ANALOG CIRCUIT DESIGN OF SPIKING NEURAL NETWORKS

Abstract

Analog circuit design for Spiking Neural Networks (SNNs) is the engineering discipline that can be used to model biological neuron’s functionality by using electronic components. This approach bases its functionality on the analogue of SNNs and biochemical neurons that allows designing effective and reliable information systems. They are applied in complex pattern identification, adaptation, motor control, flexibility, immunity to noise, and self-repair mechanisms among others. LTSpice – an effective tool for circuit simulation, has been employed to consider specified neuron models and compare it. Inspired by the model of Hodgkin Huxley (HH) Neuron that generates an imitation of the ionic currents which traverse the neuronal membrane to create action potentials, a second-orderrderivative equation of a simpleeneuronnmodel was proposed by Izhikevich(2003))[6]. Circuit simulations in LTSpice and efficiency studies have been conducted on the implementation of these neuron models. A detailed study of simple model of spiking neuron, and reconfigurable analog version of the piecewise linear neuron model[8] with CCII components is done using AD844 and AD633 in LTSpice. LTSpice has been used to implement many kinds of neuron models to make an extensive analysis about them with special consideration to working with analog circuits for spiking neurons. Subsequently the implementations AD633 analog multipliers and AD844 operational amplifiers are used where these parts are necessary. The firing patterns they exhibit are those that are observed on biological neurons, and these circuits are meant to emulate. In order to enhance the biomedical realism and accuracy of the meant spiking signal following further development, the Integrate-and-Fire neuron concept and, in addition, a membrane recovery variable, will be incorporated. When the second-order derivative equation is in use, the shape is more like the spikes seen in cortical neurons. The first part of the equation is crucial since most mechanisms of spike formation are characterized by the increase in the membrane potential. We have demonstrated the analog circuit of the proposed second-order derivative equation which was earlier proposed in the form of differential equation using AD844 and AD633. It is demonstrated that this second order derivative equation is indeed effective by constructing the neuron models in LTSpice. Also the implementation of CCII using AD844 that constitutes a major part of the membrane (iv) (9 circuit of the neuron has enhanced the area efficiency of the existing reconfigurable analog version of piecewise linear model[8] up to a great extent making its practical implementation more cost effective. Theoretical studies of the analog neuron model depend profoundly on the CCII that is implemented using the help of AD844. It also enables minimal signal attenuation and high-speed net current transfer, which are crucial for emulating neuron functions.