The point of the device model is to run transport calculations, not stability. What this means is, unless you suspect that the stability is somehow changed by applying a bias to the device, you should just carry out the stability analysis on the material itself, in a usual bulk or molecule configuration. The same typically goes for optimizing the geometry that goes into building the device. Unless the bias would change the atom positions significantly, we always just create the structure under no bias (in which case we don't need a device configuration or NEGF, which is much slower).
