In this paper, the occupancy of sidewall surface states having a clear signature in the performance of AlGaN/GaN-based self-switching diodes (SSDs) is analyzed using a semi-classical Monte Carlo (MC) simulator in a wide temperature ( T) range, from 100 to 300 K. Experimental I- V curves show an unusual current decrease at low temperature attributed to surface trapping. The dependence on T of the negative surface charge density sigma at the etched sidewalls of the SSDs is essential to explain the measurements. Two devices with different widths (80 and 150 nm) have been characterized and simulated in detail paying especial attention to the modeling of the surface states. At room temperature, MC simulations with a position-independent value of sigma are able to qualitatively reproduce the I- V curves. However, a more complex approach is required to correctly replicate the values and shape of the DC experimental curves at low temperature, below 220 K. An algorithm where sigma depends not only on T but also on the applied bias V is proposed to successfully fit the current values at every bias point. The model is able to explain the physics of the unexpected dependence of the resistance with the channel width and the sign change in the bowing coefficient, the parameters that govern the detection capabilities of the diodes.