I guess the first thing is to check if the structure of your device is the same as in the reference paper. The second thing is that you have to use the computational settings (density functional, k-point sampling) for DFT calculations that are similar to that in the paper, and also do convergence tests with respect to basis set and density mesh cut-off.
Note that the paper uses all-electron code to do the DFT calculations, whereas QuantumATK makes use of norm-conserving pseudopotentials, treating core electrons with an effective potential (called pseudopotential). So that you might need using accurate pseudopotentials such as SG15.
