Litcius/Paper detail

Optimized programming algorithms for multilevel RRAM in hardware neural networks

Valerio Milo, Francesco Anzalone, Cristian Zambelli, Eduardo Pérez, Mamathamba Kalishettyhalli Mahadevaiah, Óscar G. Ossorio, P. Olivo, Christian Wenger, Daniele Ielmini

202133 citationsDOIOpen Access PDF

Abstract

A key requirement for RRAM in neural network accelerators with a large number of synaptic parameters is the multilevel programming. This is hindered by resistance imprecision due to cycle-to-cycle and device-to-device variations. Here, we compare two multilevel programming algorithms to minimize resistance variations in a 4-kbit array of HfO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> RRAM. We show that gate-based algorithms have the highest reliability. The optimized scheme is used to implement a neural network with 9-level weights, achieving 91.5% (vs. software 93.27%) in MNIST recognition.

Topics & Concepts

MNIST databaseArtificial neural networkComputer scienceResistive random-access memoryAlgorithmKey (lock)Reliability (semiconductor)SoftwareParallel computingComputer hardwareComputer engineeringMachine learningProgramming languageEngineeringElectrical engineeringOperating systemVoltageQuantum mechanicsPhysicsPower (physics)Advanced Memory and Neural ComputingFerroelectric and Negative Capacitance DevicesSemiconductor materials and devices
Optimized programming algorithms for multilevel RRAM in hardware neural networks | Litcius