Machine Learning Based Delta Sigma Modulator Using Memristor for Neuromorphic Computing

This paper presents a memristor-based delta sigma modulator that mimics a human brain neuron for artificial neural network training. Memristive synapses are proposed in this work that can be used as weights. In recent works, memristive components have been proposed to be used in ADC and DAC. Such A-to-D converters can be trained using machine learning algorithms to get optimized results for speed and power. This paper describes the design of a 1st-order Sigma-delta ADC using a memristor working with a DC supply voltage of 1.8 V. The proposed ADC can be used for image sensors in augmented reality (AR) and virtual reality (VR) applications. The ADC consumes 24 μW of power at a sampling frequency of 100 kHz with a figure of merit of 3.6 fJ/step. It is ideal for high-resolution, low-frequency applications due to its highly linear characteristics resulting from a single-bit linear quantizer and oversampling techniques.