An On-Chip 2-D DFT Accelerator Ultrasonic Wavefront for Convolutional Neural Networks

An On-Chip 2-D DFT Accelerator Ultrasonic Wavefront for Convolutional Neural Networks

2021 IEEE USNC-URSI Radio Science Meeting (Joint with AP-S Symposium)

10.23919/USNC-URSI51813.2021.9703556

http://dx.doi.org/10.23919/usnc-ursi51813.2021.9703556

Kok-Hin Teng, Salahuddin Raju, Di Zhu, Jayce Lay Keng Lim, Daniel Ssu-Han Chen, Eva Wai Leong Ching, Sharma Jaibir, Lee En-Yuan Joshua, Eldwin Jiaqiang Ng, Kevin Chai Tshun Chuan, Amit Lal

10 February 2022

Teng, K.-H., Raju, S., Zhu, D., Lim, J. L. K., Chen, D. S.-H., Ching, E. W. L., Jaibir, S., Joshua, L. E.-Y., Ng, E. J., Chuan, K. C. T., & Lal, A. (2021). An On-Chip 2-D DFT Accelerator Ultrasonic Wavefront for Convolutional Neural Networks. 2021 IEEE USNC-URSI Radio Science Meeting (Joint with AP-S Symposium). https://doi.org/10.23919/usnc-ursi51813.2021.9703556

A 2-dimensional discrete Fourier transform (2-D DFT) is a required preprocessing step for convolutional neural networks (CNNs) to perform matrix multiplication in convolutional layers. Here we present an ultrasonic wavefront-based architecture for CNNs that harness the wave propagation diffraction physic to perform Fourier transform (FT) effectively. The computation is improved to O(N) compare to Fast Fourier transform (FFT) with O(N(log2N). In addition, analysis of the proposed ultrasonic wavefront scheme is described in this paper.

Publisher Copyright

This research / project is supported by the A*STAR - AME Programmatic Fund - Ultrasonic Wavefront Computing
Grant Reference no. : A19E8b0102

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

https://oar.a-star.edu.sg/communities-collections/articles/18415

978-1-946815-10-1

Institute of Microelectronics