Ph.D thesis (H/F): Digital spintronic neural network circuits

The Ph.D thesis will take place within the interdisciplinary INTEGNANO research group at the Centre de Nanosciences et de Nanotechnologies (Palaiseau) and the Cognitive Information Processing at the Unité Mixte de Physique CNRS/Thales (Palaiseau) . These groups associate research on nanodevices, bioinspired computing and artificial intelligence, with researchers and students of very diverse backgrounds, making them exciting environments that foster interdisciplinary thinking.

In recent years, deep learning has revolutionized the field of Artificial Intelligence, and computers are now able to recognize image, understand spoken language or even translate texts at near-Human or even sometimes above Human-level performance. However, deep learning comes with a major challenge: its considerable energy consumption.
It is now well understood that reducing the energy consumption of deep learning can come from a close association of logic and memory. This association cannot be achieved using CMOS technology only, as CMOS-based memory (SRAM) has considerable area cost and is a volatile memory. By contrast, novel technologies coming from nanoechnologies can provide ideal features for neural networks implementation. One of the most exciting opportunity are technologies such as Spin Torque MRAM that rely on spintronics, i.e. that exploit the magnetic properties of electrons. These technologies, now widely developed in industry (Intel, Samsung, GlobalFoundries, TSMC…) provide multiple features for deep learning.
The PhD project consists in implementing a novel deep learning approach, partially inspired by how the brain works, taking advantage of spintronic technologies. Our design will use digital circuits as neurons, and spintronic devices as synapses, which can be integrated monolithically within CMOS. The goal of the PhD is to design all the digital system to implement deep learning, as well as the interface circuitry between CMOS and spintronic devices.

Activities
- Development of deep algorithms dedicated to spintronic/CMOS hardware
- Simulations of the algorithms
- Neuromorphic computing with the developed hardware
- Digital ASIC design (Verilog or VHDL) using industry standard tools.
- Design of interface circuitry (e.g., sense amplifiers) between spintronics and digital CMOS

Skills
- Knowledge of deep learning
- Expertise with Python
- Expertise in digital design