GaN: better with defects?
There is no doubt that GaN-based light emitting devices are a great success. Blue, green and white LEDs made with this material now backlight the keypads and screens of billions of mobile phones, and provide light sources in bill-board displays, torches and traffic lights. Meanwhile, GaN ultra-violet lasers are powering Blu-ray players and some of the latest games consoles.
However, if sales of these devices are to continue to grow, then their manufacturers will have to target new markets. These include car headlamps, projectors, and the most lucrative sector of them all, general illumination. But success in these markets will only come if the cost-per-lumen of the devices falls substantially.
Researchers throughout the world are embarking on this quest by increasing the efficiency of GaN devices, particularly at the high drive currents needed to generate the required lighting levels demanded by the emerging applications. Many believe that brighter devices can be realized by reducing the defect density in these structures. Although nitrides are incredibly resilient to high defect densities - which would kill luminescence in other compound semiconductor light emitters - several theoretical and experimental studies are claiming that defect-free nitrides would be far more efficient.
I decided to take a different path to cutting the cost-per-lumen of nitride devices, which focused on reducing expenditure on the growth tools used for manufacture. However, through these efforts at developing a lower cost deposition tool, I have turned one of the core pieces of perceived wisdom about GaN on it head - I’ve discovered that increasing in the defect density in this material can actual enhance its photoluminescence intensity in the case of GaN nanoparticles.
Back in 2002 - under the guidance of Umesh Mishra and Steven DenBaars at the University of California, Santa Barbara (UCSB) - I started looking at methods to grow nitride materials that could offer a more affordable alternative to the widely used MOCVD approach that employs multi-wafer reactors with million dollar price tags. My work focused on a growth method known as ammonolysis, which is also described as ammonothermal synthesis. This process aimed to create highly crystalline, chemically pure material by heating a gallium containing starting material in an ammonia atmosphere. Wet-chemical approaches to GaN synthesis were avoided, because they can lead to the addition of impurities such as carbon, oxygen or hydrogen, which diminish the intensity of the band edge photoluminescence in GaN. These impurities would therefore hamper the efficiency of any form of optoelectronic devices.
One of the strengths of ammonolysis is its versatility for preparing GaN nanostructures - it is possible to use a wide range of gallium-containing compounds as a starting material. However, I avoided gallium alkoxides and other organo-gallium compounds because of the threat of carbon contamination, and selected metallic gallium and gallium oxide-based precursors instead.
The equipment that I used for producing GaN by ammonolysis is incredibly simple. A single-zone tube furnace that features a quartz tube that can be connected to tubing at either end fulfils the requirement for a continuous gas flow (see figure 1). Gas supplies can feed ammonia and nitrogen into this reactor, and precursor materials placed in an alumina crucible provide the source for group III metals, such as gallium and indium.
As the minimum temperature for the conversion of gallium oxides to GaN under ammonia flow is at least 800 ºC, I decided to prepare a series of GaN powders using various different reaction temperatures within the 900-1100 ºC range. These nanostructured samples,which were produced with a range of precursors, are composed of the hexagonal form of GaN, according to X-ray diffraction measurements.
The photoluminescence results produced by these samples gave me a tremendous shock. They directly opposed the widely held view in the nitride community that better quality material produces photoluminescence. In my case I found that the very opposite was true - better quality material produced far weaker emission than its poorer quality counterpart.
The more defective GaN powders were produced at 1100 ºC, a growth temperature that drives the decomposition of GaN into metallic gallium and nitrogen gas. These samples were grey in color and produced far stronger photoluminescence than those grown at lower temperatures, which were powders that displayed a light yellow color. This surprised me because I expected the lightly colored materials to have similar properties to the transparent/yellow, high-quality, epitaxially grown films.
My next goal was to try and make sense of these seemingly contradictory results. Any additional phases of gallium oxide, or any other possible contamination source for that matter, could be eliminated in all the samples by X-ray diffraction results. But what about forms of contamination that don’t have a crystalline phase? They would not be exposed by diffraction, because this technique only probes the long-range order in materials. Hydrogen, for example, which has been claimed to quench band edge related photoluminescence, would be difficult to pick up by X-ray diffraction and most other analysis methods.
Fortunately there is a technique that doesn’t just detect hydrogen, but also determines how it is bound to its nearest neighboring atoms – nuclear magnetic resonance (NMR). I then started an ongoing study of my samples using this technique in collaboration with Jerry Hu, the Spectroscopy Facility manager and NMR expert at UCSB’s Materials Research Laboratory. This was the first ever study of hydrogen contamination in GaN by 1H NMR.
During these studies I discovered that NMR is, in general, incredibly well suited to studying atomic level binding in GaN powders. In addition to revealing hydrogen impurities, it is possible to investigate the type of coordination of gallium within the material, thanks to the 71Ga isotope being NMR active.
Hu and I recorded 71Ga NMR spectra of different GaN samples, and discovered striking differences between different materials on the atomic level. To our big surprise, we found a strong correlation between the 71Ga NMR spectral features and photoluminescence (see figures 2 and 3 for details). These results indicated that the observed intense photoluminescence could be caused by nitrogen deficiencies. What’s more, they also contradicted the assumption that stoichiometically balanced GaN is needed to produce the strongest luminescence.
The next step, which we reported in 2006, involved the NMR study of isotopically labelled GaN nanostructures that were produced using similar conditions. By labelling ammonia with the 15N isotope, we could identify gallium, nitrogen and hydrogen isotopes in our samples. What we discovered was a striking similarity between the NMR data for the 71Ga and 15N isotopes in the samples grown at high temperatures (see figure 3). We were also able to assign the broad feature in the NMR spectra to crystal structure defects at the nearest-neighbor atomic level. This implies that it is very unlikely that chemical shift distributions are due to electronic effects, a cause that has been suggested by other researchers in this community. We continue to believe that these shifts are instead caused by nitrogen-deficiencies, which is a conclusion that we had drawn in 2004.
The addition of 15N NMR data has also enabled us to determine gallium-to-nitrogen ratios, while the 1H NMR measurements have offered us an insight into hydrogen contamination within the material. As expected, higher growth temperatures lead to an increase in nitrogen-deficiency and a reduction in hydrogen contamination.
A closer investigation of the X-ray diffraction results reveals that the samples that produce more intense photoluminescence contain several crystalline phases. Each of these is hexagonal GaN, but these phases contain slight variations in lattice spacing that can be attributed to the increase in nitrogen vacancies.
After completing this work, I went in search of new challenges in Dan Morse’s group at UCSB’s Institute for Collaborative Biotechnologies. But it wasn’t long before I found myself hankering for an even better understanding of the interplay between the structure and the properties of GaN, and wondering about the answers to a series of nagging questions: How does the chosen precursor influence photoluminescence? What extent of nitrogen-deficiency is needed to increase the emission intensity? Is the band edge related luminescence in GaN powders directly or only indirectly (no oxygen contamination, no structural protons) linked to the observation of nitrogen-deficient phases in the 71Ga NMR spectrum?
Today I’m devoting some of my time to answering these questions in a team that now includes Brad Chmelka’s group in the department of chemical engineering. Our understanding of the relationship between band edge related photoluminescence and the levels of nitrogen deficiency in GaN nanostructures is still based on experimental evidence, but we now are trying to tie these results to underlying physical principles. For example, we have confirmed our hypothesis that nitrogen-deficiencies are responsible for the increase in photoluminescence by converting non-luminescing GaN - which was prepared from gallium oxide under nitrogen gas at 1100 °C - into luminescing GaN. Again, the results of NMR spectroscopy revealed changes at the atomic level.
Since my efforts of ammonolysis began, several groups have independently reported increased photoluminescence after re-annealing their GaN LED epistructures. However, none of these groups appears to have established a connection between potential decomposition of GaN and an increase in emission, which stems from the introduction of nitrogen-deficiencies. Although it would be fascinating to compare the chemical composition of the GaN layers in these devices before and after annealing, one of the drawbacks of NMR is that it is that it cannot readily be used to investigate assembled device structures, due to the relatively large dimensions of an LED and the small amount of GaN that it contains. However, it might be possible to simulate the heat treatment techniques used for GaN LEDs, and then study the chemical composition of the active layers in these structures.
It should be possible to carry out this analysis using several of the techniques that we employ to determine the characteristics of our bulk samples. These studies would not only be very interesting from a scientific perspective – they could also help to spur the performance of LEDs, and enable them to enjoy success in emerging markets such as solid-state lighting.
