Farasis Energy reached a landmark achievement in battery R&D through a collaborative study featured on the cover of the February 5, 2026, issue of Nature. Our research, “Discovery Learning predicts battery cycle life from minimal experiments,” was led by Dr. Weiran Jiang, Ph.D., Vice President of R&D at Farasis Energy, and Professor Ziyou Song of the University of Michigan.
This work highlights a joint effort between Farasis Energy, the University of Michigan, and the National University of Singapore. Farasis team members Dr. Qi Jiao and Dr. Yao Ren worked alongside Dr. Jiang to facilitate the study’s transition to commercial-scale application. Their contributions established the research foundation through the delivery of comprehensive manufacturing data and large-format battery prototypes.
Serving as the industrial lead for this three-year project, Farasis Energy ensured each stage of the research remained aligned with real-world production requirements. We collaborated closely with the university teams through technical reviews and brainstorming sessions to solve complex kinetic and thermodynamic problems. This consistent exchange enabled us to infuse our manufacturing expertise into the scientific direction of the study, achieving a seamless transition from academic theory to practical use cases.
Predicting Battery Cycle Life with Discovery Learning
Our new physics-guided machine learning framework, Discovery Learning, draws inspiration from educational psychology to solve the challenge of predicting battery longevity. By integrating active learning, physics-guided learning, and zero-shot learning, our paradigm predicts cycle life using only a small fraction of the data typically required. This represents a fundamental shift from traditional development cycles often slowed by the high cost of long-term prototype testing.
“The sustainability of the battery industry depends on our ability to innovate faster than the market demands,” says Dr. Weiran Jiang. “By replacing years of physical testing with high-fidelity predictive modeling, we solve the sustainability dilemma of battery research. This technology allows us to deliver next-generation power solutions to our partners in a fraction of the time.”
Results demonstrate that Discovery Learning accurately predicts cycle life using physical features from only the first 50 cycles of just 51% of cell prototypes. With a 7.2% test error, this delivers profound efficiency gains. Under conservative assumptions, the methodology results in a 98% reduction in evaluation time and a 95% reduction in energy consumption compared to conventional industrial practices.
Lowering these resource requirements will accelerate the deployment of high-performance batteries for EVs and grid storage. We are now moving to translate these academic achievements into practical value by launching field trials of the Discovery Learning framework immediately.
The full study can be accessed at Nature: https://www.nature.com/articles/s41586-025-09951-7
