Dear Juan,
I check the DriftDiffusionInitialSolution in the diode example and find you setting the initial guess of n and p to IntrinsicElectrons and IntrinsicHoles.
Why not set the like:
In N-type side: n = abs(NetDoping), p=n_{i}^{2}/NetDoping
In P-type side: p = abs(NetDoping), n=n_{i}^{2}/NetDoping
I think this initial guess is more close the physics. But it is hardly converged when I change the initial guess.