Insider Brief
- Matlantis Inc., Japan's preferred network (PFN) in the US, has launched version 8 of its AI-powered Matlantis™ Universal Atomistic Simulator, increasing the accuracy and speed of material research through advanced machine learning technology.
- The upgraded platform uses PFP (Preferred Potential) Version 8, the first universal inter-machine learning potential (MLIP) trained on the R2SCAN dataset, allowing double the accuracy of the simulation compared to previous PBE training models, while maintaining rapid simulation speeds.
- Matlantis' new US office in Cambridge, Massachusetts aims to accelerate adoption across North America, providing cloud-based AI simulation tools that streamline material discovery in industries such as batteries, semiconductors, and catalysts.
Press Release – Preferred Networks, Inc., a major Japanese AI company. Matlantis Inc., the US hub for the material dispersion division of (PFN), has announced major updates to its Matlantis™ Universal Atomist Simulator and has announced the opening of its office in Cambridge, Massachusetts. This update introduces a new version 8 of PFP (preferred potential) called PFN's proprietary AI technology. This allows industry researchers to accelerate discovery, improve predictive performance, and unlock new frontiers in materials science at the level of simulation accuracy.
PFP version 8 marks important milestones as the first universal inter-machine learning environment potential (MLIP) trained on datasets developed in a new way known as R2Scan (restorated normalized, strongly constrained, well-standardized) features. The PFP version relies on datasets generated by a method called PBE (Perdew-Burke-Arnzerhof) functional, which is widely adopted by non-PFP MLIPs. However, it is known that PBE has certain limitations on the accuracy of simulations. This is how closely a computer-based simulation of material behavior matches actual experimental results.
Introduction of r2The scanning method is the culmination of PFN's ongoing efforts over the past few years to overcome the limitations of the accuracy of a PBE-based approach. Development of training datasets using R2The scanning method is more computationally intensive and requires 3-5 times more computing time than the PBE method. However, PFP version 8 is trained on datasets built with R2Similar to Scan and PBE, Matlantis users can double the accuracy of their simulations in the same time frame as previous versions.
“This update represents a major breakthrough,” says Okanohara Daisuke, CEO of Matlantis. “In 2021, we were the world's first to launch a commercial simulator using Universal MLIP. Now it's the first simulator Matlantis to incorporate R.2Scanning to ensure high simulation accuracy. We believe this will pave the way for an era of computer-based materials discovery. We will continue to support researchers in North America and elsewhere to discover new innovative and sustainable materials. “Co-invested by Mitsubishi companies PFN and Eneos, Japan's largest energy company, Matlantis has already been used by over 100 industrial and academic leaders around the world since its launch in July 2021. Today, Matlantis is one of the first commercially available AI-powered platforms aimed at industrial atomic simulations for industrial MLIP atomic simulations. (Density sensory theory) – Accuracy level up to 20 million times.
Matlantis allows the research team to:
- Run the simulation from the first day of use.
Matlantis is provided as cloud-based software As-a-Service (SAAS). Users can access it via their browser and begin searching for new material from the first day of use. Matlantis' machine learning interatomic potential (MLIP) is already trained on large datasets, allowing users to quickly focus on material discovery without spending time building machine learning models. - Search for a variety of undiscovered materials
As a universal atomic simulator, Matlantis covers a wide variety of materials, including batteries, semiconductors, and catalysts. - Accelerate material discovery
Matlantis allows researchers to complete the simulation in just a few hours. This speedup transforms iterative designs in material discovery and reshapes the R&D process, leading the experiment rather than simply verifying computational insights later. - Achieves unprecedented simulation accuracy
Using a new training dataset built in R2Scanning method, Matlantis, can simulate material properties with higher accuracy than typical MLIP over the same time frame, further narrowing the gap between simulation and experiment.
“With PFP 8.0, there is ultimately the possibility of universal machine learning that maintains the highest DFT level fidelity across most of the regular table,” says Professor Ju Li, Matlantis Technical Advisor, who is widely recognized in his research on Atomistic Modeling and Materials Research. “The combination of accuracy and speed allows engineers to generate multi-component systems of phase diagrams or screens in days or days rather than weeks or months. This is done directly to inform alloy designs, battery materials and other high value applications.
“We are excited to hear about Matlantis' major updates and the opening of a new US office. We are very excited to see the evolution of this platform further promotes our material development,” said Dr. Yoshida Oji, assistant director of the Center for Computational Science and Informatics Research at Resonac.
PFP 8.0 was developed using PFN's supercomputer and AI Bridge Cloud Infrastructure (ABCI) 2.0 and 3.0.
