Thèse Analyse par l'IA Informée par la Physique et les Données des Mécanismes de Dégradation et de Restauration dans les Cellules Solaires à Pérovskites H/F - Doctorat.Gouv.Fr

Établissement : Institut Polytechnique de Paris École polytechnique École doctorale : Ecole Doctorale de l'Institut Polytechnique de Paris Laboratoire de recherche : Institut Photovoltaïque d'Île-de-France Direction de la thèse : Jean-François GUILLEMOLES ORCID 0000000301148624 Début de la thèse : 2026-11-01 Date limite de candidature : 2026-10-31T23:59:59 Les cellules solaires à pérovskites, grâce à leur forte absorption lumineuse et leur bande interdite ajustable, offrent un potentiel exceptionnel pour des rendements photovoltaïques élevés, aussi bien en dispositifs simples qu'en tandems pérovskite-silicium. Cependant, leur stabilité opérationnelle reste un obstacle majeur à leur déploiement industriel. Ce projet de thèse vise à comprendre les mécanismes physiques sous-jacents à la dégradation et à la récupération des performances, en analysant comment l'éclairage, la température et la polarisation électrique influencent ces processus.
L'approche combine expériences contrôlées, modélisation physique (via des modèles de dérive-diffusion) et analyse pilotée par les données (IA informée par la physique). Les mesures périodiques de densité de courant-tension (JV) permettent de suivre l'évolution des propriétés électriques (tension en circuit ouvert, courant de court-circuit, facteur de forme) pour identifier les trajectoires de vieillissement. Des expériences ciblées, incluant des traitements d'interface et des conditions opératoires variées, testeront les hypothèses sur les mécanismes dominants.

Objectifs clés :
Identifier les mécanismes de dégradation à différentes étapes du vieillissement.
Comprendre comment les conditions opératoires influencent la récupération.
Développer des modèles prédictifs pour anticiper la réponse des cellules à de nouveaux scénarios.

Ce travail contribuera à prioriser les améliorations d'interface, évaluer les stratégies de restauration des performances en fonctionnement et de concevoir des dispositifs plus stables.

Controlled ageing and recovery experiments. Existing IPVF ageing datasets will establish the initial
electrical trajectories and guide new experiments on a documented reference architecture. Replicated
cells from several fabrication batches will be exposed to controlled illumination, temperature and bias,
first varying individual factors and then testing their interactions. Comparisons between maximum-power operation, open circuit and dark rest will examine when recovery occurs, how long it takes and whether improvements persist after normal operation resumes. Periodic JV measurements and operating-point tracking will follow each cell through these histories.
An automated dark-JV setup will resolve recovery changes missed by daily manual measurements.
Standardised illuminated measurements will connect these changes to photovoltaic performance, while matched controls will assess the effect of the measurement sequence. Targeted calibrated
photoluminescence and external quantum-efficiency measurements will help separate changes in
recombination from changes in collection. Comparisons of matched cells with different interface
treatments will test whether modifying a suspected source of loss changes the ageing trajectory.
Physical modelling and data-driven analysis. Drift-diffusion models based on SIMsalabim will connect
the measured JV evolution to charge transport, recombination and contact properties. The PhD will
extend the internship's individual-parameter comparisons to coupled changes and time-dependent
models, including ionic redistribution and its influence on electric fields and interfaces. Joint fitting of
measurements at several irradiances and selected device states will constrain the models. Ensembles of acceptable fits and sensitivity analysis will show which parameter changes are consistently required and which additional measurements would be informative.
Physics-informed artificial intelligence will combine experimental data and device simulations to identify patterns in degradation and recovery. Optimisation and machine-learning approximations of the simulations will accelerate parameter searches, while statistical inference will quantify parameter
uncertainty and correlations. These analyses will identify influential operating conditions and help select experiments that best separate candidate mechanisms. Model predictions will then be tested on
independent cells and operating histories excluded from fitting, creating a cycle in which experiments
improve the models and the models guide the next experiments. Lifetime indicators such as T80, the time to reach 80% of a reference efficiency, summarise stability but conceal how a device reaches that point. Two cells with the same T80 can follow different degradation pathways, with losses appearing at different times and affecting different electrical properties. Periodic current density-voltage (JV) measurements retain this intermediate evolution. Changes in open-circuit voltage, short-circuit current density, fill factor and the full curve shape provide information that is lost in an efficiency threshold or a before-and-after comparison.
Relating these electrical trajectories to physical mechanisms presents a second challenge. Different
combinations of model parameters can reproduce a JV curve equally well, so a good fit alone does not
establish which physical changes occurred. The research will address this ambiguity by combining
measurements under several conditions with complementary characterisation and experiments designed to test the competing explanations.
The central questions are which mechanisms dominate at different stages of ageing, how illumination, temperature and bias influence their interactions, and which losses recover after a change in operating conditions. A further objective is to determine which measurements best distinguish these mechanisms and whether the resulting models can predict a cell's response to a new operating history.
The project builds on my M2 internship at IPVF, which combined periodic JV measurements with genetic-algorithm fitting of drift-diffusion models. Preliminary modelling suggested that changes in recombination at the hole-selective interface could help explain early ageing in some cells. Measurements on one aged cell also suggested partial recovery in JV-derived power after switching from maximum-power operation to open circuit, motivating replicated studies of operating-point effects.

Year 1 will establish the experimental and analysis foundations by organising existing data, characterising the reference architecture and developing automated recovery measurements. Initial replicated ageing studies and calibrated models will identify the main candidate mechanisms and follow-up tests.
Year 2 will test those mechanisms through controlled changes in operating conditions and interface
treatments, supported by complementary characterisation. Coupled and time-dependent models will be developed to explain degradation beyond early ageing and the dynamics of recovery.
Year 3 will test predictive performance across independent cells and operating histories and assess
transfer to another device architecture or operational dataset. The final work will consolidate the
scientific conclusions, reusable analysis software, publications and doctoral dissertation.
The expected contribution is an experimentally tested account of how degradation and recovery depend on device properties and operating conditions, together with methods for extracting this information from repeated electrical measurements. By connecting lifetime losses to their physical origins, the work will help prioritise interface improvements, evaluate recovery strategies and design more stable perovskite devices.

candidat sélectionné : Louis YUE TING SHING

• Ionic