The automated resolution of cognitive tasks primarily relies on learning algorithms applied to neural networks which, when executed on standard architectures, lead to a power consumption several orders of magnitude larger than what the brain would require. This consumption can be drastically reduced by using hardware computing systems with an architecture inspired by biological or physical models, based on nanodevices mimicking the properties of neurons such as the emission of stochastic or synchronous spikes.

Numerous theoretical proposals have shown that nanoscale and multifunctional magnetic tunnel junctions (MTJ) are particularly adapted. The next step is to have a large number of such components interacting on-chip via reconfigurable coupling weights, and demonstrate experimentally an advantage on standard cognitive tasks, both in terms of performance and energy consumption.

This Master 2 research project is part of a larger collaboration between Leti and Spintec, and can later be extended to a PhD project. We propose to i/ study the coupling dynamics of off-chip coupled stochastic MTJs, ii/ perform the characterization of small hybrid (CMOS+sMTJ) integrated circuits acting as binary stochastic neurons.  The candidate will become familiar with the inner workings of the neural network, such as the importance of topology and stochasticity on runtime and efficiency, using Matlab or Python toy models. An understanding of the device physics, in particular the impact of stack design and nanofabrication choices on the measured characteristics will also be needed.

We are looking for outstanding applicants with a strong background in the physics of electronic and magnetic components, as well as nanofabrication and electrical characterization techniques. Basic knowledge in neuromorphic computing is an advantage. An enthusiasm for scientific research and the ability to communicate in a pluridisciplinary environment will be highly valued.