Behavioral Modelling of Analog Computing Circuit for a Synapse of any Neuron in Artificial Neural Networks

Analog implementation and behavioral modelling of neural network computing for artificial intelligence is demonstrated in this work. The main advantages of this solution are the low count of semiconductor devices and low power consumption. The calculations are performed in real time and not with clock cycles, as compared to the digital approaches. A behavioral model of an analog multiplying circuit for a synapse of any neuron in artificial neural networks was developed in the paper to demonstrate the proof of concept and simulate its operation. The operation of the analog circuits resulted in very good concordance with the mathematical formula it is supposed to implement. The application of the behavioral model concerning the technological variation of parameters like the MOS transistors threshold voltage resulted in demonstrating a low margin of error. In conclusion, this circuit proves to be very suitable for neural networks in real-time, low power and low device count operations.