Researchers seek material solutions to GaN deficiencies at ISBLLED 2006

At many of today s conferences you will often see a presentation by a major LED manufacturer outlining a roadmap that begins with the indicator lights of the 1970s and points towards the future growth of solid-state lighting. This map gives the impression that performance improvements are inevitable and it is easy to avoid thinking about the technological breakthroughs that are required to maintain progress.

However, the message was a little different at the recent International Conference on Blue Lasers and LEDs (ISBLLED). At the Montpellier meeting Werner Goetz, director of epitaxial technology at Philips Lumileds, didn t speak in vague terms about the need to improve the output of phosphor-converted blue and green LEDs to 150 lm/W to win business in general lighting, LCD backlighting, automotive and projection applications - he actually detailed the improvements in various LED characteristics that are needed to hit this target.

In his opinion, to produce 150 lm/W green LEDs for red-green-blue light sources will require an increase in extraction efficiency from 65 to 85% and a hike in internal quantum efficiency (IQE) from 20 to 40%. These gains will have to be delivered in high-power devices operating at typically 350 mA. For phosphor-converted blue LEDs similar extraction efficiency improvements are required, but the IQE must be increased from 50% to almost 100% - a value that is already produced in GaAs-based red and infra-red LEDs. Goetz says that there are already laboratory results at Philips Lumileds and other claims within the industry of 85% extraction efficiency in GaN-based LEDs. "We now have to focus on increasing the IQE through improvements in material quality."

Goetz believes that all of the material quality improvements must be compatible with uniform, reproducible and high-throughput manufacturing, and he says that today s issues can be sub-divided into problems associated with either the n-type, active or p-type regions of GaN LEDs. The low p-type doping efficiency, low solubility of the magnesium dopant and high levels of impurities such as oxygen and carbon are all associated with the GaN/AlGaN p-type region that provides hole injection and electron confinement. Meanwhile, the performance of the active region is hindered by low efficiencies at higher currents, difficulties associated with growing high-quality InGaN material at longer wavelengths and polarization effects. The latter could be improved through the use of non-polar substrates. Goetz says that the typically silicon-doped n-type region suffers from tensile strain and growth-related issues that are linked to the difficulties associated with deposition on a sapphire substrate.

According to Goetz, there are also several reasons why it is particularly difficult to grow highly efficient green LEDs. Firstly, the additional indium content increases the piezoelectric strain between the InGaN and GaN layers, which leads to increased carrier leakage and a reduced IQE due to the quantum-confined Stark effect (see figure 1). Higher-indium-content layers also require growth at lower temperatures, but under these conditions adatoms are less mobile, the impurity incorporation and the density of "V"-shaped defects increases and it is more difficult to crack ammonia - the standard nitrogen source.

Goetz notes that another problem with today s devices is reduced external quantum efficiency at high currents (see figure 2). According to him, all commercial InGaN-based LEDs, regardless of color, have peak efficiency at a current density of 1-10 A/cm2. Goetz said that improvements can be made through device design to reduce current density, but also added that the origin of the IQE reduction is not fully understood.

Many of the challenges outlined by Goetz are already being addressed by several research groups who presented at the conference. Russell Dupuis from Georgia Institute of Technology described an increase in green LED output produced by switching from a p-doped GaN layer to a p-doped InGaN layer.

Dupuis explained that the conventional fabrication of green LEDs structures involves high-temperature growth of a p-doped GaN layer, which degrades the active region by increasing indium compositional fluctuations in the InGaN quantum well. His research group has tried to reduce this effect with lower temperature growth of a magnesium-doped In0.04Ga0.96N layer that increases hole concentration and conductivity.

The team compared MOCVD-grown LEDs made with a magnesium-doped GaN layer deposited at 930 ºC and a magnesium-doped In0.04Ga0.96N layer grown at 840 °C. In both cases the quantum wells were grown at 740 °C, and an inductively coupled plasma etch with a silicon-dioxide mask defined 230 × 230 µm mesa devices that were given Ti/Al/Ti/Au and Ni/Au n-type and p-type ohmic contacts, respectively.

At a drive current of 100 mA the device with the InGaN p-doped layer had an electroluminescence peak almost three times higher than the standard structure, and at lower currents this difference was even more pronounced. The new structure s only downside is a small increase in the forward voltage from 3.1 to 3.2 V, which Dupuis attributes to the abrupt heterojunction between the p-type InGaN layer and the upper GaN quantum-well barrier. However, he thinks that the magnesium-doped InGaN layer could improve current spreading and reduce diode series resistance, and that the forward-voltage value is still less than most of the latest devices.

One of Goetz s other suggestions to improve GaN LED output - the use of "non polar" substrates that have a different crystalline orientation - was a hot topic among the delegates, including GaN LED and laser pioneer Shuji Nakamura. Unfortunately, his first results using these substrates showed that the device output powers were significantly lower than those of standard LEDs. The best blue devices made by the University of California, Santa Barbara professor s team had an output of 1 mW at 20 mA drive current, while the best green LEDs produced 0.26 mW at 250 mA.

Nakamura believes that these low output powers are due to either stacking faults or point defects. Although the output powers may be lower than expected, the novel devices are showing some of the other benefits that Nakamura had predicted. For example, the emission wavelength of the non-polar LEDs does not change with different drive current, which means that their color is extremely stable.

These devices also have a lower series resistance and the carrier concentration in the p-doped layer is an order of magnitude higher than for standard devices. Their polarized emission makes them suitable for direct backlighting of LCD displays, which normally require a polarizing light filter to operate with conventional LED backlights. "The only problem is output power," concluded Nakamura.

The materials-related problems in the n-type region of the LED that Goetz described can be addressed by switching from sapphire to GaN substrates. It is an approach that has been pursued by Sumitomo Electric Industries Katsushi Akita, who explained: "Our motivation is to produce GaN with lower threading dislocation densities that could lead to high extraction efficiencies at high current densities."

Sumitomo s engineers compared the performance of MOCVD-grown LEDs with identical active layers that are built on either c-plane sapphire or HVPE-grown GaN that had a threading dislocation density of less than 1 × 106 cm-2. They grew two types of LED on each substrate: one featured a 3 nm thick quantum well and the other a 5 nm thick quantum well. The results of external quantum efficiency (EQE) measurements performed with a pulsed injection current (10 kHz duty cycle, duty = 5%) are shown in the table.

Akita believes that the decrease in EQE at higher currents in both of the 3 nm quantum-well devices is due to carrier localization and that the devices fabricated on GaN contain non-radiative centers (NRCs) that are unrelated to threading dislocations. The density of these NRCs is lower in the LEDs with 5 nm thick quantum wells because these devices have a lower indium concentration (10% rather than 14%), said Akita, before backing up this claim with cathodoluminescence images of the structures that showed fewer defects.

All the chips were then mounted p-side down to improve thermal management and increase light extraction, but this only benefited the LED with 5 nm thick quantum wells that was grown on GaN. For this device a 20 mA drive current produced 17 mW at 30% EQE, while a 200 mA current produced 148 mW at 26% EQE.

Other presentations focused on the use of low-polarity GaN epitaxial layer overgrowth to eliminate stacking faults and the benefits of selective-area growth by HVPE. The need to generate substantial improvements in external quantum efficiency is critical to the future success of solid-state lighting, and the discussions at ISBLLED should go some way to ensuring that LED manufacturers device roadmaps can be trusted.