|
|
|
| Hardware Implementation and Discharge Performance Comparison of Nerve Synapses Based on FPGA |
| Chen Kai1,2, Lu Mai1*, Yi Feihong1,2, Wang Chao1 |
1(Key Lab of Opto-Electronic Technology and Intelligent Control of Ministry of Education, Lanzhou Jiaotong University, Lanzhou 730070, China)
2(The School of Automatization & Electric Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China) |
|
|
|
|
Abstract At present, there are few studies on the hardware implementation of chemical synaptic, the use of FPGA chip to realize chemical synapses is of great value to the realization of neural network hardware. In this work, based on the DSP Builder model, the DSP Builder model of the chemical synapse was reasonably splited, and then each module was compiled and run in the software environment corresponding to the FPGA, and finally downloaded to the FPGA core chip, using the FPGA hardware to realize five kinds of chemical synapses based on different mechanisms. The correlation coefficient method was used to compare the pre-synaptic neuron action potential, post-synaptic neuron action potential and synaptic current amplitude between the simulation results and the hardware results in the same cycle. Five chemical synapses realized by hardware could better transmit action potentials, but each model consumed different resources. The internal multiplier resources consumed by Model 3 (69%) were about 2 times that of Model 5 (31%), indicating that the higher the mathematical complexity of the synapse model, the more multiplier resources it consumed. The comparison of the correlation coefficient method showed that the correlation of model 3 was the highest, which was 0.791 3, and the correlation coefficient of model 4 was the lowest, which was 0.693 5. Although Model 3 had the high mathematical complexity and hardware resource consumption, its performance was the best. The five synaptic models implemented by hardware could better represent the one-way transmission of chemical synapses. The model 5 had low hardware resource consumption and high correlation. The model 5 was recommended as the first choice for hardware realization of chemical synapses.
|
|
Received: 27 April 2018
|
|
|
|
|
|
[1] 王青云, 石霞, 陆启韶. 神经元耦合系统的同步动力学 [M]. 北京:科学出版社, 2008: 21-22.
[2] Mead CA. Analog VLSI and Neural Systems [M]. Boston: Addison-Wesley, 1989: 10-30.
[3] Mahowald M, Douglas R. A silicon neuron [J]. Nature, 1991, 354(6354): 515-518.
[4] Weinstein RK, Lee RH. Architectures for high-performance FPGA implementations of neural models [J]. Journal of Neural Engineering, 2006, 3(1): 21-34.
[5] Weinstein RK, Reid SM, Lee RH. Methodology and design flow for assisted neural-model implementations in FPGAs [J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2007, 15(1): 83-93.
[6] Theodore Y, Gert C. Analog VLSI biophysical neurons and synapses with programmable membrane channel kinetics [J]. IEEE Transcations on Biomedical Circuits and Systems, 2010, 4(3): 139-148.
[7] Marko N, Mayr C, Partzsch J, et al. Synapse dynamics in CMOS derived from a model of neurotransmitter release [C]//The 20th European Conference on Circuit Theory and Design (ECCTD). Linkoping: IEEE, 2011: 198-201.
[8] Pradyumna SG, Rathod S. Analysis of CMOS inhibitory synapse with varying neurotransmitter concentration, reuptake time and spread delay [C] //2015 International Symposium on VLSI Design and Test (VDAT-2015). Ahmedabad: IEEE, 2015: 1-5.
[9] Sadia A, Rezaul H. A VLSI circuit emulation of chemical synaptic transmission dynamics and postsynaptic DNA transcription [J]. IEEE Transactions on Very Large Scale Integration(VLSI) Systems, 2016, 24(2): 678-691.
[10] 张荣华, 王江. FPGA在生物神经系统模型仿真中的应用 [J]. 计算机应用研究, 2011, 28(8): 2949-2953.
[11] 张荣华, 王江. 神经元网络的 FPGA硬件仿真方法 [J]. 计算机应用研究, 2011, 28(10): 3707-3710.
[12] 李宏伟, 吴庆祥. 脉冲神经网络中神经元突触的硬件实现方案 [J]. 计算机系统应用, 2014, 23(2): 17-21.
[13] 王金龙, 逯迈, 胡延文, 等. 基于现场可编程门阵列的Hodgkin-Huxley模型神经元动作电位的数值模拟功能实现 [J]. 生物医学工程学杂志, 2015, 32(6): 1302-1309.
[14] 闻佳, 逯迈, 陈小强, 等. 基于HH 模型的神经元网络的数值模拟与FPGA实现 [J]. 航天医学与医学工程, 2017, 30(1): 38-45.
[15] Hodgkin AL, Huxley AF. A quantitative description of membrane current and its application to conduction and excitation in nerve [J]. Journal of Physiology, 1952, 117(4): 500-544.
[16] Hodgkin AL, Huxley AF. Currents carried by sodium and potassium ions though the membrane of the giant axon of Loligo [J]. Journal of Physiology, 1952, 116(4): 449-472.
[17] Hodgkin AL, Huxley AF. The components of membrane conductance in the giant axon of Loligo [J]. Journal of Physiology, 1952, 116(4): 473-496.
[18] Hodgkin AL, Huxley AF. The dual effect of membrane potential on sodium conductance in the giant axon of Loligo [J]. Journal of Physiology, 1952, 116(4): 497-506.
[19] Hansel D, Sompolinsky H. Synchronization and computation in a chaotic neural network [J]. Physical Review Letters, 1992, 68(5): 718-721.
[20] Destexhe A, Mainen Z, Sejnowski T. An efficient method for computing synaptic conductances based on a kinetic model of receptor binding [J]. Neural Computation, 1994, 6(1): 14-18.
[21] Wang Xiao-Jing, Buzsáki G. Gamma oscillation by synaptic inhibition in a hippocampal inter neuronal network model [J]. The Journal of Neuroscience, 1996, 16(20): 6402-6413.
[22] Rabinovich MI, Abarbanel HD, Huerta R, et al. Self-regularization of chaos in neural systems: experimental and theoretical results [J]. IEEE Transaction Circuits and Systems, 1997, 44(10): 997-1005.
[23] Savtchenko LP. Bilateral processing in chemical synapses with electrical ‘ephaptic’ feedback:a theoretical model [J]. Mathematical Biosciences, 2007, 207(1): 113-137.
[24] 夏阳, 尧德中. 神经信息学基础 [M]. 成都:电子科技大学出版社, 2015: 20-43.
[25] 张荣华, 王江. 复杂物理模型仿真的多进程流水线方法 [J]. 计算机应用研究, 2011, 28(10): 3694-3698.
[26] 周金治, 唐肖芳. 基于相关系数分析的脑电信号特征选择 [J]. 生物医学工程学杂志, 2015, 32(4): 735-739.
[27] 尹宁. 基于脑电的磁刺激穴位复杂脑功能网络研究 [D]. 天津: 河北工业大学, 2013.
[28] 常小龙, 丁国良, 娄建安. 神经元网络同步放电的抗扰特性[J]. 上海交通大学学报, 2014, 48(10): 1485-1490.
[29] Glyzin SD, Yu A, Kolesov N, et al. On a method for mathematical modeling of chemical synapses [J]. Differential Equations, 2013, 49(10): 1193-1210.
[30] 刘利华, 刘深泉. 化学突触传递的数值模拟[J]. 北京生物医学工程, 2011, 30(2): 117-119.
[31] Yang Shuangming, Wang Jiang, Deng Bin,et al. Cost-efficient FPGA implementation of basal ganglia and their Parkinsonian analysis[J]. Neural Networks, 2015, 71: 62-75.
[32] 谢明文. 关于协方差、相关系数与相关性的关系 [J]. 数理统计与管理, 2004(3): 33-36. |
|
|
|
