GaN growers model their way to higher quality materials
One process where this knowledge can yield great benefits is MOCVD. How will a new reactor design affect the way the precursors and carrier gases flow in the growth chamber? What are the best growth parameters to start with when developing a new epitaxial structure? These are difficult questions to answer, but equipment manufacturers and growers are making increasing use of computer modeling to gain some insights.
The use of such computational techniques has been helped in no small way by the advent of powerful number-crunching desktop computers. Computational fluid dynamics (CFD) is no longer synonymous only with custom-written software and room-sized supercomputers. Today s top of the range personal computers are sufficiently powerful to model the internal workings of an MOCVD reactor and modeling software is commercially available, bringing CFD to a wider audience.
Saving time and money
The use of modeling techniques helps to reduce development time and cost by removing some of the guess work and getting as near as possible to the right answer first time. "At the present level [of reactor design and complexity] it is impossible to design new systems without modeling. The trial and error approach is too expensive," said Alex Gurary, head of the MOCVD reactor development group at Emcore.
If an accurate model for a particular reactor configuration can be made, then for a given set of process conditions, such as temperature and pressure, it becomes possible to model the fluid dynamics of the system. This will indicate if laminar flow over the substrate is maintained or if undesirable effects such as gas recirculation are more likely.
The accuracy of the output from a computer modeling exercise depends on the quality of information input into the model. Understanding the chemistry of the system is particularly important in MOCVD, where the reactions between different metalorganic chemicals and the reaction by-products affect the fluid dynamics and therefore the epitaxial quality. For established compound semiconductors such as GaAs the chemistry is well understood, allowing the modeling of processes with considerable accuracy.
"For GaAs we can predict parameters like growth rate and uniformity [for a given set of reactor conditions] and use them for process development," said Gurary. "For other materials the chemistry is less well understood. This is particularly true for GaN, where different adducts are formed in the reaction stage that can change the situation dramatically, especially for AlGaN and InGaN. At present everyone is at the stage of trying to understand the chemistry of GaN and its alloys and learning how to model the process."
Modeling to avoid turbulence
Kazuhiro Ohkawa and colleagues at the Tokyo University of Science are one of the groups of researchers involved in this learning process. Ohkawa s group has received funding through the Asian Office of Aerospace Research and Development (see AOARD box) to model the MOCVD growth of GaN. Ohkawa s recent work provides a good example of what can be done using computer simulations. The group is using a commercially available CFD code to look in three dimensions at a Nippon Sanso reactor during GaN growth under different reactor configurations and growth conditions. The group is able to correlate the modeled conditions in the reactor with the actual properties of GaN grown on sapphire when the conditions used in the model are applied to real growth situations.
The reactor is designed to give three possible gas-flow configurations (figure 1). In the one-flow setup, all of the gases, trimethyl gallium (TMG), ammonia and hydrogen, are fed in at 7° to the horizontal plane of the susceptor, with no gas flow from the subflow nozzle. In the two-flow configuration, the TMG and ammonia are fed in horizontally, while the reactor subflow carries hydrogen or nitrogen in vertically to suppress thermal convection from the hot susceptor. In the three-flow method, the horizontal main feed is split into two channels that feed the TMG and ammonia into the chamber separately.
The group performed growth under various conditions of temperature and V/III ratio. The reactor pressure was kept to 100 kPa and growth time was one hour for all experiments. GaN was deposited directly onto 2 inch sapphire wafers without substrate rotation. Growing without a low-temperature buffer layer removes the buffer layer s influence on the quality of the GaN subsequently deposited, enabling a direct correlation between the growth conditions, the modeled gas flows, the layer quality of GaN and its electrical properties. As a result, all the GaN layers had the hexagonal pyramid morphology characteristic of GaN grown onto sapphire without a buffer layer.
Figure 2 shows the output from a CFD simulation for the gas flows in the three configurations described above. For the one-flow case, a region of turbulence is seen above the susceptor. This is as a result of the gases becoming hot due to thermal convection from the heated susceptor. The gases decompose and react in the space above and not on the substrate surface. These reacting gases can be directly observed as smoke in the reactor. The growth rate in this case is very low at only about 1 µm/h and the resulting layers take the form of GaN islands.
Introducing the subflow suppresses the thermal convection (figure 2b) producing smoother gas-flow patterns in the CFD simulations for the two- and three-flow methods. The simulation in figure 2c reveals that some recirculation is occurring in the three-flow configuration. Precise gas control is needed to maintain laminar flow over the substrate, and the operating window for this flow condition under three-flow growth is very narrow.
