The gases and chemicals that are consumed during the manufacturing of LEDs make a considerable contribution to device costs. But savings are possible by switching to a new MOCVD design with a flared chamber profile. This not only trims gas consumption by up to 40 percent but also shortens the growth interruption between layers, leading to improved multiple quantum well quality, say Frank Campanale, Mike Begarney and Tom Ryan from Valence Process Equipment.

As the price of the LED light bulbs fall, sales are rising fast. But solid-state bulbs have still got a long way to go before they become the dominant lighting technology – this will require a substantial fall in the cost of the LED chips. One way to contribute to cost reduction is to make more efficient use of the gases and chemicals consumed in the making of an LED. That’s been the goal of our team at Valence Process Equipment (VPE) of Branchburg, New Jersey.

During the last five years, we have been targeting the development of a reactor design with 50 percent higher gas efficiency than traditional MOCVDs that also reduces the need for cleaning and maintenance. We have made great progress, moving quickly from a small-scale, proof-of-concept prototype, to a number of custom systems, and finally on to the release of the first of a family of production-scale MOCVD reactors.

Although the initial focus of this work has been on gas efficiency, the novel reactor design has additional benefits.  Its small chamber volume allows fast gas switching while its low thermal mass enables fast temperature ramping. In combination, these attributes lead to a very fast multi-quantum well (MQW) growth cycle time. This not only shortens the overall growth time, with the benefit of higher productivity, but also leads to QW structures with improved interface properties and higher internal quantum efficiencies. This means that LED manufacturers investing in our tool could see a reduction in overall manufacturing cost per device, as well as an increase in lumens-per-watt.

Reactor design

Our reactor, which is capable of processing more than 50 2-inch wafers in a single run, is markedly different from the two ‘widely available’ designs on the market. These incumbents use either a radial-horizontal or a vertical gas flow. If the former is employed, wafers are mounted on a rotating device toaddress the radial depletion of reactants. With the latter design, having an injector diameter the same as the wafer carrier, much of the gas is wasted as it flows over the fluid boundary layer above the wafer carrier and out of the exhaust.

In contrast, our design, US patent application 12/248,167, has a relatively small, multi-port injector mounted in a chamber with a curved top wall that flares out to the full diameter of the wafer carrier –  more than 500 mm (see Figure 1). We adopted this design concept based on three observations. Our first of these was that many vertical-flow type reactors require high gas flows to suppress the rotation-induced recirculation zone that tends to form at the outer edge of a rapidly rotating wafer carrier. Much of this gas flows straight to the exhaust.

Figure 1. The chamber has flared profile, which alters the direction of gas flow

Our second observation was that gas arriving at the centre of the wafer carrier must migrate outward over the wafer carrier to reach the exhaust – because it can’t flow through it! So substrates towards the edge of the wafer carrier are shielded by this gas from fresh gas delivered vertically downward from outer areas of the injector. So there is a limited utility in delivering more reactant gases at the outer areas of a full-sized injector.
The third observation was that it is quite difficult to uniformly distribute gas flows from relatively small supply tubes through a very large diameter injector.

But it is of extreme importance to achieve a flow front of highly-uniform velocity as the gases exit the injection surface. If there are velocity variations across the surface of the injector – as often occurs when flows from the tubes locally ‘punch-through’ areas of the injector – recirculation eddies form, compromising the laminar velocity profile of the reactor.  This limits the ability to control the thickness uniformity over the wafer carrier, causes deposition of reaction by-products on the injection surface, and allows reaction by-products to incorporate into the growing layer.

We realized that these challenges could be addressed by using a small-diameter injector in combination with a curved reactor top-wall that eliminated the inefficient volume of a traditional cylindrical chamber.  This design takes advantage of the natural tendency of the gas to expand as it flows down and outward over the wafer carrier. Our approach, then, was to develop a highly-symmetric, laminar reactor flow profile that allows reactants, injected through a smaller diameter injector, to reach the outer areas of the wafer carrier. The curved wall profile forces those gases into close proximity to the wafer carrier in order to squeeze out as much of the reactants as possible before they are ‘lost’ into the exhaust.

The result was a unique flared reactor design.  Gas enters the chamber vertically, then, guided by the curved walls and rotating wafer carrier below it, becomes increasingly diverted in a more horizontal direction as it approaches the wafer carrier. As the gas progresses through the chamber, the curved top wall forces the gas closer to the wafer carrier.  Fluid dynamics modelling results revealed that as the top, curved wall of the chamber approaches the horizontal surface of the wafer carrier, the convergence of the flow path-lines, combined with the increase of the tangential velocity of the wafer carrier with radius, produces differences in the radial and tangential velocity components of the gas when compared to a standard vertical reactor.

These differences indicate that a particular gas molecule will spend a greater amount of time in the reaction zone, allowing a higher proportion of material to be used for epitaxial growth. In addition, the profiled chamber dramatically reduces the proportion of gas flowing straight from the injector to the exhaust without interacting with the wafers.

A key decision for any designer of MOCVD tools is how to introduce the group III alkyl and group V hydride gas streams into the reactor. In our system, these gas mixtures are kept separated until they enter the chamber through a multi-zone, multi-port injector array with a diameter considerably less than that of the wafer carrier.  The gas streams enter separate chambers, and while confined there, are laterally diffused.

Gases then pass through an array of micro-tubes, where the temperature is stabilized, before entering the main volume of the reactor chamber. The multi-zone injection design allows the delivery of reactants to be adjusted proportionally to the surface area of the wafer carrier, offsetting the radial depletion that is characteristic of horizontal-flow reactors. This enables excellent uniformity, all the way to the outer edge of the wafer carrier.

The carrier is heated with a three-zone, radiant heating system that achieves temperatures above 1200 °C. The walls of the chamber are lined with a thin exterior layer, with a fluid gap immediately behind. Through appropriate choice of heat transfer fluid, it is possible to control the temperature of the walls and injector over a wide range. One of the attractive features of our MOCVD reactor is that virtually no reaction products are deposited on the injector – so regularly scheduled cleaning is not needed. This is possible because the wafer carrier is located far enough away from the injector array to prevent the deposition of reaction products on the injector.

The chamber lid is a top-loading, clamshell design, allowing easy access for loading and unloading of the wafer carrier and routine maintenance. Perhaps more importantly, the top-loading design minimises the internal volume of the chamber and maintains the chamber symmetry. This attribute, which is not present in systems that use a gate valve for wafer carrier transfer, helps realise a fully-symmetric flow field and abrupt switching of gases between layers.

The VPE GaN-500 that was released in summer 2013 can accommodate 52 2-inch wafers


The VPE GaN-500 features a novel injector/chamber assembly. One can see the profiled wall and the multi-port injector

Proven capability

Our current version, the GaN-500 MOCVD reactor, released in summer 2013, accommodates 52 2-inch wafers in three concentric rings, and is easily expanded to 59 wafers with no significant design changes. This tool’s high degree of capability is revealed by photoluminescence mapping of a typical LED structure, which we have grown using a gas flow rate of around 120 standard litres per minute – some 40 percent lower than normal flow rates in ‘widely available’ MOCVD reactors of comparable capacity. The chosen structure comprises 4 µm of GaN, a 20-period strain-relieving superlattice and an eight-period multi-quantum well, capped by a standard p-type GaN layer.

This heterostructure, which was formed using a GaN growth rate of 5 µm/hr, has an average peak wavelength of 460 nm (see Figure 2). The full-platter standard deviation (1ơ) is less than 1.8 nm, and the difference between the maximum and minimum values of the average wavelength emitted by the wafers is less than 4 nm. Within wafer uniformity is also excellent: Standard deviations are less than 1 nm, with many wafers showing less than 0.5 nm (see for example, Figure 3, which shows a photoluminescence map of a representative middle-ring wafer with a uniformity, in terms of 1ơ, of 0.417 nm).

Figure 2. A high degree of wafer-to-wafer photoluminescence uniformity is possible with the VPE GaN-500

Figure 3. Uniformity across the wafer, evaluated in terms of the standard deviation, can be less than 0.5 nm. This wafer is numbered 14 in Figure 2

We attributes these excellent values of uniformity to: good macroscopic temperature and gas composition uniformity across the whole wafer carrier; excellent strain compensation at the multi-quantum well growth temperature, which ensures that the wafer lies in good thermal contact with the pocket; and good work practice to ensure particle free placement of the wafer in the pocket. Characterisation of LED epiwafers has not been limited to photoluminescence studies. We have also investigated the thickness uniformity, crystalline quality and electrical properties of other key layers, such as the n-type GaN, p-type AlGaN and p-type GaN layers, wherein material quality is very high, validating the high level of capability associated with the reactor design.

Influencing material quality with reactor design

Although excellent wavelength uniformity is an essential attribute of any production-worthy MOCVD reactor, the factors that contribute to good uniformity don’t necessarily guarantee a brighter LED. There are many factors that influence the quantum efficiency of this device. Basic quality of the epitaxial material, dislocation density and background contamination levels are obvious examples. The quality of the MQW interfaces is also a major contributor. In a typical MQW LED the InGaN active layers must be grown at a precise temperature while the GaN barrier layers are grown at a temperature that may be over 100 °C higher. To obtain atomically sharp interfaces with an absolute minimum of inter-diffusion or contamination between the layers demands gas switching that is rapid and abrupt; fast temperature ramping and stabilization and, when the growth temperature is reached, rock-solid stability. Our reactor, thanks to its novel design, excels in all these areas. The small injector in combination with the curved wall profile reduces the volume of gas within the reaction chamber compared with a conventional, barrel-type chamber with a full-platter injector. Consequently, residual gas is swept out of the system very quickly.

The thermal response is also very rapid, thanks to an over-powered heater and a very light wafer carrier. Thermal reflectors, built into the flow guide surrounding the wafer carrier, maximise the efficiency of the heater while also shielding exhaust gases from being directly exposed to the heating elements. The end result is a system that is capable of very rapid gas switching, very rapid thermal ramping and very rapid temperature stabilisation – leading to sharp and clean interfaces and uniform indium concentration within the QWs. Both these effects can be observed in the detail of high-quality triple-axis X-ray diffraction data (see Figure 4).

Figure 4. Triple axis X-ray diffraction data and simulation from an eight period MQW structure with 20-period strain-relieving superlattice. Note the narrow and symmetrically shaped superlattice peaks and how well they match the simulated data

The intensity and widths of the higher-order satellite peaks are influenced by the ‘sharpness’ of the QW structure. Fast thermal cycling and stabilization also has a throughput benefit. For an eight-period MQW, total growth time can be reduced by as much as 40 minutes.

But the biggest benefit is the effect on photoluminescence or electroluminescence intensity. After optimisation of the MQW growth temperature control and cycle time, we have seen up to 20 percent higher intensities from identical structures grown under otherwise identical conditions (see Figure 5).

Figure 5. The left and right plots show photoluminescence intensity before and after improved temperature ramping and stability during quantum well growth. These improvements can lead to better interface quality

This gain in emitted intensity, along with a faster growth time that contributes to a higher throughput and up to a 40 percent reduction in gas and alkyl consumption, demonstrates that  our reactor may be ideal for LED manufacturing. In short, we believe that chipmakers armed with this tool can build LEDs cheaper and brighter.