Ammonia plasma passivation of GaAs in downstream microwave and radio-frequency parallel plate plasma reactors

The poor electronic properties of the GaAs surface and GaAs–insulator interfaces, generally resulting from large density of surface/interface states, have limited GaAs device technology. Room-temperature ammonia plasma (dry) passivation of GaAs surfaces, which reduces the surface state density, is investigated as an alternative to wet passivation techniques. Plasma passivation is more compatible with clustered-dry processing which provides better control of the processing environment, and thus, improves interface integrity. Passivation was monitored in real-time and in situ using photoluminescence (PL). In addition, the passivated surfaces are inspected using x-ray photoelectron spectroscopy. Passivation with two different plasma excitation methods, downstream microwave (2.45 GHz) and rf (13.56 MHz) parallel plate, are compared, and effects of operating parameters such as pressure, flow rate, and power are examined. In both methods plasma-generated H atoms reduce the surface state density by removing excess As and As2O3 during the first few seconds of the plasma exposure. This step is followed by formation of Ga2O3 which takes place on a longer time scale (5–10 min). While the final passivation result appears to be similar for both methods, surface damage by ion bombardment competes with passivation in the parallel plate method, reduces the PL yield and adversely affects the long term stability of the passivated surface. Although it is common to heat the sample during passivation, we show that NH3 plasma passivation is possible at room temperature without heating. Low-temperature processing is important since passivation can be done at the end of device processing when it is undesirable to expose the device to elevated temperatures. The absence of ion bombardment damage combined with efficient generation of H atoms in the downstream microwave treatment, make this scheme a preferred dry passivation process, which could be easily and inexpensively clustered with existing GaAs processes