Machine Learning Enables Optimization of Diamond for Quantum Applications

Spins in solid-state materials, molecules, and other chemical systems have the potential to impact the fields of quantum sensing, communication, simulation, and computing. In particular, color centers in diamond, such as negatively charged nitrogen vacancy NV$^-$ and silicon vacancy centers SiV$^-$, are emerging as quantum platforms poised for transition to commercial devices. A key enabler stems from the semiconductor-like platform that can be tailored at the time of growth. The large growth parameter space makes it challenging to use intuition to optimize growth conditions for quantum performance. In this paper, we use supervised machine learning to train regression models using different synthesis parameters in over 100 quantum diamond samples. We train models to optimize NV^- defects in diamond for high sensitivity magnetometry. Importantly, we utilize a magnetic-field sensitivity figure of merit (FOM) for NV magnetometry and use Bayesian optimization to identify critical growth parameters that lead to a 300% improvement over an average sample and a 55% improvement over the previous champion sample. Furthermore, using Shapley importance rankings, we gain new physical insights into the most impactful growth and post-processing parameters, namely electron irradiation dose, diamond seed depth relative to the plasma, seed miscut angle, and reactor nitrogen concentration. As various quantum devices can have significantly different material requirements, advanced growth techniques such as plasma-enhanced chemical vapor deposition (PE-CVD) can provide the ability to tailor material development specifically for quantum applications.