Microcontroller based platform for the experimental analysis of noise-induced spiking activity in artificial neural networks

Experimental and theoretical progresses in the field of neuroscience have been paralleled by outstanding advancements in the field of microelectronics and nanotechnologies in recent years, thus boosting the development of neuromorphic computing systems. In the years it has been pointed out that despite the stochasticity introduced by noise and fluctuations in the brain, it looks that neural information processing is not only noise tolerant, but even more noise assisted. Therefore, the effect of noise as a computational resource has been introduced in ITRS Emerging Devices Chapter starting from 2013. [http://www.itrs2.net] In this thesis such a resource for computation and learning in networks of spiking neurons will be introduced based upon brain inspired architectures to be applied and explored in signal networks of spiking neurons. In this work we develop a neural network of artificial spiking network operating at millisecond range and connected by a simplified version of excitatory synapses and specific inhibitory synapses designed to be compatible with specifications of the neurons. The network can host up to 16 neurons enabling us to mimic small micro columns and small networks, and it is equipped with switchable noise generators designed for current injection in the neurons. The experimental setup has been integrated with ArduinoDue acquisition system in order to grant parallel acquisition of up to 16 channels, 115200 baud rate and software for spike timing measurements and statistics. By using the experimental setup a number of different network configurations has been explored to address the role of loops in small layered networks and of noise spectrum.