«1. INTRODUCTION AND IMPACT OF WHITE LIGHT EMITTING DIODES The basic structure of an efficient double-heterostructure (DH) light emitting diode (LED) ...»
Background Story of the Invention of
Efficient Blue InGaN Light Emitting Diodes
Nobel Lecture, December 8, 2014
by Shuji Nakamura
University of California, Santa Barbara, CA, USA.
1. INTRODUCTION AND IMPACT OF WHITE LIGHT EMITTING DIODES
The basic structure of an efficient double-heterostructure (DH) light emitting
diode (LED) is summarized in Figure 1. This optoelectronic device is composed
of semiconductor materials and is fabricated by sandwiching an active, emit- ting layer between an n-type and p-type layer. The n-type semiconductor layer has an abundance of high-energy electrons, whereas the p-type semiconductor has an abundance of available, empty sites in which the electron may reside at a lower energy level. These sites are also referred to as holes, are positively charged, and are mobile. The energy difference between the high and low energy electron state is referred to as the bandgap of the material. For DH LEDs, the bandgap of the active layer is smaller than that of the n-type and p-type layers.
When forward biasing a DH LED using a battery (or any other direct cur- rent source), electrons and holes are injected into the active layer from n-type and p-type layer, respectively. The electrons and holes recombine radiatively in the active layer, thereby emitting photons. This act is very efficient for DH LEDs as the electrons and holes are confined to the active layer due to the smaller bandgap of the active layer with respect to the n-type and p-type cladding layers (see also Figure 7). The resulting photon has an energy approximately equal to the bandgap of the active layer material. Modifying the bandgap of the active layer creates photons of different energies.
69 70 The Nobel Prizes FIGURE 1. Schematic depiction of a double heterostructure (DH) light emitting diode (LED) in operation while being powered by a 2.8 V battery. Within the active, emitting layer, electrons and holes recombine and emit light equal to the bandgap of said layer.
High-energy electrons are sourced from the negative terminal of the battery and return to the positive terminal after losing their energy to a photon in the active layer.
In the 1980s, all known material systems possessing the necessary material properties for blue light emission had shortcomings negating the possibility of creating an efficient blue LED. Gallium nitride (GaN) was one possible candi- date, though, at the time, no p-type or active layer could be created. These chal- lenges were ultimately overcome, leading to the first efficient blue LED using GaN in 1993 by Nakamura et al. . Figure 2 shows a close-up image of a bare and packaged blue GaN LED.
Using blue LEDs, highly efficient white light sources become possible. This can be achieved by converting part of the blue light emitted from an LED to yellow using a phosphor . To the human eye, the combination of blue and yellow light is perceived as white. A white LED can be created by embedding phosphors in a plastic cap which surrounds a blue LED (see Figure 3). Higher quality white light can also be created by mixing blue light with other colors as well, including red and green .
With the availability of white LEDs, a variety of applications can be significantly improved, if not enabled all together. But arguably, the most important impact of the white LED is its ability to generate white light at an efficiency that was previously impossible. The efficacy, a measure of perceived light power relative to the provided electrical power, of white light has improved over the Story of the Invention of Efficient Blue InGaN Light Emitting Diodes 71
centuries, starting with oil lamps (0.1 lm/W) in the 15,000s B.C., incandescent bulbs (16 lm/W) in the 19th century, fluorescent lamps (70 lm/W) in the 20th century, and LEDs (300 lm/W) in the 21st century (see also Figure 15).
With this significant improvement, substantial energy savings are now possible. It is currently estimated that in 2030 approximately 261 TWh of electrical energy will be saved due to widespread use of white LEDs . This corresponds to an electricity savings of approximately 40% in 2030. Furthermore, this reduction in energy usage eliminates the need for at least 30 1-GW power plants by 2030 and avoids generating 185 million tons of CO2.
2. MATERIAL OF CHOICE: ZnSe VS. GaN In the 1980s, there were two materials considered as possible candidates for efficient blue LEDs: zinc selenide (ZnSe) and GaN .
ZnSe could be grown on single crystal gallium arsenide (GaAs) substrates, yielding high structural quality material given the very small lattice mismatch of 0.3% between ZnSe and GaAs. For GaN, on the other hand, no lattice-matched substrate was available and researchers were forced to grow it on sapphire. The large lattice mismatch (~ 16%) resulted in heavily defected material with a high density of dislocations.
When I joined the field in 1989, ZnSe was grown on GaAs with dislocation densities less than 103 cm–2. It was very popular among scientists, given the high crystal quality and the prevailing notion that a dislocation density below 103 cm–2 is needed to achieve optically functional LEDs with a high efficiency and a long lifetime . Most researchers worked in this field. GaN, however, was grown on sapphire, yielding dislocation densities on the order of 109 cm–2.
Unsurprisingly, few researchers were working in this field except, most notably, fellow Nobel Laureates Professor Isamu Akasaki and his graduate student at the time, Hiroshi Amano.
A striking example to highlight the popularity of ZnSe, as compared to GaN, is provided by looking at the attendance of researchers at the most popular conference for applied physics in Japan. At the Japan Society of Applied Physics (JSAP) conference in 1992, there were approximately 500 individuals attending the ZnSe sessions, whereas for GaN, there were around 5, including the chair Professor Isamu Akasaki, speaker Hiroshi Amano and myself, as a member of the audience. Not only was ZnSe more popular at the time, GaN was actively discouraged with researchers stating “GaN has no future” and “GaN people have to move to ZnSe material.”
3. DEVELOPMENT OF GaN
My entry into the field started in April of 1988, when I went to the University of Florida as a visiting researcher. The main purpose of my visit was to learn how to use a MOCVD (Metal Organic Chemical Vapor Deposition) system to growth GaAs crystals on a silicon substrate, as I had no experience in how to use a MOCVD. During my stay there, I worked together with graduate students and they all asked me if I had a Ph.D. I said no. At the time, I only had a Master’s.
Next, they asked me if I had published any scientific papers. Again, I said no, I had never published a single paper. Consequently, they treated me as a technician. In the U.S., this meant one has to help the researcher and one’s name would Story of the Invention of Efficient Blue InGaN Light Emitting Diodes 73 not appear on papers or patents. Gradually, I became very frustrated with this arrangement.
One year later, in March of 1989, I came back to Japan. It was my dream to get a Ph.D. degree. In Japan, at the time, it was possible to be awarded a Ph.D.
if one published five scientific papers. This type of degree was called a paper degree and one did not need to go to the university to get the degree. It was therefore my ultimate dream to publish at least five papers and get a Ph.D.
With this in mind, I noted that the ZnSe field was publishing lots of papers.
As I had never published a paper, I had no confidence in publishing a paper. In the GaN field, only very few papers had been published, mainly from Professor Isamu Akasaki and Hiroshi Amano. I was therefore confident that I could publish lots of papers, though had no confidence that I could actually invent the blue LED. My only objective was to get a Ph.D. That’s it.
So, after returning to Japan in March of 1989, I wanted to grow GaN using a MOCVD reactor. I purchased a commercially available MOCVD reactor for 2 million U.S. dollars. But this MOCVD reactor was designed for growth of GaAs. At the time, Professor Akasaki and his student Amano had developed a novel, research-scale MOCVD reactor for growth of GaN . Their design required exceptionally high carrier gas velocities (around 4.25 m/s) yielding GaN, though the high carrier velocities presented challenges pertaining to uniformity, scalability and reproducibility. Furthermore, their reactor design could only be used for small area growths, thereby lacking the necessary properties for commercialization. Since I was working for a company, I had to find a way to grow high quality GaN on large area, 2-inch diameter sapphire substrates.
Another challenge related to growing high quality GaN was the use of high concentrations of aluminum in the MOCVD reactor. While the development of the aluminum nitride (AlN) buffer layer by Akasaki and Amano was a major breakthrough providing high quality GaN film growth with a mirror-like surface morphology , the use of aluminum caused significant problems to the MOCVD reactor resulting in poor reproducibility in subsequent GaN growths.
Eliminating the use of high concentrations of aluminum during growth was strongly desired.
After my purchase of a MOCVD reactor, I attempted a significant number of growths over the course of a few months, but consistently failed. Either no growth of GaN occurred or the grown layer was black. GaN should be transparent. I realized this was a big problem, especially considering the substantial investment in the tool. That is when I decided I had to modify the reactor.
For the next 1.5 years, I modified the reactor design. In the morning, I would go to work and modify the reactor. In the afternoon, I would perform a couple 74 The Nobel Prizes of growths and analyze the results. I would repeat this pattern for 1.5 years until I invented a novel MOCVD reactor design with a low carrier gas flow which I called a two-flow MOCVD (Figure 4 a) . Using this reactor, I was able to get very uniform and high quality 2-inch GaN growth. The main breakthrough of this reactor was the introduction of a subflow (Figure 4 b) which gently pushed the carrier gases down to the substrate, thereby also improving the thermal boundary layer.
This was the most important breakthrough in my life and was instrumental toward all future breakthroughs in GaN research. One significant advancement (a) (b) FIGURE 4. (a) Schematic of a two-flow MOCVD for GaN growth and (b) schematic of the effect of the newly introduced subflow on the carrier gases.  (Reprinted with permission. Copyright 1991, AIP Publishing LLC.) Story of the Invention of Efficient Blue InGaN Light Emitting Diodes 75 this tool immediately enabled was the development of a GaN buffer layer which was superior to the AlN buffer layer, in part due to the elimination of aluminum from the growth system. With the invention of the two-flow MOCVD and the GaN buffer layer, it was possible to achieve the highest quality GaN material in the world. One measure for crystal quality is the value of the electron mobility in a crystal. Fewer defects result in fewer scattering events, which enhances overall mobility of the electrons. Mobilities for GaN grown directly on sapphire (no buffer layer) by Akasaki and Amano resulted in values around 50 cm2/Vs , whereas use of the two-flow MOCVD yielded 200 cm2/Vs . Use of an AlN buffer layer improved the mobility to values as high as 450 cm2/Vs for Akasaki and Amano . Use of a GaN buffer layer and the two-flow MOCVD values as high as 600 cm2/Vs were measured at room temperature (see Figure 5) .
This was a clear sign that the two-flow MOCVD was producing GaN material of higher quality on larger area substrates, a key step towards commercialization of GaN based devices.
The next significant development in creating an efficient blue LED occurred in 1992 when I was able to clarify why p-type GaN had remained so elusive for 20 years. While Akasaki and Amano achieved a major breakthrough in 1989 by demonstrating local p-type GaN after treating magnesium doped GaN (GaN:Mg) with low-energy electron beam irradiation (LEEBI) , its origin
was not understood for another three years. In 1992, I clarified that hydrogen was the source of passivating p-type GaN . A few years later, theoretical computations by Jörg Neugebauer and Chris Van de Walle confirmed hydrogen passivation in Mg-doped GaN .
For MOCVD growth of GaN, ammonia (NH3) is used as the nitrogen source. Ammonia dissociates during growth and atomic hydrogen is introduced into the GaN crystal. If Mg is present in the crystal, the hydrogen atom forms a magnesium hydrogen complex (Mg-H), thereby preventing Mg from acting as an acceptor . Thermal annealing of the GaN:Mg sample in a hydrogen-free environment above approximately 400 °C permits hydrogen to diffuse out of the crystal, thereby breaking up the Mg-H complex . As thermal annealing can be performed quickly and simultaneously on multiple substrates of any size in parallel (an act not achievable using LEEBI), it has become the industrial standard process for p-type activation of GaN. The formation of local p-type GaN using LEEBI treatments can be explained by local heating of the GaN:Mg by the electron beam, causing the hydrogen to locally diffuse out of the crystal and permitting the affected Mg atoms to act as acceptors yielding p-type GaN.