Carbon-Doped GaN on SiC Materials for Low-Memory-Effect Devices

Abstract

I. INTRODUCTION Gallium nitride (GaN) high electron mobility transistors (HEMT) are today used in military radar and telecom applications. However, there are still several issues that remain to be solved primarily on the material side. The best performing HEMT devices today are those produced on GaN on silicon carbide (SiC). SiC is used on account of its high thermal conductivity and relatively close lattice match with AlN and GaN. Thermal management of the devices is an issue that needs to be improved and, especially for telecom applications, the memory effect must be reduced in order to simplify linearization of the systems. We present epitaxial GaN on SiC of extraordinary high quality where the thermal barrier to the SiC substrate is minimized and where devices made on this material exhibit a substantially lower memory effect than commercially available devices today. II. RESULTS The epitaxial growth of GaN on SiC substrates starts with the deposition of an aluminium nitride (AlN) wetting layer in order to ensure that the GaN grows in a two-dimensional fashion. The structural quality of the AlN is often rather poor and gives rise to a thermal barrier resistance (TBR) that accumulates heat in the device. This results in a decrease in the device efficiency but may result in a catastrophic thermal breakdown of the device in the worst case. We have optimized the in-situ pre-treatment of the substrate surface and also the initial growth procedure of the AlN so that the AlN layer is grown in a layer-by-layer mode, resulting in a very high structural quality of the AlN and also the subsequent GaN that is grown on top of the AlN. The typical values of the full width of half maximum of XRD rocking curves for the C-doped GaN buffer and the AlN (102) planes are around 200 and 100 arcsec, respectively. The absence of pores in the AlN layer also allows this layer to be made very thin which further reduces the TBR of the structure. Figure 1 shows atomic force microscopy images of a non-optimized (left) and an optimized (right) growth of the 35 nm-thick AlN nucleation layers. The entire growth has been successfully scaled up to 100 mm semi-insulating SiC substrates. The non-uniformity over a 100 mm epiwafer of the total thickness and the sheet resistance are typically less than 1% and 2%, respectively. High electron mobility of ~2100 and ~2300 cm2/V-s was achieved from the 2 nm-thick GaN cap/14 nm-thick Al0.29Ga0.71N/GaN structures with and without a 1 nm-thick AlN exclusion layer, respectively, which is as a result of the low dislocation density and sharp channel interface. On the device level, the distortion of amplitude modulated (AM) signals due to memory effects induced by deep electron traps with long time constants, imposes the practice of linearization schemes using digital pre-distortion which are impractical if the time constants are long, requiring large computational capacity and memory. Reducing the amplitude and time constant of the memory effects will greatly improve the AM performance and reduce the cost to produce the systems. The origin of the trapping is not fully understood however it can be reduced by exchanging the dopant in the buffer layer from Fe, which is the standard dopant, to C. In figure 2, the memory effect is illustrated in a commercial HEMT device with Fe doping and a device produced using C doping in the buffer. The measurement is done by applying a 1 µs long voltage pulse between the drain and source from a quiescent working point. Unfortunately the change of dopant also carries the price of a reduced output power from the device. This issue is currently addressed and a solution is close. III CONCLUSIONS We present material enabling high-power and high-frequency GaN devices to be made, which have substantially better performance and suits the use of GaN HEMT devices in mobile base stations. We highlight here in particular the low TBR, the high mobility, and the 50 times lower memory effect. Figure 1