Computing with networks of spiking neurons on a biophysically motivated floating-gate based neuromorphic integrated circuit
Computing with networks of spiking neurons on a biophysically motivated floating-gate based neuromorphic integrated circuit
复制标题
DOI：
10.1016/j.neunet.2013.02.011
复制
发表时间：
2013-09-01
期刊：
影响因子：
7.8
通讯作者：
Hasler, P.
中科院分区：
文献类型：
--
作者：
Brink, S.;Nease, S.;Hasler, P.
给出了在一种新型神经形态集成电路上进行的几个尖峰网络实验的结果。讨论了这些网络的计算意义，包括任意时空模式的生成和识别、赢家通吃竞争、节奏输出的稳定生成和易失性存储器等应用。书中还提到了与真实生物神经系统行为的类比。从计算效率的角度讨论和比较了实现相同计算的备选方案，得出结论：在神经形态硬件上实现神经网络比在传统数字硬件上实现模型方程的数值积分具有更高的功耗效率。(C)2013爱思唯尔有限公司。保留所有权利。
Results are presented from several spiking network experiments performed on a novel neuromorphic integrated circuit. The networks are discussed in terms of their computational significance, which includes applications such as arbitrary spatiotemporal pattern generation and recognition, winner-take-all competition, stable generation of rhythmic outputs, and volatile memory. Analogies to the behavior of real biological neural systems are also noted. The alternatives for implementing the same computations are discussed and compared from a computational efficiency standpoint, with the conclusion that implementing neural networks on neuromorphic hardware is significantly more power efficient than numerical integration of model equations on traditional digital hardware. (C) 2013 Elsevier Ltd. All rights reserved.