Wrestling better performance from battery chemistries is a tough task. Chemical engineers are faced with solving a multi-variable optimisation problem, juggling energy density, cycle life, cost, safety and charge time – all under intense competitive pressure. For decades, advances had to be coaxed from nature by means of meticulous, slow and expensive experimental iteration. Today, that paradigm has been transformed by computational simulation, compressing development timelines dramatically, writes Peter Donaldson.

The process of synthesis, characterisation and cell testing for a single promising candidate used to take months, and exploring a family of materials could take a decade or more. High-throughput quantum mechanical calculations, pioneered by initiatives such as the Materials Project, have created a digital catalogue of more than 150,000 compounds, enabling today’s researchers to screen them for voltage, structural stability and the abundance of their component elements in a database query taking just seconds. This compresses the initial discovery and down-selection phase from years to weeks.

Perhaps the most significant bottleneck that simulation alleviates is cycle life testing. Validating a 1000 cycle life for a new chemistry with real-world testing previously took a year or more, but physics-based continuum models, built on the foundational Doyle-Fuller-Newman framework, can now simulate 1000 cycles in days. By modelling degradation mechanics such as unwanted growth of the solid–electrolyte interface (SEI), particle cracking and active material loss, developers can predict failure modes before the first cell is built.

Powerful as simulation is, however, it is a tool like any other in that it has limitations imposed both by computation and by the fundamentals of physics and electrochemistry. The timescale problem, for example, means that while it is possible to observe a single chemical bond break with quantum accuracy, it is not yet feasible to watch a stable SEI form over hundreds of cycles in a complete cell. Multi-scale modelling is a useful workaround in which parameters from faster, less accurate methods are passed up to slower, more coarse-grained models, but this is complex and can propagate errors. The same applies across scales of length from a few hundred atoms to a full electrode.

Also, the most accurate quantum chemistry methods such as CCSD(T) are too computationally expensive and slow for screening materials, while density functional theory (DFT) is fast enough but can be inaccurate for parameters crucial to batteries such as band gaps, reaction barriers and van der Waals forces.

Furthermore, the predictions made by machine learning models are only as good as the training data they receive, which can be sparse for such novel cell types as those with solid electrolytes or silicon-rich anodes. AI can also be something of a black box in that it can predict that a new compound will have a high voltage, but may not explain why, which can lead to unreliable predictions if the AI extrapolates beyond its training data.

Fundamental physical principles, meanwhile, present hard barriers that simulation can help engineers to approach but not break. For example, the maximum theoretical energy density of any chemistry is governed by the Gibbs free energy of its reaction. Simulation can help engineers find ways to mitigate the problems that come with any chemistry so that real-world cells come closer to their potential maximum, but it can’t eliminate those problems – or the fundamental limit.

The SEI presents its own problems stemming from its complex, dynamic and self-assembling nature, making it notoriously difficult to simulate accurately across relevant timescales. Extreme complexity also makes real-world degradation difficult to predict through simulation because it involves phenomena that are both non-linear and coupled. For example, a crack can expose fresh electrode material, leading to SEI growth, leading to lithium depletion, leading to increased impedance.

While simulation cannot overcome these limits alone, it can get better at generating intelligent, data-driven hypotheses, which can then be rapidly tested and refined in automated labs. This ‘closed-loop’ acceleration is the next frontier in the race for better batteries.

Click here to read the latest issue of E-Mobility Engineering.