합성곱 신경망 병렬 연산처리를 지원하는 저전력 곱셈 프로세싱 엘리먼트 설계

Low-Power Multiplication Processing Element Hardware to Support Parallel Convolutional Neural Network Processing

  • 박은평 (목원대학교 지능정보융합학과) ;
  • 박종수 (목원대학교 전기전자공학과)
  • Eunpyoung Park ;
  • Jongsu Park
  • 투고 : 2024.03.29
  • 심사 : 2024.04.26
  • 발행 : 2024.04.30

초록

CNN은 이미지 인식분야에서 높은 성능을 보이지만 반복적인 학습이 진행될 경우 많은 데이터 연산처리로 인한 시스템 자원부족으로 학습 시간이 오래 걸리고 많은 전력을 소모한다는 단점이 있다. 이에 본 논문에서는 합성곱 신경망 연산처리의 핵심 요소인 곱셈 프로세싱 엘리먼트에서 곱셈연산을 수행할 때 발생되는 스위칭 엑티비티를 줄이기 위해 승수와 피승수의 교환율을 늘리는 저전력 부스 곱셈기를 기반으로 하는 프로세싱 엘리먼트를 제안한다. 합성곱 신경망 병렬 연산처리를 지원하는 저전력 곱셈 프로세싱 엘리먼트는 Verilog-HDL을 사용하여 설계되었고, Intel DE1-SoC FPGA Board에 구현하였다. 실험은 성능평가에 대표적으로 MNIST의 숫자 이미지 데이터베이스를 대상으로 기존 제안된 곱셈기의 교환율과 비교하여 성능을 검증하였다.

CNNs tend to take a long time to learn and consume a lot of power due to lack of system resources with many data processing units when there are repetitive handles that do not have high performance in the image field. In this paper, we propose a handling method based on a low-power bus that can increase the exchange rate of multipliers and multiplicands within the convolution mixer, which is a tendency activity that occurs when a convolution mixer has multiplication, which is the core element of combination. Convolutional neural networks have proprietary low-power shared processor support and the design was implemented on an Intel DE1-SoC FPGA board using Verilog-HDL. The experiments validated the performance by comparing it with the exchange rate of the multiplier originally proposed by Shen on MNIST's numeric image database.

키워드

참고문헌

  1. D. Perdios, M. Vonlanthen, F. Martinez, M. Arditi and J.-P. Thiran, "CNN-Based Ultrasound Image Reconstruction for Ultrafast Displacement Tracking," IEEE Transactions on Medical Imaging, vol. 40, no. 3, pp. 1078-1089, March 2021.
  2. N. A. S, V. Chaturvedi and M. Shafique, "FRNet: A Feature-Rich CNN Architecture to Defend Against Adversarial Attacks," IEEE Access, vol. 12, pp. 26943-26956, 2024.
  3. G. Gao, Z. Xu, J. Li, J. Yang, T. Zeng and G. -J. Qi, "CTCNet: A CNN-Transformer Cooperation Network for Face Image Super-Resolution," IEEE Transactions on Image Processing, vol. 32, pp. 1978-1991, 2023.
  4. Nan-Ying Shen and O.T.-C. Chen, "Low-power multipliers by minimizing switching activities of partial products," 2002 IEEE International Symposium on Circuits and Systems (ISCAS), Phoenix-Scottsdale, AZ, USA, 2002, pp. IV-IV, doi: 10.1109/ISCAS.2002.1010397.
  5. Chang-Young Han, Hyoung-Joon Park and Lee-Sup Kim, "A low-power array multiplier using separated multiplication technique," IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing, vol. 48, no. 9, pp. 866-871, Sept. 2001.
  6. Yilun He, "Image Processing System Based on Convolutional Neural Networks," Ph.D. dissertation, Department of Computer Engineering, Paichai Univ., Daejeon, Korea, 2018.
  7. Jongsu Park, Jinsang Kim, and Won-kyung Cho, "Low-Power Multiplier Using Input Data Partition", Journal of Korean Institute of Communications and Information Sciences, vol. 30, no. 11A, pp. 1093-1097, 2005.
  8. Akmaljon Palvanov, and Young Im Cho,"Comparisons of Deep Learning Algorithms for MNIST in Real-Time Environment," International Journal of Fuzzy Logic and Intelligent Systems, vol. 18, no. 2, pp. 126-134, 2018.