En poursuivant votre navigation sur ce site, vous acceptez le dépôt de cookies dans votre navigateur. (En savoir plus)

PhD thesis (W/M) Study of ultrafast Mott transitions for the development of an energy efficient artificial intelligence

This offer is available in the following languages:
Français - Anglais

Date Limite Candidature : vendredi 27 mai 2022

Assurez-vous que votre profil candidat soit correctement renseigné avant de postuler. Les informations de votre profil complètent celles associées à chaque candidature. Afin d’augmenter votre visibilité sur notre Portail Emploi et ainsi permettre aux recruteurs de consulter votre profil candidat, vous avez la possibilité de déposer votre CV dans notre CVThèque en un clic !

General information

Reference : UMR6502-ETIJAN-002
Workplace : NANTES
Date of publication : Friday, May 6, 2022
Scientific Responsible name : Etienne JANOD
Type of Contract : PhD Student contract / Thesis offer
Contract Period : 36 months
Start date of the thesis : 1 October 2022
Proportion of work : Full time
Remuneration : 2 135,00 € gross monthly

Description of the thesis topic

State of the art.
The world in which we live is increasingly influenced by the massive use of Information and Communication Technologies (ICT). However, a major counterpart to the extraordinary progress of ICTs is their energy consumption, which represents 10% of the world's electricity in 2021, with an annual growth of 9%. These enormous energy expenditures could be drastically reduced by changing the architecture of current computers, which operate on the 75-year-old Von Neumann paradigm, to an architecture based on neural networks mimicking the potentially much more energy-efficient human brain. This Artificial Intelligence-based approach is currently performed by software running on standard "Von Neumann" computers, making it efficient but still just as energy-hungry. A real breakthrough in energy efficiency would be to implement hardware neural networks based on synapses and artificial neurons.
One possible way to meet this challenge is to turn to quantum materials, i.e. materials whose properties result from strong correlations and/or from the topological nature of their electronic states. In particular, the insulator-to-metal transitions (IMT) existing in the class of Mott insulators are currently attracting great interest worldwide, and the term 'Mottronics' has been coined to represent the concept of electronic technology exploiting such transitions of Mott. The IMN team demonstrated that the electric field, a control parameter suitable for microelectronics applications, offered a simple means of controlling the TIM in Mott insulators. Moreover, we have established and patented that this "electrical Mott transition" can be used to implement the two essential elements of a hardware neural network, the artificial synapse and the artificial neuron. Recently, the IMN team obtained funding for a regional project called MOTT-IA (2022-2026), whose main objective is to build the first hardware neural network using synapses and Mott neurons. This application-oriented project is based on current knowledge and understanding of Mott's electrical transition.
However, the discovery of the Electric Mott Transition is still very recent. Its application potential and ultimate performance, especially in terms of switching speed, remain to be evaluated. Fundamental studies are therefore necessary at this stage. Our recent studies show that the electrical Mott transition is related to the creation of hot electrons, which in turn drive a lattice compressive response. The key role played by hot electrons strongly suggests that, beyond the electric field, light is also capable of inducing similar effects. Interestingly, the impressive development of ultrafast pump-probe techniques over the past two decades now allows probing electron and lattice dynamics after the initial creation of hot electrons by short laser pulses in the femtosecond range (1 fs = 1e-15s). Such techniques, quite appropriate for a fundamental study of non-equilibrium Mott insulator-to-metal transitions, are not available at the IMN. Therefore, we have launched a few years ago a collaboration with the Institute of Physics of Rennes (IPR), where a strong expertise in ultrafast pump-probe techniques exists. The IMN and the IPR have joined forces within the framework of the International Associated Laboratory France - Japan (IMLED, 2017-21), which will become an International Research Laboratory (IRL) focused on the Dynamic Control of Materials (DYNACOM, 2022- 26). The thesis presented here will therefore be the fundamental counterpart of the applied project MOTT-IA, and will be carried out in close collaboration between the IMN and the IPR, with an additional involvement of our Japanese partners.

Scientific obstacles, objectives, methodology and expected results.
The main objective of the thesis is to understand the fundamental mechanisms involved in the electrical Mott transition, which govern the ultimate performance of the two disruptive devices of Mottronics, namely memories (or synapses) and Mott neurons. The expected results of this thesis work will firstly have a direct impact on the Mott-IA project (evaluation of ultimate performance, understanding of mechanisms with a view to industrial development, understanding of the impact of pulse protocols on the material) and on the other hand will open new perspectives on the future of Mottronics.
In this thesis, a first axis will aim to understand which physical parameters govern the ultimate switching times of the electrical Mott transition. Indeed, we have shown that this was linked to the appearance of a conductive filament. However, this phenomenon of essentially electronic origin must lead to a reaction of the crystal lattice in the filament, which could play an essential role in stabilizing the conductive state, a property at the origin of Mott memories and synapses functionalities. For this, we propose on the one hand to study the temporal response of Mott insulators subjected to electrical or light pulses. The use of time-resolved pump-probe experiments, involving excitation by ultrashort laser pulses (roughly 100 fs) will be of particular interest. This excitation indeed induces, like the electric pulse, the creation of hot electrons, and also gives access to the intrinsic time scales of the electronic responses (less than 1 ps) and of the crystal lattice (1-1000 ps). In practice, we will seek to:
- establish the ultimate characteristic time necessary to induce a phase transition out of Mott-metal insulating equilibrium which concerns 100% of the volume. For this, time-resolved reflectivity pump-probe experiments will be carried out at IPR Rennes on several Mott insulators synthesized at IMN, both in the form of single crystals and thin layers,
- understand the nature and dynamics of the structural response of the out-of-equilibrium Mott insulator to metal transition. Our recent work suggests that the structural response of Mott insulators following the creation of hot electrons is inverse to that of most solids, since it corresponds to a compression of the cell volume. We propose to address this question by time-resolved X-ray diffraction (tr-XRD) measurements. These technique is mostly located in large instruments and will therefore depend on the success of the experimental requests. However, IPR has a great expertise in the measurements of tr-XRD which will increase the chances of success of the requests for experience.
In the second axis of this thesis, we will seek to establish the proof of principle of an "all optical" Mott neural network operating one million times faster than a neural network driven by electrical impulses. For this, it will be necessary to validate two new concepts. A first challenge will consist in trying to induce non-volatile insulator-metal transitions driven by single-shot ultrafast light pulses. As part of the ongoing IMN-IPR collaboration, we have performed technically quite similar experiments. Finally, we will aim to implement an artificial neuron controlled only by light, following our recent joint work validating the concept of electro-optical neurons of Mott. For this, we will use specific equipment available at IPR that allows generating multiple light pulses separated from each other by a few tens of picoseconds.
Overall, the first axis described above will constitute direct support for the Mott-IA project, in particular by guiding the choice of the optimal material and by making it possible to optimize the protocols for applying electrical pulses. The second axis will aim to evaluate future developments in Mottronics, in particular toward ultra-fast Artificial Intelligence.

Complementarity of teams / contribution of participants.
The two teams involved in this FAST-IA thesis project, one belonging to the Institut des Matériaux de Nantes Jean Rouxel (IMN) and the other to the Institut de Physique de Rennes (IPR), know each other well and have worked closely together for several years. In particular, they have worked together in two recent ANR projects (Elastica 2016-20 and Electrophone 2019-22), as well as in the framework of a LIA (2017-21) and an IRL (2022-2025) France- Japan. In this project, their role will be very complementary. The IMN brings both the subject of the electric Mott transition, its expertise in the field of strongly correlated quantum materials, as well as its ability to synthesize well-characterized materials both in the form of single crystals and thin layers. The IPR brings to this collaborative project its very strong skills in the physics of phase transitions, in particular out of equilibrium, as well as its great mastery of time-resolved pump-probe techniques.

Role of the PhD student.
The role of the PhD student in this project will be essential. Indeed, he/she will be responsible on the one hand for the preparation and advanced characterization of different Mott insulating compounds both in the form of single crystals and thin layers. For this, he will benefit from both the expertise in the field of the IMN team, as well as access to the laboratory's synthesis and characterization platform. On the other hand, he/she will be heavily involved in the time-resolved measurements that will be carried out at the IPR. In addition, he will actively participate in measurement campaigns on large instruments (such as ESRF, SwissFEL, MAX IV, Soleil), on which our two laboratories regularly obtain two experiments per year. The geographical proximity between the two laboratories will greatly facilitate the direct exchanges and travel necessary to carry out this thesis. The PhD student might spend a few weeks in Japan for pump-probe experiments with ultimate temporal resolutions (less than 10 fs), within the framework of the IRL DYNACOM which involves the Universities of Tohoku (Sendai), Tokyo and the Tokyo Institute of Technology. The candidate will present his results at national and international conferences and write scientific articles related to his work.

Work Context

This thesis is financed within the framework of an interdisciplinary project of the CNRS. It involves two laboratories, the IMN located in Nantes (www.cnrs-imn.fr) and the IPR located in Rennes (www.ipr.univ-rennes1.fr), located approximately 100 km from one other.
The doctoral student will be enrolled in the 3M doctoral school (Materials Matter Molecules) of the University of Nantes and will be attached to the Physics of Materials and Nanostructures team. For the Rennes part, he/she will work in the Materials and Light Department.
The doctoral student will be co-supervised by the two scientific managers of the project, Maciej Lorenc (IPR) and Etienne Janod (IMN). The supervision will also involve the other researchers of the two teams of the IMN and the IPR.

Constraints and risks

Work on two laboratories separated by 100 km
Common risks associated with the use of lasers

Additional Information

The profile sought for this thesis is that of a Master's degree in condensed matter physics, with special attention paid to candidates who have received advanced training in optics.
In addition, other skills will be important to carry out this thesis, such as:
- Master the English language (read, spoken, written),
- know how to lead a project,
- have good writing skills and the ability to communicate and promote the work (presentation of posters, participation in congresses, etc.)
- know how to work in a team.

We talk about it on Twitter!