Skala

Skala, the deep-learning-based exchange-correlation (XC) functional for density functional theory (DFT) developed by Microsoft Research AI for Science, is now available in the Azure AI Foundry catalog. Trained on an unprecedented quantity of diverse, high-accuracy data, Skala achieves experimental-level accuracy within the region of chemical space represented in its training set (atomization energies). It achieves accuracy competitive with the best-performing hybrid functionals across general main group chemistry, while retaining the computational efficiency of semi-local DFT.  Skala represents a major milestone in advancing the predictive power of computational chemistry and other sciences by enabling reliable, scalable predictions for simulating molecules.

Skala is a breakthrough in the accuracy/cost tradeoff in density functional theory (DFT), the workhorse method that thousands of scientists use every year to simulate matter at the atomistic level. Bringing the accuracy of DFT in line with experimental accuracy addresses a fundamental barrier to shifting the balance of molecule and material design from being driven by laboratory experiments to being driven by computational simulations.  

While DFT offers scientists a computationally cost-effective way to address many problems, practical applications rely on approximations to the unknown exchange-correlation (XC) functional, which represents a crucial part of the energy of a molecule or material. Most existing XC functionals are constructed using a limited set of increasingly complex, hand-crafted features that improve accuracy at the expense of computational efficiency. Yet, no current approximation achieves the accuracy and generality for predictive modeling of laboratory experiments at chemical accuracy.

Microsoft Research (MSR) AI for Science has developed Skala, the first modern deep learning-based exchange-correlation functional. Unlike traditional XC functionals, Skala bypasses commonly used expensive hand-designed input features and instead learns complex non-local representations that are used to make energy predictions in a data-driven manner.  

This is enabled by training the model using an unprecedented amount of high accuracy data which we generate in-house and in collaboration with world-leading experts of highly accurate but more expensive electronic structure methods. 

This achievement shows for the first time that deep learning offers a clear and computationally scalable path to building an accurate, efficient, and broadly applicable model of the universal XC functional. Enabling DFT to shift from interpreting experimental results to predicting them reliably unlocks enormous potential across domains—from drug design to battery development—where accurate, affordable simulations can reduce costly lab work. 

Skala is now available in the Azure AI Foundry catalog and GitHub to the scientific community. 
Skala represents our first milestone model that is suitable for main group chemistry molecular systems. As we continue to expand the training dataset to extend coverage to other important areas, including transition metals and periodic systems, future versions of Skala are poised to further enhance the predictive power of first-principles simulations. By making this work available to the scientific community, we hope to enable widespread testing and gather valuable feedback that will guide future improvements. 

• Skala on Foundry
• Skala on GitHub
• AI for Science DFT website