Automated Spectroscopy for Energy Materials research group

Why We Need Automated Operando Spectroscopy for Energy Materials

The materials that make up solar cells, batteries, fuel cells, and electrolysers convert energy from one form to another at interfaces. Understanding why some materials work and others fail requires knowing how and why interfaces change as a result of operation. Currently, operando spectroscopy is the only way to access this information — but it is slow, expensive, and statistically sparse: a typical study might only collect operando spectra (spectra under operating conditions) for three materials. Through no fault of the researcher, random fluctuations in the material, the measurement, or the sample preparation can dominate the results. The literature is therefore built on a patchwork of sparse, hypothesis-driven studies that cannot be reliably compared. Our group is dedicated to solving this problem. Below you will find an example of the kind of research problem we are tackling, and how we are solving it to give you a flavour of the work.
01

Our only commercially viable electrolyzer technology relies on one of the scarcest elements on Earth

Proton exchange membrane (PEM) electrolyzers are the only technology capable of producing green hydrogen at the rates and pressures demanded by industry. The reaction takes place in two parts: first water is split into oxygen and protons at the anode, then the protons are reduced to hydrogen at the cathode. The oxygen evolution reaction (OER) is extremely slow and requires an effective catalyst. This task is so aggressive that only one catalyst survives: iridium, one of the rarest elements in the Earth's crust. To meet only the UK's initial targets for green hydrogen production would require the entire annual global production of iridium. We therefore should seek to reduce iridium usage or eliminate it entirely.

Scatter plot of crustal abundance versus annual production for key electrochemical catalyst elements, showing iridium as the most scarce
Figure 1. Crustal abundance versus annual production for elements relevant to electrocatalysis. Iridium (Ir) sits far at the far end of the trend line, combining vanishingly low abundance with minimal production — placing them in the highest-concern category. Earth-abundant elements such as iron (Fe), manganese (Mn), and silicon (Si) lie orders of magnitude higher on both axes. Data taken from Vesborg et al.
02

The current theoretical framework for OER catalysis has not led us to new materials

OER catalysts work by binding to water and repeatedly pulling a proton and an electron off of it, forming a series of stabilised intermediates (*OH, *O, *OOH) that eventually release oxygen gas (Fig. 2a). If any one of the intermediates binds too strongly or too weakly, the reaction slows down because the catalyst is poisoned (strong binding) or inactive (weak binding). Thus, we expect the best materials to follow 'volcano' relations - where middle binding strengths produce the most active catalysts (Fig. 2b, red line). This is very neat until you test the theory by measuring a material's performance and then using computational chemistry to calculate its binding energy descriptor. With such a poor correlation, it is hard to see a clear way forward to designing performant catalysts, let alone performant and stable catalysts!

A simple oxygen evolution reaction pathway showing *OH, *O, and *OOH intermediates
Figure 2a. A simple oxygen evolution reaction pathway showing *OH, *O, and *OOH intermediates. Increasing the potential applied to the electrode drives the reaction forward until catalytic turnover is achieved.
Experimental volcano plot of OER activity versus descriptor for oxide catalysts
Figure 2b. Experimental/theoretical Volcano Plot. Data from Seh et al.
03

Electrode interfaces form in unpredictable ways during operation

The problem with this dataset is not that anyone made a mistake. It's that the surface has to be assumed to be a non-interacting, crystalline surface made from the perfect termination of the crystal. However, surfaces are dynamic. Catalytic intermediates can interact strongly with the electrolyte, and complex layers form via the dissolution and redeposition of the catalysts back onto the surface. The most extreme example of this is the formation of an amorphous overlayer, which is very hard to calculate the properties of (Fig. 3a). For example, efforts to reduce Ir by moving from IrO2 to SrIrO3 show that an amorphous IrOxHy layer forms (Fig. 3b). We need to measure the behaviour of the surface under operating conditions to understand what is really happening.

Disordered amorphous interfacial layer forming at a crystalline electrode surface under operating conditions
Figure 3. A disordered amorphous layer (~2–5 nm) can form under operating electrode surfaces, separating the crystalline electrode from the electrolyte. This layer might have very different properties to the underlying crystal, and is difficult to model computationally. The only way to understand its binding properties is to measure it under operating conditions.
SrIrO3 crystal structure
SrIrO3 before and after reaction measured by Wan et al.
05

Operando measurements are slow, so the literature is built on sparse data

Setting up any kind of operando measurement requires painstaking work. A carefully chosen model material is synthesised, a cell is assembled, and measurement is typically performed by manually controlling several separate instruments, one of the simplest examples being separately controlling a benchtop UV-vis spectrometer and a potentiostat. In the worst case scenario a user has a 4–6 hour booking and a spectrum at each potential can be acquired every half hour (Fig. 4a). This means only 8–12 spectra can be collected. As it turns out binding energy can be measured by potential-resolved UV-vis spectra (a.k.a. spectroelectrochemistry). However, the sparse data collected in this way is inherently unreliable. 4–6 hours is a very long time for a dynamic surface so systematic bias is inevitable. Sparse points and error limit this format to qualitative conclusions. The solution is to automate the measurement so that a spectrum can be collected every few seconds (Fig. 4b). This produces a reliable dense dataset that can be used to reliably extract binding energies.

Workflow diagram of a single operando spectroelectrochemistry run showing the repeat loop and resulting sparse dataset
Figure 4a. A single operando spectroelectrochemistry run requires material synthesis, cell assembly, and repeated stabilisation–measurement cycles at each applied potential. The result is a handful of spectra and a sparse set of data points that cannot reliably resolve the true underlying functional relationship.
Animated spectroelectrochemistry data collection sweep
Figure 4b. Continuous spectroelectrochemical sweep producing the dense dataset.
06

Automating a series of experiments standardises experiments across labs

Every manual process we perform adds a small systematic bias. This is not anyone's fault; we have our own ways of working and seeing things. However, combined with sparse data and few samples we are left with a literature that is difficult to compare and reproduce. By chaining all our measurements together in a single automated workflow we can standardise the process and remove systematic bias. Combined with rich datasets and combinatorial libraries (where hundreds of materials are made simultaneously) we can finally build a literature that is structured and reproducible enough for deep learning. The goal of this particular project is to measure binding energies of real surfaces in a large-scale, systematic manner to ask a simple question: how well do directly measured binding energies correlate with catalytic performance? If the correlation is strong we can use binding energy as a descriptor to design new catalysts. If the correlation is weak we need to rethink our theoretical framework.

Animated demonstration of automated cell venting and filling procedure
Cell preparation. Automated vent and fill cycle.
Animated overview of the automated operando spectroscopy workflow
Workflow. Full automated measurement sequence.