Substantial reductions in chip production costs will spur the uptake of LED-based solidstate lighting. One way to do this is to start to manufacture these emitters with multi-wafer 6-inch tools that set a new benchmark for reproducibility, argues Aixtron’s Rainer Beccard.

The LED business is booming. These chips are generating attractive cash flows from backlighting the screens of netbooks, lap tops and TVs and this solidstate device is about to break into lucrative new territory: general illumination. The leading LED manufacturers have had their eyes firmly fixed on this goal for many years and their dream is now turning into reality, thanks to the8 release of the first commercial lighting products.

At Aixtron, which is based in Aachen, Germany, we have a strong track record in supporting the tremendous progress of LED manufacturers. Our effort has focused on continuous improvement in the throughput of MOCVD reactors, echoing the developments of other toolmakers in the silicon industry.

Our first design of MOCVD reactor for growing GaNbased LED epistructures accommodated 2-inch substrates, and over the years we have unveiled reactors that can house more wafers with larger diameters. This effort has culminated in our release of the future-proof Aixtron AIX G5 HT earlier this year. This tool offers simultaneous deposition of GaN and its related alloys on eight 6-inch wafers, the size that many LED chipmakers will look to migrate to over the next few years. In addition, this reactor can be configured for the growth of multiple 8-inch wafers.

The economies of scale realized by changing to a 6-inch process are obvious: better utilization of the MOCVD reactor area; less edge exclusion; more efficient handling; and better precursor utilization in the epitaxial process. However, it is not possible to produce high-quality, 6-inch LED epiwafers by simply taking established processes and applying them to these larger wafers. That’s because such large wafers create their own challenges due to their size, weight, and thickness, and the entire MOCVD environment has to be designed to suit them. In addition, the MOCVD tool must be capable  of high yields and fast cycle times, as otherwise this would negate the productivity advantage gained by the migration to large wafers.

We considered these issues when we defined our requirements for our 6-inch MOCVD tool. We decided that the reactor must be capable of producing epiwafers with uniformity high enough to translate to an overall gain in yield over previous generations of MOCVD tools. For the same reason, we had to build a reactor that set a new benchmark for wafer-to-wafer, run-to-run and tool-to-tool reproducibility.

To maximize throughput, our reactor would have to operate without cleaning and baking between growth runs. In addition, we set out to build a tool that required very little preventative maintenance, generated very few particles, and was highly automated. For example, customers had to have the option of buying a version of this tool with automatic loading and unloading.

Our flagship reactor, the AIX G5 HT, fulfils all these goals by realizing stable, reproducible and uniform growth processes on wafers up to 8-inch in diameter (see Figure 1). While designing this reactor, we paid careful consideration to the two fundamental aspects that determine the capabilities and performance of any MOCVD reactor: the thermal conditions; and the gas flow dynamics and chemical reactions, in both the gas phase and the solid phase.



Figure 1.Aixtron G5 HT Planetary Reactor in 56 x 2-inch configuration (above). Top view of this reactor in 8 x 6-inch configuration (right). The tool can also be configured for 4-inch and 8-inch wafers

 

Minimizing temperature variations

The AIX G5 HT features a novel type of gas injector that introduces perfectly laminar gas flows into the reactor.This condition can be realized at high growth pressures(close to atmospheric pressure) and growth rates ofup to 30 μm/hr. Thanks to this approach, uniform gasphase depletion occurs for all wafer sizes.

One pre-requisite for the growth of high-quality epiwafers is excellent temperature uniformity across the wafer — deviations must be less than 1 °C. This must be realized for both the low temperatures associated with multiquantum well (MQW) growth, and the far higher temperatures employed for growth of the other regions of the LED.

To realize the excellent temperature uniformity that holds the key to uniform film deposition, we employ our proprietary Planetary Reactor design that features on all of our multi-wafer tools. The satellite disks that hold the wafers rotate individually on a rotating planet disk, which is heated by an RF coil. For large wafer sizes such as 6-inch, this principle leads to an inherent advantage for the subsequent backend process. This stems from the high degree of rotational symmetry associated with the temperature distribution on the satellite, and the wafer that it supports.

To improve reactor performance even further, we have optimized the design of the RF coil and the satellite disk. Thanks to this, our tool delivers unprecedented levels of uniformity on 6-inch wafers.



Unfortunately, complete absence of temperature variations on the satellite disk is no guarantee of highly uniform film deposition. That’s because there are differences in the lattice constants and thermal expansion coefficients of the sapphire substrate and the nitride-based LED heterostructure. Strain that results affects all wafers, although the bowing that it causes gets more pronounced as wafer size increases.

Needless to say, bowing is an impediment to uniform LED properties. To prevent this, it is possible to deposit the epiwafers on thick sapphire substrates (above 1 mm), employ in situ curvature measurements to monitor and correct for bow, and last but not least, insert special layer stacks into the LED heterostructure that minimizes bow. Armed with these techniques, it is possible to realize excellent photoluminescence uniformities using the 8 x 6- inch configuration of the G5 reactor.



This high level of uniformity leads to great yield figures. Based on the above uniformity data, an exact calculation of the area yield shows that more than 98 percent of the wafer area is in a 5 nm bin.

A worthwhile analysis of yield must not be restricted to a single wafer – it must consider wafer-to wafer uniformity and reproducibility. To deliver on both these fronts, we have devoted substantial effort to optimizing the design of the reactor and the materials that it is built from. On top of this, we have made further gains by controlling the temperature of each individual wafer.

To realize individual temperature control, the temperature from the top of each satellite is measured with a pyrometric device. The gas flow of the gas-foil rotation drive of each satellite is then adjusted accordingly, bringing the temperature of the wafer back to its desired value.

Keep on running

Reproducibility is a key issue in high volumemanufacturing environments employing many identicalMOCVD tools running standardized growth recipes. Ifhigh yields are to be realized day-in, day-out, then everyreactor must deliver exactly the same performance andresults from one run to the next without any re-calibration.



We have analyzed the root causes of non-stability invarious MOCVD systems and determined that they arepredominantly related to small temperature drifts in thereactor set-up. Consequently, with our G5 reactor wehave strived for a design with inherent temperaturestability. One of the key features of this particular reactoris its novel graphite ceiling plate. In the Planetary Reactordesigns, the ceiling plate defines the upper thermalboundary of the reactor. Even though it is not activelyheated, it does influence the reactor’s thermalmanagement.

The great strength of the new graphite ceiling is that its emissivity is unaffected by the deposition of materials onto its surface. This means that the thermal properties of the reactor are fixed, rather than depending of the number of growth runs already performed. This results in unprecedented reproducibility of all LED properties, from run to run and between different reactors.

Another route to increasing productivity of an MOCVD system is to reduce its cycle times. The G5 reactor excels in this regard. Not only does it enable very high growth rates that cut material deposition times — it also has very short times associated with the non-growth processes that form part of the production run.

These gains stem predominantly from the introduction of the graphite ceiling. As noted before, this ceiling plate does not have to be frequently exchanged to ensure thermal stability, because there is absolutely no thermal drift. What’s more, the process conditions used for the ceiling mean that any deposits create a very solid film. This is stable, does not peel off and never generates particles, so there is no need whatsoever to exchange or clean the ceiling between LED growth runs. Additionally, there is no need for in situ bakes, conditioning runs, or exchange of any reactor parts.

The upshot of all of this is that growth runs can be performed continuously, without interruption. This slashes “downtime” associated with cleaning and maintenance. In more quantitative terms, the throughput of the AIX G5 HT is more than double that of the previous MOCVD tool generation, thanks to the combination of larger capacity, large wafers and shorter cycle time.

Avoiding human contact

The features associated with the G5 will appeal to manyof the bigger LED manufacturers, including those having ahistory in the silicon or display business environment.Many of these firms will lead the transition from smallwafer sizes to 6-inch wafers, and are likely to viewautomation as a pre requisite for unlocking the fullpotential of large substrates.

From a yield point of view, manual wafer handling carries an inherent risk of error. Over time this diminishes yield, with larger wafers leading to bigger losses than smaller ones. Consequently, the advantages associated with automation for manufacturing on 6-inch wafers heavily outweigh any downsides, especially once the cut in the non-productive cycle time of the MOCVD tool is accounted for.

Our incorporation of automation on the G5 tool has been realized without making any compromise to the performance of the MOCVD reactor or its processes. The transfer module, which provides automated loading, is very reliable and simple to use. A robotic system accesses the reactor through a gate valve, picks up a satellite disk together with the wafer, and then replaces it with another satellite disk housing a fresh substrate. It only takes a few minutes to exchange a complete reactor load, a process that is performed after only a short cooling phase (hot load capability). The satellites housing the epiwafers are taken away and unloaded and reloaded while the G5 starts its next MOCVD growth run.

It is possible to operate a single G5 reactor with a transfer system. However, to cut overall capital expenditure and save space, if an LED chipmaker has several of these MOCVD units, they can share transfer systems.

The light ahead

Over the next few years there will be major changes inLED manufacturing. The emergence of solid state lightingwill encourage many leading LED chipmakers to increaseproduction capacity, and prompt heavyweights in otherindustries to enter this sector. This will lead to bigger LEDfabs, which will start to resemble the silicon foundries.

These bigger fabs will focus efforts on rapidly improving throughput and productivity, which will include the introduction of 6-inch LED processes. Sapphire substrates of that size are already available, and are complemented by the latest MOCVD tools, such as our AIX G5 HT. Whether this is configured as an automated tool or as an MOCVD cluster tool, it will easily meet the foreseeable throughput, cost, performance and yield requirements of the coming years.



Figure 2. Photoluminescence map of a typical LED multiquantum well. Standard deviation across the entire 6-inch wafer (no edge-exclusion) is 0.9 nm