%0 Journal Article
%T Energy-Efficient Stochastic Computing with Superparamagnetic Tunnel Junctions
%V 13
%N 3
%P 034016
%X Most computing schemes that employ superparamagnetic tunnel junctions control them with analog currents, leading to substantial Ohmic losses and requiring digital-to-analog converters. Here the authors forego current control and embed these junctions in digital circuits to produce programmable randomness, which leads to a neural network that can recognize handwritten digits at only 150 nJ per inference. This energy efficiency is made possible by the general insight that, while nanodevices provide useful dynamics for innovative computing, understanding their integration with digital logic systems is crucial to developing viable applications.
%G en
%J Physical Review Applied
%A Daniels, M.W.
%A Madhavan, A.
%A Talatchian, P.
%A Mizrahi, A.
%A Stiles, M.-D.
%D 2020/03/05