Characterization of HfOx 1T1R Memristors for Analog Programming

One of the greatest challenges for neuromorphic hardware is the implementation of compact, reliable and programmable synaptic connections. The emergence of nanoscale memristors presented a very promising device to mimic biological synapses. In recent years, hybrid CMOS-memristors technologies allowed for implementing neuromorphic chips. However, currently available memristors present some limitations to be programmed with analog conductance values, showing stochastic behavior during programming and relaxation effect after reaching the desired value. In this work, we have applied an iterative programming algorithm to target 7 different conductance levels in a 1T1R crossbar fabricated in 130nm CMOS technology, measuring both the stochasticity of the devices and the relaxation effect.