Abstract
Equipping edge devices with deep neural networks (DNNs) could result in a revolution in human interactions with surrounding environments as edge devices would be able to perform more complex tasks. However, DNNs are power-hungry, performing billions of computations in terms of one inference. Applying approximate computing techniques reduces the cost of the underlying circuits so that DNN inferences would be performed more efficiently where negligible inference accuracy loss is acceptable.
There are many approximate multipliers proposed for various applications until now. However, the function of only a few of these approximate designs has been explored while performing inference of DNNs. Furthermore, little attention has been given on applying various approximation techniques into different layers of DNNs. In this chapter, first, a detailed stepwise approach for designing a re-configurable approximate Booth multiplier using the common available approximate techniques is presented. Then, it is shown that to get the optimum accuracy out of the available approximation techniques, it is necessary to have a re-configurable multiplier to apply various approximation techniques into different layers of DNNs. Lastly, the function of the proposed multiplier in this chapter is evaluated in a real accelerator and compared to other designs.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
References
Zhang H, Ko SB. Design of power efficient posit multiplier. IEEE Trans Circuits Syst II Exp Briefs 2020;67 5:861–5.
Chen Y, Luo T, Liu S, Zhang S, He L, Wang J, Li L, Chen T, Xu Z, Sun N. et al. DaDianNao: a machine-learning supercomputer. In: 2014 47th annual IEEE/ACM international symposium on microarchitecture. Piscataway: IEEE; 2014. p. 609–22.
Choi Y, Bae D, Sim J, Choi S, Kim M, Kim LS. Energy-efficient design of processing element for convolutional neural network. IEEE Trans Circuits Syst II Exp Briefs 2017;64 11:1332–1336.
Tang Y, Zhang J, Verma N. Scaling up in-memory-computing classifiers via boosted feature subsets in banked architectures. IEEE Trans Circuits Syst II Exp Briefs 2018;66 3:477–81.
Chen KC, Ebrahimi M, Wang TY, Yang YC. NoC-based DNN accelerator: a future design paradigm. In: Proceedings of the 13th IEEE/ACM international symposium on networks-on-chip; 2019. p. 1–8.
Chen YH, Yang TJ, Emer J, Sze V. Eyeriss v2: a flexible accelerator for emerging deep neural networks on mobile devices. IEEE J Emerg Sel Top Circuits Syst. 2019;9 2:292–308.
Chen YH, Emer J, Sze V. Using dataflow to optimize energy efficiency of deep neural network accelerators. IEEE Micro 2017;37 3:12–21.
Cheng C, Parhi KK. Fast 2d convolution algorithms for convolutional neural networks. IEEE Trans Circuits Syst I Regul Pap. 2020;67 5:1678–91.
Lavin A, Gray S. Fast algorithms for convolutional neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition; 2016. p. 4013–21.
Chen YH, Krishna T, Emer JS, Sze V. Eyeriss: an energy-efficient reconfigurable accelerator for deep convolutional neural networks. IEEE J Solid-State Circuits 2016;52 1:127–38.
Sharify S, Lascorz AD, Mahmoud M, Nikolic M, Siu K, Stuart DM, Poulos Z, Moshovos A. Laconic deep learning inference acceleration. In: 2019 ACM/IEEE 46th annual international symposium on computer architecture (ISCA). Piscataway: IEEE; 2019. p. 304–17.
Albericio J, Delmás A, Judd P, Sharify S, O’Leary G, Genov R, Moshovos A. Bit-pragmatic deep neural network computing. In: Proceedings of the 50th annual IEEE/ACM international symposium on microarchitecture; 2017. p. 382–94.
Delmas Lascorz A, Judd P, Stuart DM, Poulos Z, Mahmoud M, Sharify S, Nikolic M, Siu K, MoshovosA. Bit-tactical: a software/hardware approach to exploiting value and bit sparsity in neural networks. In: Proceedings of the twenty-fourth international conference on architectural support for programming languages and operating systems;2019. p. 749–63.
Asadikouhanjani M, Zhang H, Gopalakrishnan L, Lee HJ, Ko SB. A real-time architecture for pruning the effectual computations in deep neural networks. IEEE Trans Circuits Syst I Regul Pap. 2021;68 5:2030–41.
Kim MS, Barrio AAD, Oliveira LT, Hermida R, Bagherzadeh N. Efficient Mitchell’s approximate log multipliers for convolutional neural networks. IEEE Trans Comput. 2019;68 5:660–75.
Du Z, Lingamneni A, Chen Y, Palem KV, Temam O, Wu C. Leveraging the error resilience of neural networks for designing highly energy efficient accelerators. IEEE Trans Comput Aid Des Integr Circuits Syst. 2015;34 8:1223–35.
Chen CY, Choi J, Brand D, Agrawal A, Zhang W, Gopalakrishnan K. AdaComp: adaptive residual gradient compression for data-parallel distributed training. In: Proceedings of the AAAI conference on artificial intelligence vol. 32, no. 1; 2018.
Vaverka F, Hrbacek R, Sekanina L. Evolving component library for approximate high level synthesis. In: 2016 IEEE symposium series on computational intelligence (SSCI); 2016. p. 1–8.
Park G, Kung J, Lee Y. Design and analysis of approximate compressors for balanced error accumulation in MAC operator. IEEE Trans Circuits Syst I Regul Pap. 2021;68 7:2950–61.
Sarwar SS, Venkataramani S, Ankit A, Raghunathan A, Roy K. Energy-efficient neural computing with approximate multipliers. ACM J Emerg Technol Comput Syst. 2018;14 2:1–23.
Venkataramani S, Ranjan A, Roy K, Raghunathan A. AxNN: energy-efficient neuromorphic systems using approximate computing. In: 2014 IEEE/ACM international symposium on low power electronics and design (ISLPED). Piscataway: IEEE; 2014. p. 27–32.
Kim MS, Del Barrio Garcia AA, Kim H, Bagherzadeh N. The effects of approximate multiplication on convolutional neural networks. IEEE Trans Emerg Top Comput. 2021:1–1. https://doi.org/10.1109/TETC.2021.3050989.
Chen Y, Xie Y, Song L, Chen F, Tang T. A survey of accelerator architectures for deep neural networks. Engineering 2020;6 3:264–274.
He X, Ke L, Lu W, Yan G, Zhang X. AxTrain: hardware-oriented neural network training for approximate inference. In: Proceedings of the international symposium on low power electronics and design; 2018. p. 1–6.
He X, Lu W, Yan G, Zhang X. Joint design of training and hardware towards efficient and accuracy-scalable neural network inference. IEEE J Emerg Sel Top Circuits Syst. 2018;8 4:810–21.
Venkatachalam S, Adams E, Lee HJ, Ko SB. Design and analysis of area and power efficient approximate booth multipliers. IEEE Trans Comput 2019;68 11:1697–703.
Iandola FN, Han S, Moskewicz MW, Ashraf K, Dally WJ, Keutzer K. SqueezeNet: alexNet-level accuracy with 50x fewer parameters and< 0.5 mb model size; 2016. arXiv:1602.07360.
Venkatachalam S, Ko SB. Design of power and area efficient approximate multipliers. IEEE Trans Very Large Scale Integr Syst. 2017;25 5:1782–86.
Esposito D, Strollo AG, Alioto M. Low-power approximate mac unit. In: 2017 13th conference on Ph. D. research in microelectronics and electronics (PRIME). Piscataway: IEEE; 2017. p. 81–4.
Yang T, Sato T, Ukezono T. A low-power approximate multiply-add unit. In: 2019 2nd international symposium on devices, circuits and systems (ISDCS). Piscataway: IEEE; 2019. p. 1–4.
Lu Y, Shan W, Xu J. A depthwise separable convolution neural network for small-footprint keyword spotting using approximate mac unit and streaming convolution reuse. In: 2019 IEEE Asia Pacific conference on circuits and systems (APCCAS). Piscataway: IEEE;2019. p. 309–12.
Kim MS, Del Barrio AA, Hermida R, Bagherzadeh N. Low-power implementation of Mitchell’s approximate logarithmic multiplication for convolutional neural networks. In: 2018 23rd Asia and South Pacific design automation conference (ASP-DAC); 2018. p. 617–22.
https://image-net.org/challenges/LSVRC/2012/index.php (2012). Accessed 3 Aug 2021.
Ahmadinejad M, Moaiyeri MH. Energy-and quality-efficient approximate multipliers for neural network and image processing applications. IEEE Trans Emerg Top Comput. 2021:1–1. https://doi.org/10.1109/TETC.2021.3072666.
Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, Erhan D, Vanhoucke V, Rabinovich A. Going deeper with convolutions. In: Proceedings of the IEEE conference on computer vision and pattern recognition; 2015. p. 1–9.
https://www.mathworks.com/help/deeplearning/ug/deep-learning-speech-recognition.html (2017). Accessed 3 Aug 2021.
http://download.tensorflow.org/data/speech_commands_v0.02.tar.gz (2016). Accessed 3 Aug 2021.
Acknowledgements
This work was supported by NSERC of Canada, the R&D program of MOTIE/KEIT (No. 10077609, Developing Processor Memory Storage Integrated Architecture for Low Power, High Performance Big Data Servers) and Korea Electrotechnology Research Institute (An Energy-Efficient DNN-Based Environmental Sound Classifier).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this chapter
Cite this chapter
Asadikouhanjani, M., Zhang, H., Cho, K., Park, YJ., Ko, S.B. (2022). Efficient Approximate DNN Accelerators for Edge Devices: An Experimental Study. In: Liu, W., Lombardi, F. (eds) Approximate Computing. Springer, Cham. https://doi.org/10.1007/978-3-030-98347-5_19
Download citation
DOI: https://doi.org/10.1007/978-3-030-98347-5_19
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-98346-8
Online ISBN: 978-3-030-98347-5
eBook Packages: Computer ScienceComputer Science (R0)