proc. SPIE 12415, Physics and Simulation of Optoelectronic Devices XXXI, 124150O (10 March 2023); https://doi.org/10.1117/12.2650112

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available