Context
Non-volatile memories (NVMs) offer several advantages for Multiply-Accumulate (MAC) operations in Artificial Intelligence (AI) applications.
The first and key advantage is a reduced power consumption: NVMs consume less power compared to traditional volatile memories (like SRAM and DRAM) because they do not require power to maintain their state. This can lead to significant energy savings, especially in large-scale AI models that involve extensive computations. In-memory computing with NVMs also minimizes the need to transfer data between the memory and processing units, which is a major source of energy consumption in traditional architectures.
Fast Access Times and Parallel Processing allows multiple MAC operations to be performed simultaneously, thus speeding up AI computations, and thus offering both speed and performance for a higher density.
In particular, ferroelectric and spintronic solutions are well-suited for neuromorphic computing architectures, which mimic the neural structure and operation of the human brain. Contrarily to other physical principles (e.g. phase-change, resistive memories), the offer the large endurance needed to train large AI models.
The question of the energy consumption of the information and communication Technologies is becoming an important environmental and geopolitic issue. We recently proposed in several Nature article (and in several patents ) new spintronic and ferroelectric device, in particular including that combined with high spin-orbit coupling elements, ferroelectrics have also a natural potential to generate an electrically-switchable and highly efficient spin-charge interconversion . This effect can be used to develop a new generation of low-power ferroelectric devices. The host team We are currently building a start-up on this topic (spin-off from CEA, Grenoble Alpes University, CNRS), through a valorization project based on these discoveries.
Position
You will benefit from the expertise of the Topological Spintronics team of the Spintec Laboratory on these new generations of spintronic/ferroelectric devices and of the IC Design and AI teams of the laboratory on the use of non-volatile devices for MAC operation. The goal is to design integrated circuits demonstrating the potential of these devices for AI neuromorphic solutions. You will design the reading and writing circuit allowing to take advantage of these devices for multiply and accumulate operations. You will also work in close collaboration with the experimental team developing the devices, and benefit from the existence of a large collective momentum in our teams towards the development and the integration of these devices, with ongoing ANR and EU projects, and more importantly with a valorization project aiming at creating a start-up based on this technology.
Requested qualities: PhD with skills in computing/design/AI, topics involving inventiveness, cooperative work, interest towards innovative microelectronics applications, and towards intellectual property creation. Possible will to pursue a career in a start-up environment.
How to apply
Please send CV to the following contacts:
Jean-Philippe Attané and Laurent Vila
Bibliography
1). Noël, Attané, Vila et al., Nature 580.7804 (2020): 483.
i). P. Noël et al., Nature 580, 483 (2020) ; S. Varotto et al. Nature Electronics 4 (10), 740-747 (2021) ; Trier et al., Nature Reviews Materials, 1-17 (2021)
ii). WO2020136267 (FR1874319), WO2022129306 (FR2013673), FR2108563, FR2108564, FR2200906
