Battery development is increasingly shaped by what engineers can predict before a material reaches the lab. For electric vehicle (EV) programs, that shift matters because material choices influence how a battery performs over time and how reliably it can be scaled.
Taku Watanabe, head of global customer success at Matlantis
AI-assisted simulation is giving researchers a way to evaluate more candidate materials earlier in the process. One example is Machine Learning Interatomic Potentials (MLIPs), which are AI models that estimate how atoms interact inside a material. These tools can help development teams narrow the field before moving into synthesis and cell validation.
Taku Watanabe, head of global customer success at Matlantis, explains how AI and atomistic simulation are influencing battery materials discovery for EV applications.
Here is what he had to say…
How is AI changing the way EV battery materials are identified and evaluated?
AI is enabling researchers to explore battery materials at a scale that was previously impractical. As EV manufacturers push for higher energy density, faster charging, longer lifetimes, and reduced dependence on critical minerals, the number of potential material combinations has expanded dramatically.
Machine-learning-based simulation methods can evaluate properties such as voltage, stability, lithium mobility, and reaction energetics across large chemical spaces, helping researchers identify materials that balance multiple performance requirements simultaneously.
Specifically, the adoption of Machine Learning Interatomic Potentials (MLIPs), the same class of AI-based atomistic models, can allow simulations to run orders of magnitude faster than traditional first-principles methods while maintaining near-quantum accuracy. This speed-up can be crucial for screening thousands of candidate structures in a timeframe that matches the rapid development cycles of the EV industry.
This is particularly important for next-generation cathodes, solid electrolytes, and other systems where small compositional changes can have a significant impact on performance.
How much progress has been made in connecting atomistic-scale simulations with battery performance at the cell and pack level?
The connection between atomistic simulations and real-world battery performance has improved considerably over the last decade. Material-level properties such as diffusion rates, phase stability, and degradation pathways can now be incorporated into larger electrochemical and thermal models.
The challenge is that battery performance emerges across multiple scales, from atomic interactions to electrode architecture and pack-level thermal management. While no model can yet predict every aspect of battery behavior from first principles, multiscale modeling is becoming increasingly effective at identifying how material choices influence charging performance, cycle life, safety, and overall system efficiency.
EV battery development increasingly depends on earlier insight into how materials perform before cell testing.
How are AI and simulation changing the timeline from initial material discovery to battery cell development?
Battery development is often slowed by the need to validate materials through multiple rounds of synthesis, characterization, and cell testing. AI and simulation help reduce that burden by identifying potential limitations earlier in the process. Researchers can evaluate factors, such as structural stability, ionic conductivity, and electrochemical performance before committing significant experimental resources.
This allows development teams to focus their efforts on the most technically viable candidates and spend less time pursuing materials that are unlikely to meet commercial performance targets.
This acceleration is primarily driven by MLIPs, which can achieve near-density functional theory (DFT) accuracy while running 1,000 to 10,000 times faster than traditional DFT calculations. DFT is a quantum-mechanical modeling method used to calculate material properties at the atomic scale.
This significant reduction in simulation time can enable high-throughput computational screening, allowing development teams to rapidly explore vast compositional spaces and identify the most technically viable candidates early on, which is essential for meeting the rapid development cycles of the EV battery industry.
The result is a shift in the timeline by front-loading the material selection process with high-fidelity computational validation.
Could AI eventually help researchers identify degradation mechanisms that are difficult to observe experimentally?
Yes… some of the most important degradation mechanisms occur gradually over hundreds or thousands of charge-discharge cycles and can be difficult to isolate experimentally. Examples include transition-metal dissolution, lithium trapping, oxygen loss from cathodes, and the accumulation of structural defects.
Computational models can help researchers investigate these processes at the atomic level and evaluate how they evolve under different operating conditions. This is especially valuable when multiple degradation mechanisms are occurring simultaneously and influencing one another, making it difficult to determine the root cause of performance loss through experiments alone.
Which battery materials challenges are proving the most difficult for AI models to predict?
One of the fundamental challenges is the extreme heterogeneity of battery systems. EV batteries are not single-phase materials but coupled assemblies of solids, liquids, metals, oxides, and polymers, with a large number of evolving interfaces between them. This creates a prediction problem that is qualitatively different from modeling isolated crystalline materials, since performance is governed by interactions across chemically and structurally diverse environments.
AI and machine learning (ML) models tend to perform best in relatively uniform systems where the physics is well-defined and the structure is stable. In batteries, they must instead capture behaviors spanning bulk electrodes, liquid electrolytes, solid electrolytes, binders, and interphases, and all within the same framework.
Only a limited class of models, particularly more general atomistic machine-learning interatomic potentials, are currently capable of representing this level of chemical diversity in a consistent way.
On top of this heterogeneity, long-term evolution remains difficult: degradation pathways driven by cycling, side reactions, and gradual structural changes are still challenging to track in a fully predictive manner. While universal ML potentials are improving the ability to represent multiple chemistries within a single framework, a complete predictive description of coupled heterogeneity plus lifetime evolution remains an open problem in battery modeling.
Can AI help researchers better understand failures at electrode and electrolyte interfaces?
Fast charging provides a good example of why interfaces are so important. Lithium plating, impedance growth, and electrolyte decomposition are all strongly influenced by reactions occurring at electrode surfaces. Because these reactions take place within thin, evolving interphase layers, they can be difficult to characterize experimentally.
Computational approaches are increasingly being used to investigate charge-transfer mechanisms, interfacial stability, and reaction pathways under realistic operating conditions. This understanding is becoming particularly important as manufacturers pursue faster charging rates and higher-energy battery designs.
AI-assisted atomistic simulation can help researchers evaluate material behavior during early-stage EV battery development.
How is simulation influencing the development of silicon anodes, solid-state batteries, and other emerging chemistries?
Many emerging battery chemistries face challenges that are difficult to solve through experimentation alone. Silicon anodes, for example, offer substantially higher capacity than graphite but undergo significant volume expansion during cycling.
Solid-state batteries promise improved safety and energy density but must overcome issues related to ionic conductivity, interfacial resistance, and materials compatibility. Simulation allows researchers to investigate these challenges at a fundamental level, helping identify design strategies that improve performance while highlighting potential failure mechanisms before large-scale development begins.
Which areas of EV battery development are likely to see the greatest benefit from AI-assisted materials research?
The greatest impact will likely come in areas where material complexity has become a limiting factor for innovation. This includes next-generation cathodes, solid-state electrolytes, silicon-rich and lithium-metal anodes, and advanced electrolyte formulations. Another important area is battery lifetime prediction, where understanding how degradation accumulates over years of operation remains a major challenge.
As EV batteries become more sophisticated, AI will be increasingly valuable for helping researchers navigate complex design tradeoffs involving performance, durability, safety, manufacturability, and cost.