Entalpic

⚛️Chemical reactions aren't just part of daily life; they're how we discover new materials every day at Entalpic. Yet, they remain one of the hardest phenomena to model, with an immense combinatorial space and complex atomic-scale dynamics. Last week, we had the pleasure of spending a morning with Guillaume Stirnemann’s group at Ecole normale supérieure - PSL Research University, following their recent CECAM workshop on machine learning for chemical reactivity. It was a great opportunity to compare perspectives and discuss how our respective approaches can complement each other. In particular, we presented our new foundation model for atomistic systems (Triforces) and discussed how MLIPs can be integrated into chemical reactivity workflows. We also shared some of our recent open-science efforts around LeMat-GenBench, charge-density prediction, and AI-driven catalyst discovery. Finally, LeMat-Rho was also mentioned: our dataset of inorganic crystal structures with computed DFT properties, including energies, forces, and charge densities. A big thank you to Guillaume and the whole team for the stimulating discussions. It’s always exciting to see different communities converge around the same scientific challenges. We’re looking forward to seeing where these exchanges lead. 😊 #AI4Science #ComputationalChemistry #MachineLearning #MaterialsDiscovery #ChemicalReactivity

✨ 🇰🇷 Ali Ramlaoui and Joseph Musielewicz presented our new machine learning model for atomic-level property predictions, called TriForces, at International Conference on Machine Learning - ICML 2026 in Seoul! They brought two main additions to classic ML Interatomic Potentials (MLIP) to make them more generalizable and easier to fine-tune. → The first one was a self-supervised pre-training approach to obtain more meaningful representations, instead of training directly on energy and forces. → They use separate streams to learn composition, structure, and how the two interact before combining them, making the resulting representation easier to reuse across tasks. Beyond presenting the work, what stood out were the interesting conversations. Many people immediately recognized the difficulty of jointly training on energies and forces, and there was strong interest in self-supervised learning for materials. 👀 Keep an eye on Entalpic’s LinkedIn for any future events where you can see our team explain our program in detail!

What happens when a team of Entalpists take on a treasure hunt in 40°C heat? You discover that a team of scientists and engineers can spend 20 minutes defending the wrong interpretation of a clue before blaming the map. That was one of the highlights of our annual seminar at Château de Villeray. 🏰 We came back from this Entaltrip with a lot of good conversations and even stronger team spirit. Now it's back to building. We have an exciting second half of 2026 ahead, and we're looking forward to what's next. 🚀 If that sounds like your kind of team, we're hiring across engineering, research, and business ! 👀 Take a look at our open roles: https://lnkd.in/e67Cqfq2 #Entalpic #TeamCulture #DeepTech #WorkCulture

🚀 Entalpic is releasing a new model called Triforces, currently presented at ICML ! TriForces tackles a question we've been thinking about for a while in atomistic Machine Learning: "Can we move beyond task-specific MLIPs toward reusable representations of materials? MLIPs have become good predictors & substitutes for DFT. Far less attention has been paid to the quality of the representations they learn. As a result, composition, structure and interactions become entangled in a single latent space, optimised for targeted prediction tasks, making transfer learning brittle and limiting reuse across downstream tasks. Our approach combines two ideas: 🧠 - A self-supervised learning module (denoising, masking and JEPA-style latent prediction) that learns useful representations without requiring additional DFT labels. - A three-stream architecture that explicitly separates composition, structure, and their interactions, instead of compressing everything into one representation. Why is it nice ? 🤩 - TriForces is architecture-agnostic: it plugs into existing MLIPs (MACE, UMA, ORB-v3...) - It consistently improves low-data performance and make fine-tuning easier. - It opens the door to applications beyond property prediction, such as similarity retrieval using composition, structure, or both. More broadly, I hope this pushes the field toward a different paradigm: first learning rich, meaningful representations of materials, then fine-tuning them for specific tasks (energies, forces, stress, charge density, and beyond). Reusability and steerability may become just as important as raw benchmark accuracy, as people now want very good specialised models. Huge congratulations to Ali Ramlaoui, who led this project from start to finish, and to the whole team: Joseph Musielewicz, Hannah Bull, Victor Schmidt, and Hugues Talbot Fragkiskos Malliaros at CentraleSupélec and Inria. Credits to GENCI & Stephane Requena for the compute 🖥️ ⚠️ Ali & Joe are at [ICML] Int'l Conference on Machine Learning this week presenting it ! Reach out to them if you wanna chat science with them ! Where ? Hall A #1113 Tuesday 10:30AM KST

⚗️ At Entalpic, we are building an AI platform for ALD materials innovation, from precursor discovery to deposition process optimization and thin-film design. We’re excited to introduce the advisors who support our mission. ✨ Michael Nolan is Head of Group Materials Modelling for Devices and CMOS++ Research Cluster Lead at the Tyndall National Institute (Cork, Ireland). He is internationally recognized as one of the leading experts in atomistic simulation for semiconductor manufacturing. His group has pioneered the use of first principles to understand and predict reaction mechanisms in Atomic Layer Deposition (ALD), Atomic Layer Etching (ALE), and Hybrid Molecular Layer Deposition (MLD), materials processing technologies that are at the heart of advanced semiconductor device and chip fabrication. His work has helped establish atomistic modeling as a practical tool for engineering surface chemistry at the atomic scale. We are delighted to have him on board! Fun fact: Michael will always be found at a conference proudly displaying his heavy metal shirts 🤟 #ALDep #ALEtch #MLDep #Semiconductors #ComputationalChemistry #MaterialsScience #AI4Science

🎉 Entalpic is giving a talk at the "LNE Journée Technique" on Nanomaterials and AI on July 2nd (10:30am) at LNE Salle des Conférences. Indira Fabre will present our work on AI-driven Engineering of Materials at the Atomic Scale. Thermo Fisher Scientific, TEMAS Solutions and other interesting chemistry actors will also be presenting. If you're attending, come find us ! #ALD #AI4Science #MaterialsDiscovery #Semiconductors #Entalpic #AtomicPrecision

🤗 Entalpic is growing 🤗 We’re hiring 2 Computational Chemists + 1 Intern to strengthen our Modelling Team & help us push the frontier of AI-driven materials discovery ! ⚛️ Surface Chemistry: atomic layer modelling (ALD/ALE), reaction mechanisms, MLIPs, multi-scale modelling. ⚡Electronic Structure & Functional Materials: beyond-DFT methods, dielectric and optical properties, for the semiconductor industry. 🤖 Intern: high-throughput DFT workflows, machine learning, reaction networks. If you enjoy running large-scale quantum simulations, building new workflows, questioning assumptions, and working on industrial problems where the science is still being written, we would love to hear from you 👇 Application Links in comments 👇 If someone comes to mind, please SHARE 🤗

🙌 There will be three occasions to meet the Entalpic team at Tampa from June 28th to July 1st at the ALD/ALE 2026 event. Luis Pinto, Alexandre Duval, and Mathieu Galtier will be presenting: -A talk about multi-objective precursor discovery using steerable generative AI (5:15 – 5:30 pm · June 29th) -A pitch about high-throughput in-silico exploration of ALD reaction mechanisms (3:15 – 3:30 pm · July 1st) -A presentation about our Digital twin framework combining AI, computational chemistry, and experimental data. (4:15 – 4:30 pm · July 1st) Come speak with the team and discuss about your challenges at the times listed above, or reach out to meet during the event! #ALD #AI4Science #MaterialsDiscovery #Semiconductors #ThinFilms #Entalpic

Excited to share that Ali Ramlaoui and I will be presenting our work on TriForces (https://lnkd.in/eZvFrPf9) at #ICML in Seoul this year! If you're going to be there July 6-11, let me know — I'd love to grab coffee or chat during a poster session! And if you use graph neural networks for chemistry (MLIPs or property prediction) definitely come check out our poster. We have shown how to make atomistic GNN representations more transferable, sharply improving accuracy in low-data regimes, and achieving SOTA results on most Matbench tasks.

⚛️ Energies and forces have powered much of the recent progress in atomistic ML modeling. Charge densities, however, contain a richer representation of materials: the quantum-mechanical distribution of electrons that underlies bonding, structure, and properties. As part of the LeMaterial open science initiative, we are releasing LeMat-Rho: 70,000 r²SCAN charge-density calculations generated through a stringent PBE→r²SCAN workflow. The dataset is now available on Hugging Face. From accelerating DFT workflows to enabling new model architectures, we look forward to seeing what the community builds with it! The full write-up is on LeMaterial. You can find the link in the comments 👇 #MaterialsDiscovery #AI4Science #ComputationalChemistry #Entalpic #OpenScience #MachineLearning