Current students

The research project explores the impact of non idealities in crosspoint memory arrays, focusing on mitigation and compensation techniques. Indeed, crosspoint arrays are key players in the expansion of In-Memory Computing paradigm, thanks to its integration density, the low power requested and the intrinsic capability of performing matrix-vector multiplication (MVM). However, with increasing the size of arrays, several non idealities become impactful. Among the most significant ones we can enlist the parasitic resistances along wires (IR drop), the variability of the adopted emerging memory devices and the resistance shown by the driving and read-out stages. The aim of the research project is to obtain general purpose solutions capable of rejecting the various parasitisms, in order to unleash the potential inherent to the massive parallelism of crosspoint architecture. In this framework, a special attention is given to the application of crosspoint arrays as deep learning accelerators, thus with fixed set of conductive weights. 

The research project will insist on finding the proper combination of device, operating point, computing architecture and compensation techniques to demonstrate a final application (e.g. binary neural network for image classification). It would be interesting also to explore different learning paradigms, to better exploit the physic characteristics of the devices in an application-oriented context, and to merge the acquired knowledge in memristive devices with neuromorphic architectures based on CMOS technology.