It is shown that the inclusion of dopant segregation, negative differential mobility and thermal assisted tunnelling to the calculation of dark current activation energy provides a much better fitting to the experimental results than geometrical or Poole-Frenkel models and handles well its voltage asymmetry and nonlinearity. Besides, the transport model obtained is able to fit the current-voltage curves for both bias polarities with almost the same transport parameters. A practical method to estimate the segregation length is discussed. A consequence of this method is a more accurate set of values for the transport parameters of the structure, considering direct and reverse bias.