NEUROMORPHIC VLSI REALIZATION OF THE HIPPOCAMPAL FORMATION AND THE LATERAL SUPERIOR OLIVE

Abstract

In this work, the focus is on realizing the function of the hippocampal formation (HF) and the lateral superior olive (LSO) in electronic circuits.

The first major contribution of this dissertation is to realize the function of the HF in silicon. This was based on the GRIDSmap model and the Bayesian integration. For this, two novel circuits were designed and integrated with others. The first circuit was that of a Bayesian integration synapse which can perform Bayesian integration at the single neuron level. The second circuit was that of a velocity integrator which is so compact that it can enable integration of the entire system on a single chip compared to its predecessors which would have needed 27 chips!

However, since the computational neuroscience models of the hippocampal place cells do not explain all the characteristics observed empirically, a novel model for the place cells, based on the sensori-motor integration of inputs is proposed. This is the second major contribution of this thesis.

The third major contribution is to demonstrate a VLSI system which can perform azimuthal localization based on population response of the LSO. This system was based on the Reed and Blum's model of the LSO. For this, a novel circuit of a second order synapse and that of a conductance neuron was designed and integrated with other circuits. This synapse circuit can produce an output current whose peak is delayed and is proportional to the number of inputs it receives.

The HF is thought to aid in spatial navigation and the LSO is thought to be involved in azimuthal localization of sounds both of which are useful for autonomous robotic spatial navigation. Hence, silicon realization of these two will be useful in robotics which is an area of interest for the neuromorphic engineers.