Peter Schindler1
Northeastern University1
The discovery of thermally stable materials with surfaces that exhibit an ultra-low work function would allow thermionic energy conversion of heat directly to electricity with high efficiencies and enable next-generation electron emission devices such as THz sources and fluorescent light bulbs. In contrast, surfaces with ultra-high work functions are crucial for devices where large contact barriers are required to suppress electron leakage such as in dynamic RAM and modern transistor architectures. Further, the work function is crucial for band alignment in heterostructures and interfaces. Recently, data-driven approaches based on high-throughput first principles computation have emerged as a new paradigm to facilitate the search through vast chemical spaces for new materials with tuned properties. Most material databases largely lack to report surface properties like the work function and surface energy, as each bulk material typically has dozens of distinct low-index crystalline surfaces and terminations.<br/>Here, we report on recent progress in our high-throughput workflow using density functional theory (DFT) to calculate both the work functions and the cleavage energies of over 55,000 surfaces that we created from ~3,700 bulk materials (with a zero bandgap and including up to ternary compounds). Moreover, we developed a physics-based approach to design surface descriptors and established a surrogate machine learning model to predict the work function. Our machine learning model achieves a mean absolute test error 4 times lower than the baseline, comparable to the accuracy of DFT. This surrogate model enables rapid predictions of the work function (~10<sup>5</sup> faster than DFT) across a vast chemical space. This facilitates the discovery of metallic surfaces that have an extreme work function but also a low surface energy paving the way for new materials solutions in thermionic energy conversion, electron emission devices, and contact electronics.