Abstract

This paper presents a 2-D programmable dendritic neuron array consisting of a 3 x 32 dendritic compartment array and a 1 x 32 somatic compartment array. Each dendritic compartment contains two types of regenerative nonlinearities: a NMDA synaptic nonlinearity and a dendritic spike nonlinearity. The chip supports the programmability of local synaptic weights and the configuration of dendritic morphology for individual neurons through the address-event representation protocol. Neurons can be stimulated and recorded using the same protocol. A novel local cable circuit between neighboring compartments allows one to construct different dendritic morphologies. This chip provides a hardware platform for studying the network behavior of neurons with active dendrites and for investigating the role of different dendritic morphologies in neuronal computation. Based on experimental results from a chip fabricated in a 4-metal, 2-poly, 0.35 mu m CMOS technology, this work shows one instance of how dendritic nonlinearities can contribute to neuronal computation, that is, the dendritic spike mechanism can dynamically reduce the mismatch-induced coefficient of variation of the somatic response amplitude from about 40% to 3.5%, and the response timing jitter by a factor of 2.