Memory is at the center of modern electronics. However, the way that it is used is far from optimal: considerable energy is wasted due to the von Neumann bottleneck inherent to the design of computers. This postdoc project aims at addressing this situation by designing “intelligent memories” that work much more like Human memory than computer memory. This takes special meaning as resistive memories such as memristive devices are currently emerging and are widely believed to express their full potential in bioinspired architectures rather than conventional memories1. The postdoc will investigate theoretically and experimentally how such memories can implement naturally ideas from Bayesian reasoning and therefore allow true intelligent memory design. In particular, he/she will investigate how some typical memristive device concerns (crosstalk, stochasticity) can become beneficial in this context.
This effort is part of a large project that recently received 1.5M€ of funding through the European Research Council (ERC) for the invention of natively intelligent memory.
Scientific and technical work: The postdoc will draw a parallel between memristive devices crossbars and models of reasoning (based on Bayesian theory) that can be used to implement intelligent memory. In particular, he/she will investigate how memristive devices can store conditional probability distributions and implement an inference mechanism.
The postdoc project involves both theoretical studies and experiments. The ratio between computer and experimental work can be adapted to the candidate’s background. The postdoc will simulate design and simulate basic blocks for Bayesian inference. The methodology will be based on Matlab and a circuit-level simulator (Cadence). Experiments involve characterization of the statistical properties of memristive devices crossbars. The project is therefore ideal for an experimentalis who wants to open up to simulation work, or for a theorist/designer who wants to open up to experimental work.
The project will be located at the Centre de Nanosciences et de Nanotechnologies (https://sites.google.com/site/damienquerlioz/) of the Université Paris-Sud. The Centre is located in a beautiful green campus in the Parisian suburbs. Downtown Paris is accessible by frequent RER B train. The project includes very substantial collaboration with Fabien Alibart at Institut d’Electronique de Microélectronique et de Nanotechnologie in Lille, a leading expert on memristive device fabrication and characterization in a bioinspired context.
A PhD in Electrical Engineering or Physics is expected for this challenging project. At least some experience with programming and device characterization is expected. Curiosity and taste for learning new material from new fields is essential. Knowledge of Bayesian probability is not expected before the postdoc.
1D. Querlioz et al, “Bioinspired Programming of Memory Devices for Implementing an Inference Engine“, Proceedings of the IEEE. 103, p. 1398, 2015 (2015)
Skills to be learnt: The postdoc will learn about novel ideas for computing and new memory technology. He/she will learn several methodologies of simulation and experimental characterization. The project is highly adapted to a subsequent career in academia in an Electrical Engineering or Applied Physics department.
Funding is available through an ANR (until March 2017) and an ERC (starting March 2017) project. The project can start immediately but starting date is flexible. Funding is available for 2 years (renewable).