Silicon Controlled Rectifier based neuronal model for low power applications

Abstract

Human brain processes the information and makes decision efficiently by consuming very low amount of power. Graphics processing unit (GPUs) and Field Programmable Gate Arrays (FPGAs) are being used nowadays for low power portable applications but power consumption is not yet comparable to biological brain. In addition to less energy dissipation, there is requirement of mimicking real neuron with simple neuronal structure so that it can cater the need for implementation of less complex spiking neural networks. In an attempt to fulfill both the above requirements, a simple leaky integrate and fire neuron based artificial neuron has been proposed where the tunability of parameters like discharge time of capacitor is tunable. The tunability of parameters also leads to flexibility of design. The proposed design is novel as the design consists of basic electronic components such as SCR, resistors, diodes and capacitors which makes design simple. The design has been simulated in LTspice and simulation results shows the spiking activity of neuron when a sinusoidal signal acts as input from the source. The proposed design consumes power in nJ and can further be reduced by scaling the electronic components thus opens the path for large scale spiking neuron networks implementations for various applications such as pattern recognition.