Neuromorphic Data Converters Using Memristors

Abstract

Data converters are ubiquitous in data-abundant mixed-signal systems, where they are heterogeneously distributed across the analog–digital interface. Unfortunately, conventional CMOS data converters trade off speed, power, and accuracy. Therefore, they are exhaustively customized for special purpose applications. Furthermore, intrinsic real-time and post-silicon variations dramatically degrade their performance along with the technology downscaling. Using machine learning techniques and neuromorphic computing, these issues can be overcome. This chapter presents four-bit neuromorphic analog-to-digital (ADC) and digital-to-analog (DAC) converters using memristors that are trained using the stochastic gradient descent algorithm in real-time to autonomously adapt to different design and application specifications, including multiple full-scale voltages, sampling frequencies, number of resolution bits, and quantization scale. Theoretical analysis, as well as simulation results, show the collective resilient properties of our converters in application reconfiguration, logarithmic quantization, mismatches calibration, noise tolerance, and power optimization. Furthermore, large-scale challenges are discussed and solved by leveraging mixed-signal architectures, such as pipelined ADC. These ADC and DAC designs break through the tremendous speed-power-accuracy tradeoff in conventional data converters and enable a general-purpose application architecture with valuable results for neuromorphic computing.