1. FPGA-based small-world spiking neural network with anti-interference ability under external noise.
- Author
-
Guo, Lei, Liu, Yongkang, Wu, Youxi, and Xu, Guizhi
- Subjects
- *
NEUROPLASTICITY , *SPEECH perception , *GATE array circuits , *NOISE , *LARGE-scale brain networks , *NEURAL circuitry - Abstract
Neuromorphic hardware has become hotspot in the field of brain-like computing due to its advantages. However, the presence of external noise imposes challenges with respect to maintaining normal function of neuromorphic hardware. Biological brains have self-adaptability to external noise, meaning that a brain-like hardware with bio-plausibility can be expected to improve robustness. The purpose of this paper is to implement a highly fitted brain-like hardware with anti-interference ability (AIA) while preserving bio-plausibility. We propose a method of implementing a small-world spiking neural network (SWSNN) with bio-plausibility based on a field-programmable gate array (FPGA), in which the nodes are Izhikevich neuron modules, the edges are synaptic plasticity modules, and the topology is a small-world network. Then, the AIAs of the FPGA-based SNNs with different external noises are evaluated by two anti-interference indices. Further, taking a speech recognition task as the case study, the AIAs of these FPGA-based SNNs are verified in application. Finally, the AIA mechanism of the FPGA-based SNNs is discussed. Our results demonstrate that: (i) In the FPGA-based SWSNN, the FPGA-based Izhikevich neuron modules and the synaptic plasticity modules highly fit to the corresponding simulation results, and the topology conforms to the small-world property of human functional brain networks. (ii) Based on two anti-interference indices, the FPGA-based SWSNN outperforms the FPGA-based SNNs with other topologies, which is further verified by the speech recognition accuracy. (iii) Our discussions hint that the synaptic plasticity is intrinsic factor of the AIA, and the topology is a factor affecting the AIA. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF