基于SRAM的通用存算一体架构平台在物联网中的应用

曾剑敏; 张章; 虞志益; 解光军

doi:10.11999/JEIT210010

基于SRAM的通用存算一体架构平台在物联网中的应用

doi: 10.11999/JEIT210010

1.
合肥工业大学电子科学与应用物理学院合肥 230601
2.
中山大学微电子科学与技术学院珠海 519082

基金项目: 国家自然科学基金(U19A2053, 61674049)，中央高校基本科研基金(JZ2020YYPY0089)，中国科学院红外成像材料与器件重点实室开放课题(IIMDKFJJ-19-04)

详细信息

作者简介:
曾剑敏：男，1991年生，博士生，研究方向为存算一体架构设计

张章：男，1982年生，教授，主要研究方向为集成电路设计及基于混合器件的AI神经形态芯片设计

虞志益：男，1977年生，教授，研究方向为处理器架构设计、基于非易失器件、面向人工智能深度学习等应用的电路与系统设计

解光军：男，1970年生，教授，主要研究方向为集成电路及新型器件电路设计

通讯作者:
张章　zhangzhang@hfut.edu.cn

中图分类号: TN47
计量
- 文章访问数: 1122
- HTML全文浏览量: 921
- PDF下载量: 144
- 被引次数: 0
出版历程
- 收稿日期: 2021-01-05
- 修回日期: 2021-04-09
- 网络出版日期: 2021-04-30
- 刊出日期: 2021-06-18

Applications of Generic In-memory Computing Architecture Platform Based on SRAM to Internet of Things

1.
School of Electronic Science and Applied Physics, Hefei University of Technology, Hefei 230601, China
2.
School of Microelectronics Science and Technology, Sun Yat-sen University, Zhuhai 519082, China

Funds: The National Natural Science Foundation of China(U19A2053, 61674049), The Fundamental Research Funds for Central Universities (JZ2020YYPY0089), The Key Laboratory of CAS (IIMDKFJJ-19-04)

摘要

摘要: 最近，存算一体(IMC)架构引起了广泛关注，并被认为有望成为突破冯诺依曼瓶颈的新型计算机架构，特别是在数据密集型(data-intensive)计算中能够带来显著的性能和功耗优势。其中，基于SRAM的IMC架构方案也被大量研究与应用。该文在一款基于SRAM的通用存算一体架构平台——DM-IMCA的基础上，探索IMC架构在物联网领域中的应用价值。具体来说，该文选取了物联网中包括信息安全、二值神经网络和图像处理在内的多个轻量级数据密集型应用，对算法进行分析或拆分，并将关键算法映射到DM-IMCA中的SRAM中，以达到加速应用计算的目的。实验结果显示，与基于传统冯诺依曼架构的基准系统相比，利用DM-IMCA来实现物联网中的轻量级计算密集型应用，可获得高达24倍的计算加速比。
- 物联网 /
- 超越冯诺依曼架构 /
- 存算一体 /
- 计算型SRAM
Abstract: In-Memory Computing (IMC) architectures have aroused much attention recently, and are regarded as promising candidates to break the von Neumann bottleneck. IMC architectures can bring significant performance and energy-efficiency improvement especially in data-intensive computation. Among those emerging IMC architectures, SRAM-based ones have also been extensively researched and applied to many scenarios. In this paper, IoT applications are explored based on a SRAM-based generic IMC architecture platform named DM-IMCA. To be specific, the algorithms of several lightweight data-intensive applications in IoT area including information security, Binary Neural Networks (BNN) and image processing are analyzed, decomposed and then mapped to SRAM macros of DM-IMCA, so as to accelerate the computation of these applications. Experimental results indicate that DM-IMCA can offer up to 24 times performance speed-up, compared to a baseline system with conventional von Neumann architecture, in terms of realizing lightweight data-intensive applications in IoT.
- Internet of Things (IoT) /
- Beyond von Neumann architecture /
- In-Memory Computing (IMC) /
- Computational SRAM

HTML全文

图 1 SRAM列内逻辑运算示意图

下载: 全尺寸图片幻灯片

图 2 DM-IMCA硬件架构图

下载: 全尺寸图片幻灯片

图 3 存内向量计算自动化及自适应示意图

下载: 全尺寸图片幻灯片

图 4 测试平台

下载: 全尺寸图片幻灯片

图 5 测试OTP加密所花费的时间对比

下载: 全尺寸图片幻灯片

图 6 加法哈希函数运算中字符串在IMC-SRAM中的映射

下载: 全尺寸图片幻灯片

图 7 测试加法哈希算法进行加密所耗费的时间对比

下载: 全尺寸图片幻灯片

图 8 二值神经网络中激活和权重在IMC-SRAM中的映射

下载: 全尺寸图片幻灯片

图 9 测试二进制卷积运算所花费的时钟周期数对比

下载: 全尺寸图片幻灯片

表 1 IMC指令集编码格式

指令	字段	名称
储配置指令： memCfg Rn	31–27	op (11001)
	26–4	Reserved
	3–0	rn
地址配置指令： addrCfg R3, R2, R1	31–27	op (11000)
	26–20	r3
	19–13	r2
	12–6	r1
	5–0	Reserved
计算指令： opcode vl (vl: vector length，矢量长度)	31–27	op (11010)
	26–23	function
	22–15	vl
	14–0	Reserved

下载: 导出CSV

表 2 IMC指令集

指令类型		指令	操作/功能
配置指令	存储配置	memCfg	配置寄存器Rn
配置指令	地址配置	addrCfg	配置存内计算地址寄存器R1～R3
计算指令	逻辑计算	mand	$c = a\& b$
		mor	$c = a\|b$
		mxor	$c = a \oplus b$
		mnor	$c = \sim(a\|b)$
		mnand	$c = \sim(a + b)$
		mnot	$c = \sim a$
	算术计算	madd	$c = a + b$(有符号)
		maddu	$c = a + b$(无符号)
		mop	$c = - a$
		minc	$c = a + 1$
		mdec	$c = a - 1$
		msl	$c = a \ll 1$
		msr	$c = a \gg 1$
	存储操作	mcopy	$c = a$

下载: 导出CSV

表 3 OTP算法

输入: plaintext P[N], key K[N], N为数组P和K的字长
输出: ciphertext C[N]
(1)　for (unsigned i = 0; i < N; i++) do
(2)　C[i] = P[i]⊕K[i];
(3)　end for

下载: 导出CSV

表 4 加法哈希算法

输入: char *key, uint32_t len, uint32_t prime
输出: result
(1) uint32_t hash, i
(2) for (hash = len, i = 0; i < len; i++) do
(3) hash +=key[i]
(4) end for
(5) return (hash % prime)

下载: 导出CSV

表 5 二值乘法真值表

a_i	w_i	f_i=a_i×w_i
–1	–1	1
–1	1	–1
1	–1	–1
1	1	1

下载: 导出CSV

表 6 存内矢量XOR运算和popcount操作算法

输入: int $A[N]$, $B[N]$, $N$为数组A和B长度
输出: S
(1) $M[i] \leftarrow {\rm{0x}}0001,i \in [0,N - 1]$
(2) $D[i] \leftarrow 0,i \in \in [0,N - 1]$
(3) $S \leftarrow 0$
(4) $A[i] \leftarrow A[i] \oplus B[i],i \in [0,N - 1]$
(5) $C[i] \leftarrow A[i]\& M[i],i \in [0,N - 1]$
(6) $D[i] \leftarrow D[i] + C[i],i \in [0,N - 1]$
(7) for (i = 0; i < 31; i++) do
(8) $C[i] \leftarrow A[i] \gg 1,i \in [0,N - 1]$
(9) $C[i] \leftarrow A[i]\& M[i],i \in [0,N - 1]$
(10) $D[i] \leftarrow D[i] + C[i],i \in [0,N - 1]$
(11) end for
(12) for (i = 0; i < N; i++) do
(13) $S \leftarrow S + D[i],i \in [0,N - 1]$
(14) end for
(15) return $S$

下载: 导出CSV

参考文献(36)

[1]	JIN Xiaolong, WAH B W, CHENG Xueqi, et al. Significance and challenges of big data research[J]. Big Data Research, 2015, 2(2): 59–64. doi: 10.1016/j.bdr.2015.01.006
[2]	SCHILIZZI R T. The square kilometer array[C]. Proceedings SHE, Ground-based Telescopes, Glasgow, UK, 2004: 62–71.
[3]	JONGERIUS R, WIJNHOLDS S, NIJBOER R, et al. An end-to-end computing model for the square kilometre array[J]. Computer, 2014, 47(9): 48–54. doi: 10.1109/MC.2014.235
[4]	ZASLAVSKY A, PERERA C, and GEORGAKOPOULOS D. Sensing as a service and big data[J]. arXiv preprint arXiv: 1301.0159, 2013.
[5]	FLORIDI L. Big data and their epistemological challenge[J]. Philosophy & Technology, 2012, 25(4): 435–437. doi: 10.1007/s13347-012-0093-4
[6]	REINSEL D, GANTZ J, and RYDNING J. Data age 2025: The digitization of the world from edge to core[R]. IDC White Paper–#US44413318, 2018.
[7]	SIEGL P, BUCHTY R, and BEREKOVIC M. Data-Centric computing frontiers: A survey on processing-in-memory[C]. Proceedings of the Second International Symposium on Memory Systems (MEMSYS'16), Alexandria, USA, 2016: 295–308. doi: 10.1145/2989081.2989087.
[8]	PATTERSON D, ANDERSON T, CARDWELL N, et al. A case for intelligent RAM[J]. IEEE Micro, 1997, 17(2): 34–44. doi: 10.1109/40.592312
[9]	ROGERS B M, KRISHNA A, BELL G B, et al. Scaling the bandwidth wall: Challenges in and avenues for CMP scaling[J]. ACM SIGARCH Computer Architecture News, 2009, 37(3): 371–382. doi: 10.1145/1555815.1555801
[10]	DAS R. Blurring the lines between memory and computation[J]. IEEE Micro, 2017, 37(6): 13–15. doi: 10.1109/MM.2017.4241340
[11]	ZHU Qiuling, AKIN B, SUMBUL H E, et al. A 3D-stacked logic-in-memory accelerator for application-specific data intensive computing[C]. 2013 IEEE International 3D Systems Integration Conference (3DIC), San Francisco, USA, 2013: 1–7. doi: 10.1109/3DIC.2013.6702348.
[12]	AHN J, HONG S, YOO S, et al. A scalable processing-in-memory accelerator for parallel graph processing[C]. Proceeding of the ACM/IEEE 42nd Annual International Symposium on Computer Architecture (ISCA'15), Portland, USA, 2015: 105–117. doi: 10.1145/2749469.2750386.
[13]	AHN J, YOO S, MUTLU O, et al. PIM-enabled instructions: A low-overhead, locality-aware processing-in-memory architecture[C]. The 42nd Annual International Symposium on Computer Architecture (ISCA'15), Portland, USA, 2015: 336–348. doi: 10.1145/2749469.2750385.
[14]	CHI Ping, LI Shuangchen, XU Cong, et al. PRIME: A novel processing-in-memory architecture for neural network computation in ReRAM-based main memory[J]. ACM SIGARCH Computer Architecture News, 2016, 44(3): 27–39. doi: 10.1145/3007787.3001140
[15]	HALAWANI Y, MOHAMMAD B, LEBDEH M A, et al. ReRAM-based in-memory computing for search engine and neural network applications[J]. IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 2019, 9(2): 388–397. doi: 10.1109/JETCAS.2019.2909317
[16]	ZHANG Bin, CHEN Weilin, ZENG Jianmin, et al. 90% yield production of polymer nano-memristor for in-memory computing[J]. Nature Communications, 2021, 12(1): 1984. doi: 10.1038/s41467-021-22243-8
[17]	JELOKA S, AKESH N B, SYLVESTER D, et al. A 28 nm configurable memory (TCAM/BCAM/SRAM) using push-rule 6T bit cell enabling logic-in-memory[J]. IEEE Journal of Solid-State Circuits, 2016, 51(4): 1009–1021. doi: 10.1109/JSSC.2016.2515510
[18]	AGA S, JELOKA S, SUBRAMANIYAN A, et al. Compute caches[C]. 2017 IEEE International Symposium on High Performance Computer Architecture, Austin, USA, 2017: 481–492. doi: 10.1109/hpca.2017.21.
[19]	ZHANG Yiqun, XU Li, DONG Qing, et al. Recryptor: A reconfigurable cryptographic cortex-M0 processor with in-memory and near-memory computing for IoT security[J]. IEEE Journal of Solid-State Circuits, 2018, 53(4): 995–1005. doi: 10.1109/JSSC.2017.2776302
[20]	ZENG Jianmin, ZHANG Zhang, CHEN Runhao, et al. DM-IMCA: A dual-mode in-memory computing architecture for general purpose processing[J]. IEICE Electronics Express, 2020, 17(4): 20200005. doi: 10.1587/elex.17.20200005
[21]	AGRAWAL A, JAISWAL A, ROY D, et al. Xcel-RAM: Accelerating binary neural networks in high-throughput SRAM compute arrays[J]. IEEE Transactions on Circuits and Systems I: Regular Papers, 2019, 66(8): 3064–3076. doi: 10.1109/TCSI.2019.2907488
[22]	YIN Shihui, JIANG Zhewei, SEO J S, et al. XNOR-SRAM: In-memory computing SRAM macro for binary/ternary deep neural networks[J]. IEEE Journal of Solid-State Circuits, 2020, 55(6): 1733–1743. doi: 10.1109/jssc.2019.2963616
[23]	LIU Rui, PENG Xiaochen, SUN Xiaoyu, et al. Parallelizing SRAM arrays with customized bit-cell for binary neural networks[C]. 2018 55th ACM/ESDA/IEEE Design Automation Conference (DAC), San Francisco, USA, 2018: 1–6. doi: 10.1109/DAC.2018.8465935.
[24]	KANG Mingu, GONUGONDLA S K, PATIL A, et al. A multi-functional in-memory inference processor using a standard 6T SRAM array[J]. IEEE Journal of Solid-State Circuits, 2018, 53(2): 642–655. doi: 10.1109/JSSC.2017.2782087
[25]	BELLOVIN S M. Frank miller: Inventor of the one-time pad[J]. Cryptologia, 2011, 35(3): 203–222. doi: 10.1080/01611194.2011.583711
[26]	SOBTI R and GEETHA G. Cryptographic hash functions: A review[J]. IJCSI International Journal of Computer Science Issues, 2012, 9(2): 461–479.
[27]	SEOK B, PARK J, and PARK J H. A lightweight hash-based blockchain architecture for industrial IoT[J]. Applied Sciences, 2019, 9(18): 3740. doi: 10.3390/app9183740
[28]	SZE V, CHEN Y H, YANG T J, et al. Efficient processing of deep neural networks: A tutorial and survey[J]. Proceedings of the IEEE, 2017, 105(12): 2295–2329. doi: 10.1109/JPROC.2017.2761740
[29]	LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278–2324. doi: 10.1109/5.726791
[30]	SIMONYAN K and ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv: 1409.1556, 2014.
[31]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR'16), Las Vegas, USA, 2016: 770–778. doi: 10.1109/CVPR.2016.90.
[32]	COURBARIAUX M, HUBARA I, SOUDRY D, et al. Binarized neural networks: Training deep neural networks with weights and activations constrained to +1 or –1[J]. arXiv preprint arXiv: 1602.02830, 2016.
[33]	RASTEGARI M, ORDONEZ V, REDMON J, et al. XNOR-Net: ImageNet classification using binary convolutional neural networks[C]. The 14th European Conference on Computer Vision, Amsterdam, The Netherlands, 2016: 525–542. doi: 10.1007/978-3-319-46493-0_32.
[34]	FOLEY J D, VAN F D, VAN DAM A, et al. Computer Graphics: Principles and Practice[M]. Boston: Addison-Wesley Professional, 1996: 1–1175.
[35]	WAN Yi and XIE Qisong. A novel framework for optimal RGB to grayscale image conversion[C]. The 2016 8th International Conference on Intelligent Human-Machine Systems and Cybernetics (IHMSC), Hangzhou, China, 2016: 345–348. doi: 10.1109/IHMSC.2016.201.
[36]	PRATT W K. Introduction to Digital Image Processing[M]. Boca Ration: CRC Press, 2013: 1–756.