Automated design of thermally stable heteroleptic precursors by computational screening
Dr. Simon Elliott
Director of Atomic Level Process Simulation – Schrödinger Inc.
Ligand choice in ALD/CVD precursors is crucial for throughput, stoichiometry, impurities and process temperature. Heteroleptic precursors (containing more than one type of ligand) are one way to compromise between conflicting chemical requirements. However to date we have barely ‘scratched the surface’ of the vast chemical space of possible heteroleptic precursors. As an example to illustrate the magnitude of the design problem, 8,855 chemically-distinct complexes can be formed by combining four of 20 ligands around a tetravalent metal center. Clearly, an exhaustive experimental analysis is not possible. Instead we look to computational screening to narrow down the search to the most promising options. Here we present a computational approach for screening metal precursors with respect to thermal stability. The computational strategy is illustrated on the example of Zr precursors for zirconium nitride, used as a hard coating to protect industrial parts in corrosive environments.
CVD precursors should decompose unimolecularly at elevated temperature in the reactor and we therefore seek complexes with moderately low energies for homolytic bond dissociation (i.e. into a pair of radical fragments). ALD on the other hand requires the gas-phase stability of precursors to be as high as possible, as this dictates the upper temperature limit for ALD.
We first enumerate over ligands. Even a small ligand library of just six O-free, N-bearing ligands (amines, amidinates, cyano and guanidinate, with various alkyl groups) gives 6^4=1296 possible complexes, but many are symmetrically equivalent or over-coordinated, leaving 81 for optimization with density functional theory (DFT). To study thermal stability requires a second phase of enumeration, over the 1280 different bonds that can be broken in these complexes, yielding a set of radical fragments that are also computed at the DFT level. It is clear that every step in this computational workflow requires robust and efficient automation.
DFT reveals that the lowest bond dissociation energies are obtained for cleaving intact ligands from the metal. The least thermally stable complexes are found to be zirconium amides with bidentate amidinates/guanidinates, and so we predict that these would be the best CVD precursors. We observe very little steric effect when replacing methyl groups with ethyl groups. By contrast, cyano groups have a stabilizing influence, suggesting that similar electron-withdrawing groups would be useful as spectator ligands in heteroleptic precursors for ALD. On the basis of this sample system, we discuss the general requirements for chemical enumeration software and the limits faced by automation.