将SNN部署到类脑处理器的映射优化算法研究

doi:10.3778/j.issn.1002-8331.2407-0164

摘要/Abstract

摘要： 近年来，具有生物合理性和能效优势的脉冲神经网络（SNN）受到广泛关注。然而，目前在类脑处理器上部署SNN的映射方案存在通信延迟高、拥塞严重、能耗高和节点连接性不足等问题，从而削弱了其实用性和执行效率。为解决这些问题，提出了基于KL（Kernighan-Lin）和波尔兹曼退火差分进化（Boltzmann anneal differential evolution，BADE）的改进部署算法，用于将SNN映射到资源受限的类脑处理器上。该算法包括两个步骤：分区和映射。在分区阶段，通过在递归KL算法中引入全局优化策略（GRBKL）来最小化集群之间的通信延迟；在映射阶段，提出利用吸引子导向的BADE算法（BAFDE）寻找最小化通信延迟和最大拥塞的分配方式。用五个SNN实例对该算法进行了评估，结果表明，与SNEAP和SpiNeMap等方法相比，所提出的算法显著降低了通信延迟（分别降低了55.41%和94.73%）和最大拥塞（分别降低了81.27%和97.79%）。

关键词: 脉冲神经网络（SNN）, 类脑处理器, 启发式算法, 片上网络（NOC）

Abstract: In recent years, spiking neural network (SNN), which offers biological plausibility and energy efficiency, has garnered widespread attention. However, current mapping schemes for deploying SNN on neuromorphic processors face issues such as high communication delays, severe congestion, high energy consumption, and insufficient node connectivity, thereby reducing their practicality and execution efficiency. To address these problems, an improved deployment algorithm based on KL (Kernighan-Lin) and BADE(Boltzmann anneal differential evolution) has been proposed to map SNN onto resource-constrained neuromorphic processors. This algorithm comprises two steps： partitioning and mapping. In the partitioning phase, a global optimization strategy (GRBKL) is introduced into the recursive KL algorithm to minimize communication delay between clusters. In the mapping phase, an attractor-guided BADE algorithm(BAFDE)is proposed to find an allocation scheme that minimizes communication delay and maximum congestion. Finally, the algorithm is evaluated using five SNN instances, and the results show that, compared to methods like SNEAP and SpiNeMap, the proposed algorithm significantly reduces communication delay (by 55.41% and 94.73%, respectively) and maximum congestion (by 81.27% and 97.79%, respectively).

Key words: spiking neural network(SNN), brain-inspired processor, heuristic algorithm, network-on-chip(NOC)

陈奥新, 陈亮, 李千鹏, 王智超, 徐东君. 将SNN部署到类脑处理器的映射优化算法研究[J]. 计算机工程与应用, 2025, 61(11): 156-165.

CHEN Aoxin, CHEN Liang, LI Qianpeng, WANG Zhichao, XU Dongjun. Research on Mapping Optimization Algorithms for Deploying SNN to Brain-Inspired Processors[J]. Computer Engineering and Applications, 2025, 61(11): 156-165.

参考文献

[1] 苏赋, 吕沁, 罗仁泽. 基于深度学习的图像分类研究综述[J]. Telecommunications Science, 2019, 35(11): 58-74.
SU B, LYU Q, LUO R Z. Review of image classification based on deep learning[J]. Telecommunications Science, 2019, 35(11): 58-74.
[2] 奚雪峰, 周国栋. 面向自然语言处理的深度学习研究[J]. 自动化学报, 2016, 42(10): 1445-1465.
XI X F, ZHOU G D. A Survey on deep learning for natural language processing[J]. Acta Automatica Sinica, 2016, 42(10): 1445-1465.
[3] DAVIES M, SRINIVASA N, LIN T H, et al. Loihi: a neuro-morphic manycore processor with on-chip learning[J]. IEEE Micro, 2018, 38(1): 82-99.
[4] ORCHARD G, FRADY E P, RUBIN D B D, et al. Efficient neuromorphic signal processing with Loihi 2[C]//Proceedings of the IEEE Workshop on Signal Processing Systems, 2021: 254-259.
[5] MEROLLA P A, ARTHUR J V, ALVAREZ-ICAZA R, et al. A million spiking-neuron integrated circuit with a scalable communication network and interface[J]. Science, 2014, 345: 668-673.
[6] FURBER S B, GALLUPPI F, TEMPLE S, et al. The spinnaker project[J]. Proceedings of the IEEE, 2014, 102(5): 652-665.
[7] PEI J, DENG L, MA C, et al. Multi-grained system integration for hybrid-paradigm brain-inspired computing[J]. Science China Information Sciences, 2023, 66(4): 272-285.
[8] KERNIGHAN B W, LIN S. An efficient heuristic procedure for partitioning graphs[J]. The Bell System Technical Journal, 1970, 49(2): 291-307.
[9] LI H, WANG H, WANG L, et al. A modified Boltzmann annealing differential evolution algorithm for inversion of directional resistivity logging-while-drilling measurements [J]. Journal of Petroleum Science and Engineering, 2020, 188: 106916.
[10] DAYAN P, ABBOTT L F. Theoretical neuroscience: computational and mathematical modeling of neural systems[M]. Cambridge: MIT Press, 2005.
[11]张翰林. 众核类脑计算嵌入式系统设计与实现[D]. 天津: 天津大学, 2021.
ZHANG H L. Design and implementation of multi-core brain-inspired computing embedded system[D]. Tianjin: Tianjin University, 2021.
[12] 杜加琴. NOC 基于 2D MESH 的路由算法的比较[J]. 电子技术, 2012, 39(3): 9-11.
DU J Q. Comparison of 2D MESH based routing algorithm in NOC[J]. Electronic Technology, 2012, 39(3): 9-11.
[13] LIN C K, WILD A, CHINYA G N, et al. Mapping spiking neural networks onto a manycore neuromorphic architecture[C]//Proceedings of the 39th ACM SIGPLAN Conference on Programming Language Design and Implementation，2018：78-89.
[14] AMIR A, DATTA P, RISK W P, et al. Cognitive computing programming paradigm: a corelet language for composing networks of neurosynaptic cores[C]//Proceedings of the International Joint Conference on Neural Networks, 2013: 1-10.
[15] GALLUPPI F, DAVIES S, RAST A, et al. A hierachical configuration system for a massively parallel neural hardware platform[C]//Proceedings of the 9th Conference on Computing Frontiers, 2012: 183-192.
[16] DENG L, WANG G, LI G, et al. Tianjic: a unified and scalable chip bridging spike-based and continuous neural computation[J]. IEEE Journal of Solid-State Circuits, 2020, 55(8): 2228-2246.
[17] SONG S, CHONG H, BALAJI A, et al. DFSynthesizer: dataflow-based synthesis of spiking neural networks to neuromorphic hardware[J]. ACM Transactions on Embedded Computing Systems, 2022, 21(3): 1-35.
[18] BALAJI A, DAS A. A framework for the analysis of throughput-constraints of SNNs on neuromorphic hardware[C]//Proceedings of the IEEE Computer Society Annual Symposium on VLSI, 2019: 193-196.
[19] VARSHIKA M L, BALAJI A, CORRADI F, et al. Design of many-core big little μBrains for energy-efficient embedded neuromorphic computing[C]//Proceedings of the Design, Automation & Test in European Conference & Exhibition , 2022: 1011-1016.
[20] ZHANG J, HUO D, ZHANG J, et al. ANP-I: a 28-nm 1.5-pJ/SOP asynchronous spiking neural network processor enabling Sub-0.1 μJ/Sample on-chip learning for edge-AI applications[J]. IEEE Journal of Solid-State Circuits 2024, 59(8): 2717-2729.
[21] DAS A, WU Y, HUYNH K, et al. Mapping of local and global synapses on spiking neuromorphic hardware[C]// Proceedings of the Design, Automation & Test in European Conference & Exhibition, 2018: 1217-1222.
[22] EBERHART R, KENNEDY J. A new optimizer using particle swarm theory[C]//Proceedings of the 6th International Symposium on Micro Machine and Human Science, 1995: 39-43.
[23] BALAJI A, DAS A, WU Y, et al. Mapping spiking neural networks to neuromorphic hardware[J]. IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2019, 28 (1): 76-86.
[24] JI Y, ZHANG Y H, LI S C, et al. NEUTRAMS: neural network transformation and co-design under neuromorphic hardware constraints[C]//Proceedings of the 49th Annual IEEE/ACM International Symposium on Microarchitecture, 2016: 1-13.
[25] LI S, GUO S, ZHANG L, et al. SNEAP: a fast and efficient toolchain for mapping large-scale spiking neural network onto NoC-based neuromorphic platform[C]//Proceedings of the 2020 on Great Lakes Symposium on VLSI, 2020: 9-14.
[26] STEINBRUNN M, MOERKOTTE G, KEMPER A. Heuristic and randomized optimization for the join ordering problem[J]. The VLDB Journal, 1997, 6: 191-208.
[27] 高鹰, 谢胜利. 基于模拟退火的粒子群优化算法[J]. 计算机工程与应用, 2004, 40(1): 47-50.
GAO Y, XIE S L. Particle swarm optimization algorithms based on simulated annealing[J]. Computer Engineering and Applications, 2004, 40(1): 47-50.
[28] XIAO C, CHEN J, WANG L. Optimal mapping of spiking neural network to neuromorphic hardware for edge-AI[J]. Sensors, 2022, 22(19): 7248.
[29] AARTS E H L, KORST J H M. Boltzmann machines as a model for parallel annealing[J]. Algorithmica, 1991, 6: 437-465.
[30] 刘洪普, 侯向丹. 模拟退火算法中关键参数的研究[J]. 计算机工程与科学, 2008, 30(10): 55-57.
LIU H P, HOU X D. Research on the key parameters in the simulated annealing algorithm[J]. Computer Engineering & Science, 2008, 30(10): 55-57.
[31] 陈颖, 林盈, 胡晓敏. 多种群多策略的并行差分进化算法[J]. 计算机科学与探索, 2014, 8(12): 1502-1510.
CHEN Y, LIN Y, HU X M. Parallel differential evolution with multi-population and multi-strategy[J]. Journal of Frontiers of Computer Science & Technology, 2014, 8(12): 1502-1510.
[32] XIAO C, HE X, YANG Z, et al. Hierarchical mapping of large-scale spiking convolutional neural networks onto resource-constrained neuromorphic processor[J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2024, 43(5): 1442-1455.
[33] JAIN R, RAMAKRISHNAN K, CHIU D M. Congestion avoidance in computer networks with a connectionle ss network layer[C]//Proceedings of the Computer Networking Symposium, 1998: 134-143.
[34] 蒋万春, 李昊阳, 陈晗瑜, 等. 网络拥塞控制方法综述[J]. 软件学报, 2024, 35(8): 3952-3979.
JIANG W C, LI H Y, CHEN H Y, et al. Survey on network congestion control algorithms[J]. Journal of Software, 2024, 35(8): 3952-3979.