Physical implementation of pair-based spike-timing-dependent plasticity

Rahimiazghadi, Mostafa, Al-sarawi, Said, Iannella, Nicolangelo, and Abbott, Derek (2011) Physical implementation of pair-based spike-timing-dependent plasticity. Australasian Physical & Engineering Sciences in Medicine, 34. 141. p. 572.

PDF (Abstract Only) - Published Version
Download (73kB) | Preview
View at Publisher Website:


Objective Spike-timing-dependent plasticity (STDP) is one of several plasticity rules which leads to learning and memory in the brain. STDP induces synapticweight changes based on the timing of the pre- and postsynaptic neurons. A neural network which can mimic the adaptive capability of biological brains in the temporal domain, requires the weight of single connections to be altered by spike timing. To physically realise this network into silicon, a large number of interconnected STDP circuits on the same substrate is required. This imposes two significant limitations in terms of power and area. To cover these limitations, very large scale integrated circuit (VLSI) technology provides attractive features in terms of low power and small area requirements. An example is demonstrated by (Indiveri et al. 2006). The objective of this paper is to present a newimplementation of the STDPcircuit which demonstrates better power and area in comparison to previous implementations.

Methods The proposed circuit uses complementary metal oxide semiconductor (CMOS) technology as depicted in Fig. 1. The synaptic weight can be stored on a capacitor and charging/discharging current can lead to potentiation and depression.

Results and Conclusion: HSpice simulation results demonstrate that the average power, peak power, and area of the proposed circuit have been reduced by 6, 8 and 15%, respectively, in comparison with Indiveri's implementation. These improvements naturally lead to packing more STDP circuits onto the same substrate, when compared to previous proposals. Hence, this new implementation is quite interesting for real-world large neural networks.

Item ID: 45715
Item Type: Article (Abstract)
ISSN: 1879-5447
Date Deposited: 20 Jul 2017 02:50
FoR Codes: 09 ENGINEERING > 0906 Electrical and Electronic Engineering > 090601 Circuits and Systems @ 100%
SEO Codes: 97 EXPANDING KNOWLEDGE > 970109 Expanding Knowledge in Engineering @ 100%
Downloads: Total: 239
Last 12 Months: 18
More Statistics

Actions (Repository Staff Only)

Item Control Page Item Control Page