Design of efficient Winograd convolution hardware and its quantization scheme

YAN Zheng; ZHANG Chenshuo; BAI Yichuan; DU Yuan; DU Li

doi:10.20193/j.ices2097-4191.2025.0042

PDF(14567 KB)

Integrated Circuits and Embedded Systems ›› 2025, Vol. 25 ›› Issue (8) : 41-52. DOI: 10.20193/j.ices2097-4191.2025.0042

Special Issue of Emerging Computing Chip Design

Design of efficient Winograd convolution hardware and its quantization scheme

Author information +

History +

Abstract

Convolution is the most common operation in CNN networks, and the power consumption of multiplication and accumulation operations in convolution is high, which limits the performance of many CNN hardware accelerators. Reducing the number of multiplications in convolution is one of the effective ways to improve the performance of CNN accelerators. As a fast convolution algorithm, Winograd algorithm could reduce up to 75% multiplications in convolution. However, the weights of the model for Winograd convolution have a significantly different distribution, which results in longer quantization bit width to maintain similar accuracy and neutralizes the hardware reduction brought by the reduction of multiplications. In this paper, we analyze this problem quantitively and propose a new quantization scheme for Winograd convolution. The quantized Winograd computation hardware module is implemented with accuracy loss less than 1%. To further reduce the hardware cost, we apply the approximate multiplier (AM) to Winograd convolution. Compared with the conventional convolution computation block, the Winograd block saves 27.3% of the area, and the application of the approximate multiplier in Winograd block saves 39.6% of the area without significant performance loss.

Key words

convolution neural networks / Winograd algorithm / model quantization / approximate multiplier / hardware accelerator

Cite this article

EndNote

Ris (Procite)

Bibtex

Download Citations

YAN Zheng , ZHANG Chenshuo , BAI Yichuan , et al . Design of efficient Winograd convolution hardware and its quantization scheme[J]. Integrated Circuits and Embedded Systems. 2025, 25(8): 41-52 https://doi.org/10.20193/j.ices2097-4191.2025.0042

References

List( Publishing order | Descend order by publishing year | Descend order by cited within ) Chart analysis

[1]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2016:770-778. Cited in this article [1]

[2]	MOLLAHOSSEINI A, CHAN D, MAHOOR M H. Going deeper in facial expression recognition using deep neural networks[C]// 2016 IEEE Winter conference on applications of computer vision (WACV). IEEE, 2016:1-10. Cited in this article [1]

[3]	EN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2016, 39(6):1137-1149. Cited in this article [1]

[4]	REDMON J, FARHADI A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767, 2018. Cited in this article [2]

[5]	ABDEL-HAMID O, MOHAMED A, JIANG H, et al. Convolutional neural networks for speech recognition[J]. IEEE/ACM Transactions on audio,speech,and language processing, 2014, 22(10):1533-1545. Cited in this article [1]

[6]	LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2015:3431-3440. Cited in this article [1]

[7]	YUAN Z, LIU Y, YUE J, et al. STICKER:An energy-efficient multi-sparsity compatible accelerator for convolutional neural networks in 65-nm CMOS[J]. IEEE Journal of Solid-State Circuits, 2019, 55(2):465-477. Cited in this article [1]

[8]

MARCHESI

, ORLANDI

, PIAZZA

, et al. Fast neural networks without multipliers[J]. IEEE transactions on Neural Networks, 1993, 4(1):53-62.

https://www.ncbi.nlm.nih.gov/pubmed/18267703

Cited in this article [1] Abstract

Multilayer perceptrons (MLPs) with weight values restricted to powers of two or sums of powers of two are introduced. In a digital implementation, these neural networks do not need multipliers but only shift registers when computing in forward mode, thus saving chip area and computation time. A learning procedure, based on backpropagation, is presented for such neural networks. This learning procedure requires full real arithmetic and therefore must be performed offline. Some test cases are presented, concerning MLPs with hidden layers of different sizes, on pattern recognition problems. Such tests demonstrate the validity and the generalization capability of the method and give some insight into the behavior of the learning algorithm.

[9]	LIN Z, COURBARIAUX M, MEMISEVIC R, et al. Neural networks with few multiplications[J]. arXiv preprint arXiv:1510.03009, 2015. Cited in this article [2]

[10]	SARWAR S S, VENKATARAMANI S, ANKIT A, et al. Energy-efficient neural computing with approximate multipliers[J]. ACM Journal on Emerging Technologies in Computing Systems (JETC), 2018, 14(2):1-23. Cited in this article [2]

[11]	STRASSEN V. Gaussian elimination is not optimal[J]. Numerische mathematik, 1969, 13(4):354-356. Cited in this article [1]

[12]	KOUYA T. Accelerated multiple precision matrix multiplication using Strassen's algorithm and Winograd's variant[J]. JSIAM Letters, 2014(6):81-84. Cited in this article [1]

[13]	GROCHOW J A, MOORE C. Designing Strassen's algorithm[J]. arXiv preprint arXiv:1708.09398, 2017. Cited in this article [1]

[14]	PAN V Y. Computation schemes for a product of matrices and for the inverse matrix[J]. Uspekhi Matematicheskikh Nauk, 1972, 27(5):249-250. Cited in this article [1]

[15]	ALI M, YIN B, KUNAR A, et al. Reduction of multiplications in convolutional neural networks[C]// 2020 39th Chinese control conference (CCC). IEEE, 2020:7406-7411. Cited in this article [2]

[16]	PARKER D A, PARHI K K. Area-efficient parallel FIR digital filter implementations[C]// Proceedings of International Conference on Application Specific Systems, Architectures and Processors:ASAP'96. IEEE, 1996:93-111. Cited in this article [1]

[17]	WANG J, LIN J, WANG Z. Efficient hardware architectures for deep convolutional neural network[J]. IEEE Transactions on Circuits and Systems I:Regular Papers, 2017, 65(6):1941-1953. Cited in this article [1]

[18]	WINOGRAD S. Arithmetic complexity of computations[M]. Siam, 1980. Cited in this article [3]

[19]	WANG X, WANG C, ZHOU X. Work-in-progress: Optimising FPGA-based neural network accelerators using fast winograd algorithm[C]// 2018 International Conference on Hardware/Software Codesign and System Synthesis (CODES+ISSS). IEEE, 2018:1-2. Cited in this article [1]

[20]	MAJI P, MUNDY A, DASIKA G, et al. Efficient winograd or cook-toom convolution kernel implementation on widely used mobile cpus[C]// 2019 2nd Workshop on Energy Efficient Machine Learning and Cognitive Computing for Embedded Applications (EMC2). IEEE, 2019:1-5. Cited in this article [1]

[21]	DICECCO R, LACEY G, VASILJEVIC J, et al. Caffeinated FPGAs: FPGA framework for convolutional neural networks[C]// 2016 International Conference on Field-Programmable Technology (FPT). IEEE, 2016:265-268. Cited in this article [1]

[22]	LIANG Y, LU L, XIAO Q, et al. Evaluating fast algorithms for convolutional neural networks on FPGAs[J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2019, 39(4):857-870. Cited in this article [1]

[23]	PARK H, KIM D, AHN J, et al. Zero and data reuse-aware fast convolution for deep neural networks on GPU[C]// Proceedings of the Eleventh IEEE/ACM/IFIP International Conference on Hardware/Software Codesign and System Synthesis, 2016:1-10. Cited in this article [1]

[24]	LAI P W, ARAFAT H, ELANGO V, et al. Accelerating Strassen-Winograd's matrix multiplication algorithm on GPUs[C]// 20th Annual international conference on high performance computing. IEEE, 2013:139-148. Cited in this article [1]

[25]	DI X, YANG H, HUANG Z, et al. An operation-minimized FPGA accelerator design by dynamically exploiting sparsity in CNN Winograd transform[C]// 2019 32nd IEEE International System-on-Chip Conference (SOCC). IEEE, 2019:50-55. Cited in this article [1]

[26]	LAVIN A, GRAY S. Fast algorithms for convolutional neural networks[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2016:4013-4021. Cited in this article [3]

[27]	WINOGRAD S. On multiplication of polynomials modulo a polynomial[J]. SIAM Journal on Computing, 1980, 9(2):225-229. Cited in this article [1]

[28]	MADISETTI V K, YOUNG I T. The Digital Signal Processing Handbook-3 Volume Set[M]. CRC press, 2018. Cited in this article [1]

[29]	COURBARIAUX M, BENGIO Y, DAVID J P. Low precision arithmetic for deep learning[C]// ICLR (Workshop), 2015. Cited in this article [2]

[30]	NARAYANAMOORTHY S, MOGHADDAM H A, LIU Z, et al. Energy-efficient approximate multiplication for digital signal processing and classification applications[J]. IEEE transactions on very large scale integration (VLSI) systems, 2014, 23(6):1180-1184. Cited in this article [3]

[31]	HORE A, ZIOU D. Image quality metrics: PSNR vs. SSIM[C]// 2010 20th international conference on pattern recognition, IEEE, 2010:2366-2369. Cited in this article [1]