Written first for LinkedIn, kept here because the argument outlives the feed. Longer pieces live under writing.
The future: multi-scale simulations available to everyone, not just large corporations
The future I want to see is multi-scale simulation available to everyone. Not just large corporations.
The barrier today is not only compute. It is complexity. Setup. Workflow fragility. Interpretation.
Platforms lower the barrier by encoding expertise into templates and guardrails. The goal is not to remove complexity. It is to make it usable.
Democratization also means making workflows understandable, not just runnable. The goal is guided competence: defaults that work, assumptions that are visible, and results that are explainable.
This trip left me with deep admiration for the people we met, and a stronger conviction that Japan will help lead the next wave of real-world AI applications.
During the final few days of my last trip, we spent time with teams at NVIDIA, ITOCHU Techno-solutions, visited Prof. Taro Hitosugi’s lab at the University of Tokyo, and met with colleagues at Science Tokyo and Tsukuba. Thank you to everyone who welcomed us and shared their work.
What makes me excited is the depth of knowledge here. Developing a material, making a semiconductor, or getting a manufacturing process to work reliably takes years of accumulated experience. That knowledge becomes even more valuable when AI can learn from it and help researchers test new ideas.
Prof. Hitosugi’s lab is one concrete example: robots synthesizing and measuring materials, with data guiding the next experiment. Seeing that work in person makes the possibilities easier to appreciate.
The US–Japan collaboration is bringing advances in AI and computing together with deep experimental and manufacturing expertise, shortening the path from idea to something useful. Better materials and more efficient ways of making things would benefit people far beyond either country.
At Mat3ra, we’re building the infrastructure to connect materials simulations, data, and AI. I want us to help researchers on both sides of the Pacific work together, and give the next generation of scientists the tools to take this further.
A lot to learn from Japan. A lot to build together. Stay tuned!
I first heard about Tsukuba as a student, reading about one of the world's fastest supercomputers.
I first heard about Tsukuba as a student, reading about one of the world's fastest supercomputers. Now, I got to see it in person.
Mat3ra Japan and the University of Tsukuba’s Center for Computational Sciences (CCS) signed an MoU. Excited to be working together. The agreement will give participating CCS students, faculty, and researchers access to Mat3ra, with training and a dedicated liaison to help them get started. We’ll also work with CCS to expand adoption across Japan’s research community.
For me, a big part of this is education. The students learning to use simulation and AI today will be the scientists developing tomorrow’s materials for semiconductor, energy, defense, and other sectors. We want to give them the best tools they can use throughout that journey.
We also visited the RnD team at Sekisui Chemical, which has R&D operations in Tsukuba, to discuss AI and Informatics for Materials RnD.
Japan has so much materials and manufacturing knowledge to build on. I see AI as a catalyst for putting that knowledge to work, and Mat3ra as part of the infrastructure that will make it possible.
Prediction: in 5 years, no R&D team will work without a unified computational workspace
Prediction: in five years, no serious R&D team will operate without a unified computational workspace.
Software engineering went through this evolution. From local scripts to shared repositories. From ad-hoc deployments to CI. From tribal knowledge to systems.
Science is catching up.
The cost of scattered workflows is now too high. The value of shared provenance is too obvious.
A unified workspace does not mean one tool for everything. It means one place where work is captured consistently: data, compute, environments, and provenance.
So far, we’ve met with UTokyo Prof. Junichiro Shiomi, Teruyasu Mizoguchi, and Shigeru Kobayashi, caught up with our long-time partners at ENEOS Group, and visited two growing communities: RISE-A for the semiconductor ecosystem (thanks to Kiyoshi Nakazawa) and the artience group (former Toyo Ink Group) Incubation CANVAS for materials science. Discussing AI for Materials Science and cross-border collaborations supporting the Office of the Under Secretary for Science, U.S. Department of Energy.
Japan’s depth in materials and manufacturing puts it in a unique position. This expertise will become even more valuable as AI changes how R&D is done across multiple "physical" industries, including semiconductor, energy, and defense sectors. So much practical knowledge here to build upon.
Education is a big part of this. A new generation of scientists will help grow Japan’s semiconductor sector and take its materials and manufacturing capabilities further. Giving them early access to AI and powerful simulation tools—in their education and research—can rapidly accelerate that progress.
We’re building Mat3ra to be a catalyst for AI adoption, providing the infrastructure for students, researchers, and industry to bring materials data, simulation, and AI into their work.
Thank you to our hosts and partners. Excited to keep building together.
Jean-Dominique F. and I are at JPS 2026, prompting the next generation of scientists to rethink how materials discovery gets done.
Breaking down the wall between theory and experiment. For too long, computational modeling and lab experimentation operated in separate silos. We are showing how browser-native physics enables seamless co-design and co-creation—without the technical barriers holding anyone back.
Co-Creation Without Barriers: Experimentalists can launch DFT, Molecular Dynamics, and machine learning workflows straight from a browser—no HPC setup or server maintenance required.
Closing the Loop: Rapid iteration between lab synthesis and atomic-scale predictions accelerates real-world clean tech and materials discovery.
Empowering New Scientists: Equipping students with open Python, Jupyter, and DFT/MD toolkits so they can bridge theory and practice from day one.
The future of physics isn't theory or experiment—it’s both working together from the start.
If you're walking the halls at UTokyo Komaba, stop by Booth 3-2 to chat about co-design, see a live workflow, and grab a turtle sticker!
Why materials scientists benefit from adopting a “product-builder” mindset
A useful mental shift for researchers is adopting a product builder mindset.
It means: iterate quickly, measure what matters, and build reusable components.
In research, the component is often a workflow. A pipeline. A template. A dataset. When you build those, you are not just doing science. You are building leverage for future science.
This mindset also makes collaboration easier: you can hand someone a workflow that works, not just an outcome that must be re-created.
In practice, this is how research teams become more than the sum of individuals.
HPCwire features AlphaFilm and the DOE Genesis Mission
HPCwire featured AlphaFilm today — the DOE Genesis Mission project Mat3ra.com leads with the National Lab of the Rockies and the University of Tennessee, Knoxville.
Thin films for semiconductors don't yield to simulations alone. You deposit a film, then scan it to measure how defects change its behavior, adjust the recipe, and go again — for weeks, sometimes months per iteration. AlphaFilm closes that loop: physics-based and AI models on Mat3ra, combinatorial co-sputtering at the National Lab of the Rockies, automated multimodal microscopy at UT Knoxville, with AI agents deciding what to try next. The goal is to turn weeks into hours.
The material—ferroelectric aluminum scandium nitride—underpins next-generation non-volatile memory, RF filters, neuromorphic computing, and sensors built for jet engines and spacecraft.
Grateful to Sergey Barabash, Sergei Kalinin, Andriy Zakutayev, and the Office of the Under Secretary for Science, U.S. Department of Energy team — and to Alex Woodie for taking the time to write this up.
Everything around us is made of materials, yet we rarely think about it
Everything around us is made of materials.
Phones, planes, batteries, buildings, medicine… and yet most people never think about materials science.
It’s invisible infrastructure.
That is part of why I find this field so meaningful: small improvements in materials quietly reshape what is possible in energy, health, and technology.
If we can reduce the friction of discovery even a little, the impact compounds far beyond any single lab.
Why reproducibility is the biggest bottleneck in modern labs
Reproducibility is often described as a scientific virtue. In practice, it is the main bottleneck.
If you cannot reproduce a run, you cannot trust it. If you cannot trust it, you cannot reuse it. If you cannot reuse it, progress stays linear.
In modern labs, the bottleneck is rarely compute. It is provenance: what exactly happened, with what inputs, using what environment.
Make provenance automatic and reproducibility becomes normal. Then reuse becomes normal. That is where speed comes from.
When reproducibility is weak, teams compensate with meetings, manual checklists, and reruns. That looks like rigor, but it is mostly friction. Good systems replace meetings with provenance.
