The advanced state of MOCVD for UHB-LEDs
12/01/2004
Fabrication of the multicomponent structure of ultrahigh-brightness light-emitting diodes (UHB-LED) with metal organic chemical-vapor deposition (MOCVD) requires accurate control of layer thickness to better than 1%, 1% composition uniformity over the substrate surface area, and control of the heterostructure interface quality and sharpness, the latter on the order of one monolayer. The light-producing structure of UHB-LEDs often involves 50-60 deposited layers with different doping concentrations and layer thicknesses and all depositions done sequentially during a process that can take up to 5 hr in one batch reactor load.
 |
Despite the sophistication of UHB-LEDs, from a product and application perspective, these are commodity components that benefit from any improvement in mass production. While today’s advanced MOCVD reactors can achieve the required deposition specifications for producing UHB-LEDs, continuous improvement can be made in large-area deposition, through increasing the reactor size, substrate size, or both.
Advanced MOCVD for LED manufacturing
MOCVD uses a carrier gas flow containing a dilute mixture of metal organic precursors. The gas mixture flows into a reactor chamber at 50-500torr where substrates are at 600-800°C for conventional III-V materials, and as high as 1200°C for GaN. The reactive gases decompose and deposit thin epitaxial layers of III-V materials (e.g., AlGaAs, InGaAsP, InGaN, etc.) from a few nanometers to a few microns thick, as required.
 |
Advanced MOCVD reactors used for production of UHB-LEDs (see table) are heated with lamps or resistance heaters, or inductively heated from a coil that is underneath the susceptor. In Aixtron reactors, deposition for batch volume production is accomplished via radial flow of reactants from the center of a reactor’s susceptor to the outer rim, while wafers undergo “planetary” motion - a method that enhances thickness and uniformity by averaging deposition.
The precursors used for MOCVD are relatively expensive, and the deposition utilization efficiency of MOCVD reactors is ~50% (i.e., ~50% of the atoms injected into the reactor are deposited on substrates as thin films). This is close to the theoretical limit and is principally controlled by reactor design, which leaves little room for improvement - perhaps only 5% as reactor designs are further enhanced.
Throughput with MOCVD is very nearly independent of the substrate sizes used. The viable paths to higher throughput involve continually decreasing cycle time through faster, more accurate temperature ramping and control; improved maintenance and service procedures; and greater tool reliability.
CFD modeling enables advanced MOCVD
Use of computational fluid dynamics (CFD) modeling has played a significant role in understanding and developing advanced production-scale MOCVD reactors that deliver the layer and interface control required today. CFD is one way to improve reactor design and processing, shorten development cycles, and make more efficient use of development and testing costs.
Overall, the computational model of the MOCVD growth process is based on first principle prediction of flow, heat, and mass transfer by solving the corresponding transport equations on 2D or 3D domains supplemented with appropriate boundary conditions. The MOCVD process model, however, not only involves Navier-Stokes Flow (usually incompressible low-pressure internal flow) - the classical realm of CFD - but multidisciplinary physics and chemistry processes that occur during MOCVD.
In all attempts to model MOCVD reactors, the accurate prediction of detailed heat transfer in the process environment is the most crucial prerequisite for the determination of species distribution, as well as layer thickness and composition. The chemical element’s diffusive properties, gas phase, surface reaction rates, and surface incorporation are sensitive to the thermal field distribution in the equipment. Thermal radiation appears to be the governing mode of heat transfer due to the high substrate temperatures and large temperature differences involved.
 Figure 1. Calculated 2D temperature distribution in an Aixtron MOCVD Planetary Reactor used for high-temperature GaN deposition. |
For example, we have found that designing MOCVD reactors requires a clear understanding of the relation between electromagnetic fields and heat transfer where planetary movement is employed. Also essential is knowledge of the distribution of induction-power density. Characterizing these parameters using CFD leads to a reactor design that achieves superior temperature distribution (Figs. 1 and 2).
 Figure 2. An Aixtron Planetary Reactor loaded with 24 substrates. |
With such knowledge derived from computer modeling, we have been able to design reactors that are capable of a ±1° surface temperature uniformity in the growth zone; this level of temperature uniformity is essential for the successful species incorporation and resulting materials properties during the MOCVD growth process of multinary layers (e.g., InGaN, InGaAsP, etc.).
It is known that another crucial issue for process results is the design of gas delivery and injection into the process chamber. Design issues, which are specific to a given precursor, include the appropriate injection of growth-limiting group III species into the deposition zone; preventing deposition at the center of the susceptor and upstream of the wafers; the delayed mixing of group III and group V species to avoid any risk of premature reaction; and the absence of any vortices.
Future efforts
A state-of-the-art MOCVD reactor can accommodate 49 2-in. substrates, a seven-fold improvement from the first production reactor configurations, which were only capable of processing seven 2-in. substrates
un. While it has been proven that making a reactor capable of handling ~100 2-in. wafers is possible, it is very unlikely that LED manufacturers will venture beyond the deposition area equivalent to 50 2-in. wafers, simply to control the risks associated with the value of LED substrates in one production run.
The industry is working to advance beyond 2-in. substrates (i.e., to 6 in.), but substrate size almost certainly will not evolve as it has with silicon-based manufacturing, because of the ongoing limitations driven by GaAs and sapphire substrate fragility, and the restrictions dictated by the cost-effectiveness of established 2-in. LED fab lines and tools.
Work is also being conducted to develop an alternative substrate for GaN-based UHB-LEDs. Currently, the most commonly used substrate is sapphire, which is prone to too many defects in the subsequent GaN layer, precipitated by lattice mismatch, that limit the performance of the resulting UHB-LEDs. Two approaches are generally recognized to solve this problem: 1) a cheaper substrate material than sapphire, and 2) a substrate that increases performance.
The dream for a cheaper substrate material is focused on silicon. Using our experience in epitaxial growth, we are currently working with various universities and institutions in Germany and Belarus to develop GaN growth on silicon. Using silicon would allow the industry to manufacture more cost-effective LEDs, which is a prerequisite for introducing LEDs into general lighting applications. While this lab-scale work is promising, the performance (i.e., brightness, forward voltage, and ESD) of LEDs fabricated on silicon, has not achieved the levels obtainable on sapphire substrates. The belief, however, is that the required performance will be achieved eventually through more device optimization and engineering work.
The other approach involves developing a substrate that increases performance - mainly a GaN single-crystal wafer, onto which MOCVD is used to grow homoepitaxial GaN layers. Specifically, the application is for blue laser diodes, but it is also applicable for UHB-LEDs if the cost of these new substrates becomes compatible. Today, a 2-in. GaN wafer costs ~$10,000 so there is still a lot of work to be done. The substrate industry needs to develop and exploit roadmaps to reduce the costs of manufacturing these GaN single crystals. One route might involve the introduction of cost-effective, multiwafer, hydride vapor-phase epitaxy.
Acknowledgment
Planetary Reactor is a registered trademark of Aixtron AG.
For more information, contact Bernd Schulte, executive VP and COO, compound semiconductor technologies, at Aixtron AG, Kackertstr. 15-17, 52072 Aachen, Germany; ph 49/241-8909-123, fax 49/241-8909-452, e-mail [email protected].