The newest Nichia lasers, which have much better reliability than previously, use the ELO technology. After growing 2 µm of GaN on sapphire using conventional growth techniques, the wafer is removed from the reactor. A layer of SiO2 is deposited on the substrate, and then lithographically patterned into 2 µm wide stripes. The substrates are then put back into the reactor. 10 µm more GaN is then grown, using conditions resulting in a rectangular habit. The GaN grows over the SiO2. and the film eventually coalesces. The GaN over the SiO2 has a much reduced defect density, since it is difficult for the dislocations originating at the sapppire/ GaN interface to propagate to the overgrown region. In the session on epitaxial overgrowth, Tsvetanka Zheleva reported on the ELO work being done at NC State. TEM results indicating a reduction of the dislocation density in overgrown regions. Nakamura showed that the etch-pit density was essentially zero in the overgrown regions. Etch pits may only reveal one type of defect, but if they are a defect associated with a failure mechanism, then this is an important result. I heard talk about nanopipes being important in this respect. Interestingly, it is claimed that the new lasers do not differ significantly from the old lasers in any respect other than their lifetime.
Prof. Kasumasa Hiramatsu of Mie University gave an interesting overview of the ELO field, and Akitaka Kimura of NEC gave a talk on ELO whose abstract laterally overgrew its space. Results on both MOVPE and HVPE selected area growth and lateral epitaxial overgrowth were presented. It appears that the shape of the overgrowth depends on the growth conditions; you can get flat-topped pyramids, sharp pyramids and rectangular overgrowths. The masking material also affected the shape of the overgrowth. The coalescence region is a possible defect site, seen also in the Nichia lasers.
First, an explanation of that mouthful. The Stark effect is the shift of atomic absorption and emission lines of the hydrogen atom in an electric field. In semiconductors, the same effect is seen in excitons. (The Franz-Keldysh effect is the change of band-gap absorption in an electric field) The Stark effect is larger when the exciton is confined in a quantum well, because the binding energy is larger. The quantum-confined Stark effect (QCSE) was studied extensively in the 80's, in particular by a group at Bell Labs who used the effect to make optical modulators. An electric field has two effects on an exciton in a quatum well, it reduces the oscillator strength, and shifts the energy. Since nitride quantum wellls are typically normal to a piezoelectric axis, any strain will induce an electric field, thus the piezoelectric quantum-confined Stark effect. The larger the strain, the larger the field, the smaller the oscillator strength. Thus, pseudomorphic InGaN quantum wells are expected to have reduced oscillator strength, especially if the indium fraction is large, or the wells are wide. The reduced oscillator strength mean weaker radiative recombination, which means higher laser thresholds. This effect can be neutralized by the introduction of carriers, by optical or electrical pumping, for instance.
In a session titled "Phase Separation in InGaN" Shigefusa Chichibu of Science University of Tokyo and Andreas Hangleiter of University of Stuttgart both emphasized the importance of the piezoelectric effect in optical characterization of InGaN quantum wells. Apparently, both In segregation and piezoelectric effects must be considered together to explain phenomena such as the blue shift with increasing driving current and the time-resolved photoluminescence behavior.
In a related area, it appears that the experimentalists have finally caught up with the theorists doing calculations on surface structures. David Greve of Carnegie-Mellon showed beautiful scanning tunneling micrographs of the low temperature Ga-induced reconstructions of the polar faces of GaN. The Ga face, typically grown by MOVPE, has a 2x2 reconstruction with added Ga, while it appears that the N face has a 3x3 reconstruction, along with a series of more complex reconstructions with added Ga. At a poster, Ruediger Held of University of Minnesota showed RHEED patterns from homoepitaxial GaN which confirmed this picture.
The theorists, who no longer worry that they're calculating the wrong face, have been busy calculating surface and interface structures. Rosa DiFelice of Xerox showed some very interesting results on the AlN/sapphire interface, showing that both polarities could be formed, depending on the nitrogen chemical potential. Zusanna Liliental-Weber showed an example of a GaN inversion domain nucleating on a piece of AlN, which seemed to confirm this picture. Jørg Neugebauer and Krzysztof Rapcewicz gave talks elucidating the structure of the polar surfaces. They agree that both (0001) and (000-1) surfaces are Ga terminated (or capped). (See the terminology recomendations page if you don't remember which surface is which.)
Dr. Nakamura described an interesting self-pulsation phenomenon at 3.5GHz observed in some devices at higher pump currents. This indicates two things- that the lasers can be fast, but also that there is probably a saturable absorber in the laser cavity. The obvious candidates are (you guessed it) the piezoelectric quantum-confined Stark effect or composition fluctuations.
For applications, the noise in lasers is important. Nakamura reported that noise measurements on his lasers indicate a RIN of -140dB, good enough for DVD applications. (I'd appreciate being enlightened on the meaning of RIN).
Lisa Sugiura presented the Toshiba group's laser. They report a threshold current of 530mA, and a density of 10.6 kA/cm2. They achieve p-type doping with no post-growth annealing using Mg doping with an N2 carrier gas. Sugiura reported a new structure, the "Inner Stripe Laser Diode" in which the current stripe is defined by a blocking layer underneath the p-contacts. The Toshiba group suggests that the relationship between device degradation and dislocations has two components, temperature rise and nano-pipes. (See Lisa Sugiura's comments.)
Kathy Doverspike presented the Cree laser. They grow GaN on Si-face 6H-SiC, and are moving to 2" substrates "no different from what we sell." They are now using a vertical contact configuration made possible by Cree's conducting buffer layer technology, but they have also made lasers using the shorting ring technology. Their lowest threshold to date is 225 mA, 11kA/cm2, and at present they claim to be limited by a problem with p-AlGaN layers that they know how to solve. Coating the cleaved facets (done at Brown University) reduces the threshold current. The CW lifetime is less than a minute.
Akito Kuramata presented the Fujitsu group's laser. He emphasized the use of a low pressure (100 torr) VPE process. Their conducting buffer layer is about 9%AlGaN, (Cree won't say what theirs is) and they have obtained 12 kA/cm2 thresholds at 15V applied. The Fujitsu group seems to have very high hole concetrations (1x1018cm-3), which go down to 1017 for 15% AlGaN.
Three groups reported lasers in the late news session. Kiehl Sink reported a new laser from UC Santa Barbara. This is a cleaved-facet laser made by growing GaN on a-plane sapphire, in contrast to their first laser (reported in MIJ-NSR) which was grown on c-plane sapphire. Althought the a-plane sapphire cleaves nicely, the cleave was ugly through the GaN layer. Still, the threshold current density was 15-90 kA/cm2, about 640-2200 mA, 50-90V. Ross Bringans reported on the Xerox laser. The most impressive thing was how nice the dry etched facets and waveguides looked. The AFM measured 40-60Å roughness on the sides, and 5-6Å on the bottoms of etched areas. Their threshold was 25 kA/cm2. The Sony laser was presented by Hiroji Kawai. The interesting thing about their laser is the use of 4% InGaN barriers in their MQW and waveguide layers. Their threshold voltage is only 18V, with a 12kA/cm2 current density. They found lower thresholds with a smaller number of quantum wells, and use 5 wells in their reported structure.
The biggest divergence in the structures of the lasers reported is whether or not a high Al fraction AlGaN cap layer is required immediately after the GaN/InGaN multi-quantum well layers. This was introduced by Nakamura, who claimed that it was necesary to keep the indium in place during subsequent growth of the p-type AlGaN cladding layer. The Fujitsu group agrees that it is essential, but says it functions as an electron blocking layer. The Sony group needs the cap layer, but the Xerox and UCSB groups have no cap layer. It would make a nice story if the cap layer was needed to tune the piezoelectric quantum-confined Stark effect.
Eric Hellman
Based on estimation of dislocation motion, remarkable reliability and longevity of GaN-based light emitting devices despite having an extremely high dislocation density is due to extremely slow dislocation velocity. However, dislocation velocity of GaN-based materials is greatly enhanced by "temperature rise" and in that case dislocation multiplication might occur and influence device lifetime. High injected current and voltage are still required for lasers, which leads to heat generation. So it is very important to restrain the "temperature rise" by reducing device resistivity or improving heat radiation. "Nanometer pipes" are easily formed at threading screw dislocations, and they might also lead device degradation. So reducing nanopipes is also important.
In my talk, I mentioned that there are many factors that influence the device lifetime, but I suggested two factors related to dislocation. (This is very important! As I was talking about dislocation related matters in GaN-based devices at the former part of my speech, I suggested the degradation factors only related to dislocations.) I presented that two factors related to dislocations are "temperature rise" and "nanopipes". Dislocation velocities of GaN-based materials are extremely slow at room temperature compared with other III-V compound semiconductors, and this is the reason for the long lifetime of GaN-based light-emitting devices despite their having extremely high dislocation density. THIS is the point I wanted to emphasize most. "However, dislocation velocities of GaN-based materials increase drastically as temperature rises, and in that case dislocation multiplication might occur and influence device lifetime." THIS is a problem for realizing lasers, since the injected current and the voltage is still very high compared to LEDs. (I also wanted to emphasize THIS POINT.) So it is very important to restrain the temperature rise by decreasing device resistivity or improving heat radiation. And another degradation factor related to dislocation is nanopipes which are tend to generate at screw dislocations. Nanopipes decrease when growth conditions are optimazed. However, if they exist, it might lead to device degradation. So it is also important to reduce nanopipes.
My talk about dislocation matters is a result from calculation, and it is not about dislocation motion or the dislocation matters in Toshiba GaN-based devices. We don't really know the main cause of degradation in our device at present. I can only tell that if the device degradation related to dislocation occurs, "temperature rise" might be the main cause and if the "nanopipes" exist, it might be another factor.
Lisa Sugiura Advanced Semiconductor Devices Research Laboratories, RD Center, Toshiba Corp., 1, Komukai Toshiba-cho, Saiwai-ku, Kawasaki 210, Japan E-mail lisa.sugiura@toshiba.co.jp
last updated December 31, 1997 5:07:58 PM.
© 1997 The Materials Research Society
