Cognitive Systems with Stochastic Spintronics Devices – CogniSpin

This project aims at using magnetic RAM devices in an original probabilistic regime, to demonstrate low-power cognitive-type applications. In recent years, research on using memory nanodevices as “synapses” (the connections in the brain) has largely boomed, with studies both on the device and on the applications sides. This project will join this rich research dynamic, and bring a radically novel point of view that may transform this emerging field. The programming of many nanodevices, and in particular magnetic RAM (MRAM), has an intrinsic random character. Lowenergy programming pulses may program the device, but only with a finite probability. We propose to EXPLOIT this random aspect of short programming pulses to develop new ultra low power computing paradigms. Our research will focus on Spin Torque Transfer MRAMs. In these devices, the stochastic behavior is indeed controllable. Exploiting this behavior is a real reverse of way of thinking. Nondeterministic effects are normally dramatic for circuit applications, here they will allow us to perform learning with algorithms that would be complex to implement in another way. The scheme will be naturally lowpower since it involves short programming pulses. In this project, the term “synapses” is taken in its broadest meaning. We will aim bioinspired applications, where nanodevices are used similarly to biological synapses, and also other learning systems where synapses learn in ways inspired by machine learning. The final goal of this technology is to develop ultralow power embedded systems capable of extreme adaptation thanks to learning, and capable of processing natural data.

The project is interdisciplinary in nature and features different kinds of research. On the device side, we will characterize different kinds of magnetic tunnel junctions in regimes that are not typically explored, but are the best for synaptic-type applications. We will model the stochastic behavior and we will program compact (VerilogA-based) models for circuit simulations, and behavioral models for system-level simulations. These models will be made open source. In parallel, we will develop different kinds of learning rules that exploit the stochastic behavior. We will assess them through simulations of systems that use large numbers of nanodevices. For this purpose, a specific simulator will be developed. We will focus on demonstrating complex real life applications (image/video, sound, olfaction and robotics). We will evaluate the robustness of our approach to technology imperfections and unpredictable environments. We will design CMOS circuits to be associated with synaptic MRAMs and validate them by simulation. Finally, we will fabricate a small demonstrator with a small number of packaged MRAMs on printed circuit board that will demonstrate stochastic learning in practice and open the way to the realization of a full-size demonstrator with CMOS/MRAM integration in the future.

The project will benefit from an exceptional context. The coordinator has already led a pioneering an extremely successful preliminary study (“PEPS” project funded by CNRS), whose goal was to bring the preliminary results for the submission of the present ANR project. An excellent PhD student has already agreed to join the project. Finally, the project benefits from the exceptional expertise of IEF in nanoelectronics-based design, computer simulation and spintronics devices characterization.

The results of this project can have a strong social and economic impact. It is expected that a driver for future electronic devices will be ambient and cognitive intelligent devices that should simplify people’s everyday life, assist elderly and handicapped people, and help medical doctors provide tailored and timely health-care. They will require electronic systems, which can compute efficiently with the real-life data from sensors with minimum power consumption. Our novel computing paradigm can be a key tool to achieve such systems.