CODEX: Stochastic Encoding Method to Relax Resistive Crossbar Accelerator Design Requirements

Abstract

A stochastic input encoding scheme (CODEX) is presented that aims to relax the digital-to-analog converter (ADC) design requirements in memristor crossbar systems. CODEX reduces the ADC input range by encoding the input bits using Bernoulli statistics so that the bit-line current distribution becomes a narrow Gaussian. By reducing ADC input range, CODEX can be used to reduce ADC power and area or increase ADC resolution for faster in-situ training. Besides input data encoding, CODEX includes probability thresholding for sparse input data as well as a random re-sampling method for dealing with ADC overflow. CODEX is evaluated on CIFAR-10 dataset image classification and reconstruction, sentiment classification, and audio classification. The results show an averaged 68.5% reduction in ADC power and 35.5% reduction in ADC area as well as 25.8% increase in in-situ training speed when applied to the state-of-the-art ISAAC and PUMA accelerators.

Design of an ultra-broadband terahertz absorber based on a patterned graphene metasurface with machine learning

“…29 ML works by extracting useful information from raw databases and outputting independent decisions to perform tasks. With the rise of ML technology, some models combined with ML have been applied to metasurface based optical devices, such as performance prediction of color filtering, 30 metalens, 31 digital encoding, 32,33 holographic imaging, 34 etc. In recent years, a number of studies have been proposed using ML regression models to predict absorption values at intermediate wavelengths, which can be effective in reducing the simulation time and costs.…”