Tremolo-X

Innovative materials are developed today at the smallest scale – on the atomic level. Tremolo-X opens the door to nanoscale simulations: the software delivers precise insights into the structure of materials. This enables new materials to be designed virtually on the computer, reducing the time and cost of extensive laboratory experiments.

Tremolo-X is a high-performance simulation engine for nanotechnology and materials science developed by Fraunhofer SCAI. The software calculates complex many-body potentials using state-of-the-art parallelization and serves as the numerical backend of the force-field module in Synopsys‘ QuantumATK simulation toolkit.

Tremolo-X can perform molecular dynamics simulations with 100,000 to several million atoms depending on the interaction potential and hardware. This way, Tremolo-X provides the accuracy to explore and predict material properties in detail.
 

Wide range of interaction potentials

With more than 480 built-in parameter sets, Tremolo-X supports a wide spectrum of established interaction potentials, including OPLS, Tersoff, and ReaxFF. These potentials describe how atoms in a material influence one another. A highlight is full support for AI-driven models such as the linear Moment Tensor Potential (MTP) or non-linear equivariant graph neural network potentials such as MACE and MatterSim, which are especially well-suited for complex materials. Integrated training is performed on multiple graphics cards (Multi-GPU).
 

Artificial intelligence in materials research

The current version supports message-passing graph-convolution potentials such as MACE and MatterSim. These AI models analyze materials based on their atomic structure by “passing information” between atoms (message passing), enabling realistic simulations of materials with any combination of chemical elements. Tremolo-X offers one of the first multi-GPU implementations of these models, achieving excellent parallel scaling performance on modern HPC systems.
 

Powerful modeling – flexible to train

An integrated training framework allows models to be tailored precisely to specific needs. It supports multi-fidelity learning (combining data of varying quality), multi-task learning (simultaneously learning multiple properties), and pairwise training (comparing pairs of materials). This results in models optimized for the exact requirements of each application.

Tremolo-X does not have its own GUI. The high-performance engine is integrated into Synopsys' atomistic simulation software QuantumATK. For individual projects, the Virtual Material Design business area at Fraunhofer SCAI is happy to provide support.