Silicon (in the form of lithium silicides) has almost ten times the theoretical charge storage capacity of graphite, the anode material used in most commercially-available lithium-ion batteries. Replacing graphite with silicon therefore promises a substantial improvement over the state-of-the-art in electrochemical energy storage. However, it has proved difficult to realise this high theoretical capacity in a practical electrochemical cell and maintain it over repeated charge-discharge cycles. This dissertation presents experimental work probing the changes in local structure occurring during the electrochemical reactions of lithium with silicon, using neutron total scattering and nuclear magnetic resonance, together with novel processing methodologies for analysing the resulting data, in the hope of suggesting ways of improving the performance of silicon-based lithium-ion batteries.

Neutron total scattering patterns were obtained from silicon-based anode materials extracted from cells at various states of charge. These samples were composed of a heterogeneous mixture of amorphous, crystalline and disordered crystalline materials. Reverse Monte Carlo is a technique for obtaining structural information from experimental data (particularly total scattering patterns) from amorphous and disordered crystalline materials. However, previously existing Reverse Monte Carlo software could only handle homogeneous materials. Therefore, the RMCprofile software package was extended to handle data from heterogeneous samples. The improved RMCprofile was applied to the aforementioned total scattering patterns, but the much stronger scattering from the other components (themselves not well-characterised) swamped that from the lithium silicide. Future work should attempt to reduce the scattering from the inactive components, particularly the hard-to-model incoherent scattering.

NMR data were acquired in situ from silicon-nanowire-based lithium-ion batteries during repeated charge-discharge cycles, achieving much better electrochemical performance than had been seen in previous in situ experiments with silicon. Owing to the large quantities of data obtained, an automated, model-free dimensionality reduction technique was needed. The NMR data were processed using principal component analysis and a variant of non-negative matrix factorisation. With both of these methods, one of the components was found to be associated with high voltages vs. $Li|Li+$ (i.e. a fully discharged anode). This region has seen very little interest by comparison with the low voltage (high levels of lithiation) region of the charge-discharge cycle, so this discovery suggests a new avenue for future research.