Energy-Efficient Single Transistor Neuron for Reconfigurable Threshold Logic and Image Classification

This work explores the applications of germanium based energy-efficient single transistor neuron for reconfigurable threshold logic. The implementation is done with the help of single device without using any external circuitry. The nature of bias current applied at the drain of the MOSFET, and the weights associated with the inputs decide the logic that is to be implemented. A three-layer spiking neural network has also been designed using the proposed neuron to verify image classification efficiency of the network for multiclass MNIST dataset. The maximum accuracy of 96.88% was achieved from the designed network.