Hardware Modelling and Testing of a Rotating Neuron Reservoir

Rotating neural networks (RNRs) have been proven very efficient in executing in-memory computing tasks, while their potential to be implemented in full hardware massively reduces their power consumption. This paper reports the hardware modelling and simulations of an RNR using Cadence virtuoso tools, demonstrating the performance using the Mackey–Glass (MG) chaotic mathematical series for prediction. The readout layer calculations were performed using MATLAB, and the results were compared with the respective full MATLAB ideal algorithm. The Cadence/MATLAB hybrid algorithm has exhibited an NRMSE=0.042 for a reservoir size of 3 neurons, compared to the ideal MATLAB model, which gave a result of NRMSE=0.0267 for the same number of neurons.