The tight-binding model (model Hamiltonian) used in that paper is slightly different the semi-empirical tight-binding mode (extended Hückel method and Slater-Koster method). In the former case, the TB parameters are usually obtained by the maximally localized Wannier functions of first-principles calculations. For the latter cases, the parameters are usually obtained by the fitting procedure.
You can't use the TB parameters of that paper to ATK.
