Talk abstract

AlGaN/GaN HFETs offer 5-10 times higher power at microwave frequencies than conventional solid-state devices. The conventional structure relies on piezoelectric and spontaneous polarisation to form the 2DEG, with no need for modulation doping of the structure. A key feature of the structure is that there should be no parallel conduction through the GaN layer. Experimentally it is found that some growth conditions lead to insulating GaN. However, it is not entirely clear what defect or mechanism leads to this insulating material [1]. This paper indicates that transport in this insulating layer can have both n- and p-type nature, and can potentially be explained by the high dislocation density of the material.

An HFET layer structure of 28nm Al0.23Ga0.77N on ~1µm GaN (all undoped) was grown by MOVPE in a Thomas-Swan Epitor reactor. The substrate was a conducting, nitrogen doped, n-type, 4H, on-axis SiC wafer from Cree Inc. There is intense interest in growth on SiC because of its high thermal conductivity compared to sapphire. The conducting nature of the SiC allows a substrate bias to be applied to the device giving the opportunity to change the 2DEG concentration and obtain information about the electrically active centres in the GaN.

Temperature dependent conductivity measurements made between adjacent etched mesa isolated devices, showed that the conductivity of the GaN layer was ohmic and extremely low (typically <1nS at 350°C), with an activation energy of 0.9eV. Similar conductivity behaviour and activation energy was obtained with HFET layers grown on sapphire. The fact that the conduction between two isolated 2DEG regions was ohmic suggested that the material was n-type, with transport likely to be via conduction band electrons with a compensated deep level at 0.9eV. The conductivity was too low for us to carry out Hall measurement to check carrier type.

The effect of varying substrate bias on the HFET gate transfer characteristic has been studied and showed a shift in pinch-off voltage as a result of a reduction in 2DEG carrier density as the substrate bias was made more negative. The pinch-off voltage was found to vary with the square root of the substrate bias, suggesting that a depletion region of roughly constant doping with exposed negative charge is being formed in the GaN. The fact that the substrate leakage was insignificant (nA level), and the presence of a depletion region implies that the GaN layer has a p-type nature, with an acceptor doping of 1.6x10^17cm-3. If the substrate had an n-type nature, space charge limited behaviour would normally result and depletion would not be seen.

One model which could reconcile these observations relies on the inhomogeneous nature of the GaN layer. AFM indicates that these layers typically have a dislocation density of ~10^9cm-2. It is expected that the dislocations and grain boundaries can accept high concentrations of electrons [2]. Since these dislocations will extend vertically from the 2DEG to the GaN/SiC interface, it is possible that hole conduction can occur along the dislocation cores, with tunnelling, or defect mediated transfer, across the GaN/SiC interface to the SiC conduction band, allowing the GaN to appear p-type for vertical transport. If there is a background of acceptors at 0.9eV below the conduction band and a smaller concentration of compensating donors in the bulk of the GaN, the material would behave as n-type for conduction between mesas. Where a dislocation core intersected with a 2DEG, there would be a small region surrounding the dislocation which would be depleted, resulting in a small reduction in the effective mobility.

This work was carried out as part of Technology Groups 7 and 9 of the UK MoD Corporate Research Programme. © Crown Copyright 2001

[1] P. Kordos, M. Morvic, J. Betko, J. M. Van Hove, A. M. Wowchak, and P. P. Chow, J. Appl. Phys. 88, 5821-5826 (2000). [2] I. Shalish, L. Kronik, G. Segal, Y. Shapira, S. Zamir, B. Meyler, and J. Salzman, Phys. Rev. B 61, 15573-15576 i (2000).