Migration-enhanced MOCVD advances AlGaN performance

In particular, these mainstay film-growth methods of the industry are not exactly ideal for nitride deposition, an application area that will inevitably become an increasingly important part of the III-V business. Now, a related technique known as migration-enhanced (ME) MOCVD promises to solve some of the problems encountered with MBE and MOCVD, especially for making devices that have a high aluminum content, including AlGaN HEMTs and UV-LEDs.

The key difference between ME-MOCVD and conventional film growth is the way in which gases are introduced into the reaction chamber, which results in fewer collisions between metalorganic and ammonia gas molecules. The upshot is faster growth speeds and a lower density of screw dislocations, which are detrimental to device performance.

When used for III-nitride manufacture, MBE and MOCVD differ substantially in regard to the source materials used, and fundamentally in terms of their source-material transport mechanisms. These different mechanisms can be described by the so-called Knudsen parameter (Kn), which is the mean free path of the gas to reactor dimensions ratio.

The mean free path of a gas is the average distance traveled by a molecule or atom before it collides with another. For MBE, Kn is usually close to, and sometimes greater than, unity, meaning that source molecules can reach the reaction surface without experiencing any collisions. This is possible because a low reactor pressure is maintained during MBE.

On the other hand, Kn for CVD is typically in the 10-2 to 10-3 range (for low-pressure CVD) or 10-5 to 10-6 (for atmospheric-pressure CVD). This means that a large number of collisions will take place before a molecule reaches the reaction surface. As a result, CVD material transport is actually governed by fluid mechanics. Important effects include: bulk transport toward the reaction surface by diffusion; diffusion through a stagnant boundary layer that is formed above the reaction surface; and surface diffusion and incorporation/decomposition.

So, depending on the conditions used, film growth is limited either by gas-phase transport or surface-reaction kinetics. Optimized nitride growth with MOCVD is limited by gas-phase transport, except for the low-temperature deposition of the nucleation layer. In this region the growth rate and layer composition can be controlled by adjusting the flow of source material at the inlet.

The stagnant boundary layer plays an important role in the transport process. Its thickness directly affects the source materials composition gradient in the stagnant layer, which in turn determines the diffusion fluxes of incorporation and decomposition. In short, the stagnant boundary layer can effectively alter film growth and decomposition rates. Being able to play with the dimensions of the stagnant boundary layer allows more growth control in MOCVD than in MBE.

The small Kn associated with nitride growth by MOCVD can interrupt the epitaxial process because of the increased propensity for gas-phase reactions. The metalorganic source materials, especially the widely used trimethylaluminum (TMA), readily react with ammonia molecules when collisions between the two take place.

These reactions produce small particles that may fall onto the reaction surface and then act as new, unwanted nucleation centers that terminate epitaxial growth. The result is surface roughening and extended defect generation. During AlGaN growth by MOCVD, the gas-phase reactions clearly manifest themselves, causing "V" defects on the surface.

These reactions also limit the nitride growth rate possible with MOCVD because increasing the influx of source materials will result in more surface defects. Another problem is that the strong aluminum-nitrogen bond allows only a very short surface-diffusion length. This greatly reduces the ability of aluminum and gallium adatoms to incorporate into energy-favorable sites, which leads to random growth of 3D islands and deterioration of epilayer quality.

The thinking behind ME-MOCVD is to separate ammonia and metalorganic molecules, which ought to reduce the likelihood of gas-phase reactions and enhance adatom surface migration. At Sensor Electronic Technology (SET), we can optimize source waveforms and overlaps to refine material quality. And unlike other flow-modulation growth techniques, ME-MOCVD allows the freedom to work with the stagnant boundary layer.

As mentioned earlier, the stagnant boundary layer has a significant effect on material incorporation and decomposition. The layer thickness depends on the flow of gas into the reaction chamber and is proportional to the square root of the kinetic viscosity of the gas divided by its velocity. During ME-MOCVD AlGaN growth, ammonia has the greatest modulated flow. So, AlGaN growth with ME-MOCVD presents two options.

The first is to switch off the flow of ammonia and simultaneously switch on a compensation gas, for example nitrogen or argon. The kinetic viscosity and velocity of the gas can be changed to produce a consistent boundary-layer thickness, or to make the layer thinner. If the layer is made thinner, metalorganics can diffuse through the boundary layer to reach the growth surface more quickly, maintaining a fast growth rate and greatly enhancing surface migration. However, because the V/III ratio is vanishing and the stagnant boundary layer is therefore becoming thinner, decomposition is also enhanced. As a result, point defects - mainly nitrogen vacancies - become more likely.

The second option is to switch off the ammonia flow without introducing any compensation gas. This means that the main chamber flow quickly becomes dominated by hydrogen, which has a much larger kinetic viscosity than the hydrogen/ammonia mixture used in MOCVD and thus reduces the gas velocity. This approach creates a thicker stagnant boundary layer, which can still enhance surface migration (because of the vanishing V/III ratio), but at a lower decomposition rate.

We have found ME-MOCVD to be efficient in improving AlN and AlGaN material quality. Under optimum growth conditions a step-flow mode was used to grow AlN on vicinal (0001) sapphire substrates (figure 1), which enhanced aluminum-adatom surface migration.

The growth rate for AlN with ME-MOCVD was three times as fast as that seen with conventional MOCVD under the same flow of TMA. Comprehensive characterizations also revealed a superior material quality and, in particular, demonstrated a reduction in the density of screw dislocations.

Figure 2 shows a TEM micrograph of SET s patent-pending UV-LED buffer structure, which consists of an ME-MOCVD-grown AlN and AlN/AlGaN superlattice, and a thick layer of silicon-doped AlGaN. The screw dislocation density in the structure, starting from the thin AlN layer, is only 3 x 108 cm-2. This is about one-hundredth the screw dislocation density of conventional MOCVD-grown, thin AlN.

So, as well as faster deposition, AlGaN layers grown by ME-MOCVD clearly have better structural quality. Among other evidence that we have obtained through different characterization techniques, the minority carrier lifetime data is of great importance, since this is directly related to the light-emission efficiency of the device. Figure 3 shows a comparison between the lifetime measurement data obtained through ME-MOCVD- and MOCVD-grown AlGaN layers with an identical aluminum fraction and thickness.

The lifetime of AlGaN grown by ME-MOCVD is about six times as long as that produced by MOCVD (190 ps versus 30 ps), indicating a far lower incidence of non-radiative-decay processes and consequently higher light-output efficiency. At SET, we first developed ME-MOCVD for a single 2 inch wafer chamber. More recently, we have successfully scaled this technology to a multiple-wafer reactor (3 x 2 inch) set-up.

SET is actively working on III-nitride technology for two types of device, namely deep-UV-LEDs (figure 4) and HEMTs. Although these devices require different substrates, ME-MOCVD benefits both in the sense that it builds a much higher quality buffer on which the device structures reside.

The high-quality AlN and AlN/AlGaN superlattices enhance the n-AlGaN mosaic block dimensions (a larger mosaic size equates to a better epilayer quality), which directly reduces tensile strain between AlGaN blocks and alleviates cracking in thick, conducting AlGaN layers. This development has resulted in increased power output from deep-UV-LEDs, which have an emission peak at around 280 nm at a drive current of 20 mA. SET began shipping these LEDs in August 2004. The current devices have a continuous-wave power exceeding 1 mW at 20 mA. Our plan is to produce deep-UV-LEDs with a wall-plug-efficiency of more than 2% in the next two years.

On the microelectronics side, we used ME-MOCVD instead of MOCVD to fabricate the AlN transition layer between a semi-insulating 4H-SiC substrate and GaN. The AlGaN cap layer has also been grown by ME-MOCVD. Using the new technique greatly improved HEMT wafer quality and device performance. For example, the transistors showed reduced drain-current collapse and increased gate-drain breakdown voltage. SET is now capable of supplying 2, 3 and 4 inch GaN HEMT wafers, and the multiple-wafer ME-MOCVD chamber has greatly enhanced our wafer-production capacity.