The Nano-Electronic Simulation Software (NESS) features an improved model of Interface Roughness (IR), accounting for correlation lengths in two perpendicular directions and allowing anisotropic roughness. IR in \(\text {GaAs/Al}_{0.3}\text {Ga}_{0.7}\text {As}\) Resonant Tunnelling Diodes (RTDs) was investigated using both the previous and improved models, with 4 correlation lengths ( \(L_C\) ) ranging from 2.5 nm to 10 nm. For each correlation length, 25 RTD device structures with IR were randomly generated. Device variation was quantified as the standard deviation of the resonant peak current ( \(I_r\) ) and the corresponding bias voltage ( \(V_r\) ), both extracted from the non-linear RTD current-voltage (IV) characteristics. The improved model resulted in greater variation, increasing standard deviation from 6.2 mV and 9 nA to 24.2 mV and 34.7 nA for \(L_C=2.5\) nm. Standard deviation also roughly doubled as \(L_C\) increased from 2.5nm to 10nm, increasing from 6.2 mV and 9 nA to 11.7 mV and 18.8 nA for the previous IR model, and from 24.2 mV and 34.7 nA to 38.8 mV and 80.9 nA for the improved IR model. A further study of anisotropic correlation lengths resulted in variation of standard deviations. This paper hence shows the importance of simulating IR with two correlation lengths for future accurate RTD research.
