Exploration of exfoliation, functionalization and properties of MXenes via first-principles and machine learning

Abstract

The monolayers of early transition metal carbides and carbonitrides named MXenes are exfoliated from the corresponding bulk MAX phases (Mn+1AXn, M = early transition metal, A = group IIIA or IVA element and X = carbon and/or nitrogen), where MX layers are interleaved with “A” atoms. The selective etching of the A element from the MAX phases using the aqueous solution of HF as the etchant causes functionalization of the surface of MXenes, which are represented by Mn+1XnT2, (T=F, O, H and/or OH). According to experimental reports, the uncontrolled, non-uniform and mixed functionalization of MXene cause the biggest challenge in the isolation of pristine MXene. Using the first-principles calculations, we have carried out a comprehensive study for isolating pristine MXene using Nb4AlC3 MAX phase. The calculated bond-dissociation energy (BDE), density of states (DOS) and electron localization function (ELF) show that the presence of LiF instead of commonly used HF in MAX phase facilitates the isolation of pristine Nb4C3 MXene. Almost all of the MXenes are non-uniformly functionalized therefore, to get an ordered functionalization of MXene, the role of chemical potentials of all the constituents need to be determined. We extended our study and carried out a comprehensive investigation of exfoliation process of Ti3AlC2 to Ti3C2 MXene via HF insertion. Spontaneous dissociation of HF and subsequent termination of edge Ti atoms by H/F weakens Al−MXene bonds. A consequent opening of interlayer gap allows further insertion of HF that leads to the formation of AlF3 and H2, which eventually come out of the MAX, leaving behind functionalized MXene. Thermodynamic analysis shows that depending upon the chemical potentials, along with full F-termination, mixed and non-uniform functionalization of MXene can also be stabilized. Unlike other functional groups, O-termination opens the band gap in otherwise metallic MXenes. We showed that depending upon the position of the O atoms, two phases namely BB′ and CB are possible. Using the insights from charge transfer, DOS, and ELF, the stability of BB′ and CB phases is explored as a function of M atom in MXene. Further, due to the absence of inversion symmetry, the CB phase of Sc2CO2 MXene possess intrinsic dipoles. A monolayer having both ferroelectric and new-unreported antiferroelectric low-energy configurations is obtained for the first time. The reasonable polarization switching barrier ensures the potential of this material for nonvolatile memory applications. Furthermore, as we go from the monolayer to bilayer, interestingly, the transition from insulator to a nondegenerate 2D electron/hole gas system takes place. As there are various choices of functional groups available, an enormous number of MXene can be generated. Characterizing all with the conventional methods would be very time-consuming and inefficient. Hence, we utilized the machine-learning (ML) based approach to predict the various properties in an accelerated manner. A total of 23,870 MXenes are generated and stored in a functional materials database named “aNANt”. Using ML based classification model, the metal-semiconductor classification is carried out with an accuracy of 99%. Further, the Gaussian process-based regression model is developed to predict the band gap of these MXenes with GW level accuracy. The application of ML based approach is extended to accurately position the GW band edges at an absolute scale, which are predicted with a minimal root mean square error of 0.12 eV