J. S. Harris, Jr.
Solid State and Photonics Lab, Stanford University
CIS-X 328, Via Ortega, Stanford, CA 94305-4075
GaInNAs on GaAs is a very promising materials system to realize low cost, long wavelength VCSELs. Progress towards realizing these devices for high-speed optical networks is described.
The incredible growth of the Internet and data transmission has pushed the bandwidth requirements for fiber networks at an incredible rate for the past five years. While many are lamenting the recent slowdown in optical networks, it had been moving at a rate that the quickest solutions were being chosen rather than the best. It is thus a good time for a respite to reflect on and develop new approaches to the critical bottlenecks. Certainly the most critical is high-speed access to the fiber backboneÑfrom the desktop to local area (LANs) to metro area (MANs) networks. Three elements are particularly lacking for the optimum network development: 1) low cost, long wavelength vertical cavity lasers (VCSELs) which can be directly modulated at 10 Gbps, operate uncooled and are easily packaged and coupled to fiber, 2) very high density, low power, optically transparent switches and 3) fiber amplifiers that cover the entire 1.25-1.65µm low loss fiber region. I believe all three of these are essential to realize high bandwidth access and refuel the optical revolution. They are also in moderate stages of development and the current slowdown in installation provides an opportunity for these new technologies to catch up and provide a foundation for truly incredible broadband network access from the desktop. While all aspects of these technologies were discussed at length at the recent LEOS Summer Topicals meeting, I will only address the first, new technologies for low cost, long wavelength VCSELs.
|
Figure 1. Transmission distance vs. laser modulation frequency for a variety of optical fiber/laser diode sources utilized in optical networks. |
II. The critical role of such a device in optical networks is clearly illustrated in figure 1. It is clear that in order to achieve 10 Gbps bandwidth LANs and MANs, long wavelength, single mode VCSELs or DBR edge emitting lasers will be required. Long wavelength, InP based DBR lasers have fueled the fiber backbone for the past 25 years. In spite of this, they are still far too expensive and operate with thermoelectric coolers. There are several candidate materials systems for this goal and at least six approaches have been aggressively pursued for the past 4-5 years. I will first discuss the candidate materials systems and challenges to realize such devices and then describe progress on GaInNAs and its promise as the newest, but now, most attractive candidate to realize long wavelength, low cost VCSELs.
Semiconductor lasers operating in the 1.3-1.6µm region require materials with bandgaps between 0.95 and 0.78 eV. One of the requirements for alloy semiconductors is that they must be reasonably closely lattice matched to readily available binary alloy substrates (GaAs or InP). For many years it was believed that there was no suitable alloy lattice matched to GaAs that would emit at >1.1µm, so InGaAsP on InP was the only materials system that met the perceived criteria. All of the long wavelength lasers currently in use today are fabricated from this system. While InGaAsP met the needs of edge emitting lasers, it is almost impossible to meet the requirements for growth of distributed Bragg reflector (DBR) quarter wave VCSEL mirrors, particularly at 1.3µm. This is because the refractive index contrast of InP/InGaAsP is insufficient and the thermal and electrical conductivity are both too low to realize the required combination of high reflectivity and low thermal and electrical resistance. Thermal issues are a particular problem for InGaAsP based VCSELs because of their low To. Because of the thermal performance and limited gain of the active regions in VCSELs, it is of utmost importance to minimize the DBR electrical and thermal resistance. These almost impossibly conflicting requirements have pushed the investigation of alternative approaches which can be divided into “two camps,” those using InGaAsP/InP quantum well active regions, but alternative approaches for the DBR mirrors and those based upon GaAs/AlAs DBR mirror technology, but new active gain region materials that are lattice matched to GaAs. InGaAsP QW based VCSELs have been fabricated using metal mirrors{1}, wafer bonded AlAs/GaAs mirrors{2}, combined InGaAsP/InP and AlAs/GaAs metamorphic mirrors{3}, AlGaAsSb/AlAsSb mirrors{4} and dielectric mirrors{5}. GaAs based VCSEL approaches include InAs quantum dot active regions{6}, GaAsSb/InGaAs Type II quantum wells{7} and GaInNAs quantum wells{8-14}, which is the emphasis of the rest of this paper.
GaInNAs is a surprising candidate for long wavelength emission{8,9}. Contrary to the general rules of III-V alloy semiconductors where a smaller lattice constant increases the bandgap, the large electronegativity of N and its small size cause a very strong bowing parameter and the addition of N to GaAs or GaInAs dramatically decreases the bandgap. By combining N and In, GaInNAs produces a very rapid decrease in bandgap to reach the long wavelength emission region with simultaneous control over bandgap and lattice match to GaAs. Such an active quantum well material allows the fabrication of 1.3µm lasers and particularly VCSELs by combining a 1.3µm QW active region with the well-developed AlAs/GaAs DBR mirror technology.
Research on GaInNAs has revealed several additional factors that favor this system over others to realize a low cost, long wavelength VCSELs. First, for the same bandgap material, the conduction band well is deeper and the electron effective mass is larger{10}, thus providing better confinement for electrons and better match of the valence and conduction band densities of states. This leads to a higher T0 and higher operating temperature, higher efficiency and higher output power. Second, most of the energy band engineering used to minimize heterojunction voltage drops use intermediate graded layers of AlxGa(1-x)As, all of which are lattice matched to GaAs and do not require difficult compositional control over both column III and column V constituents in a quaternary layer to maintain lattice match. Third, compositional control and uniformity of GaInNAs grown by MBE is relatively easy compared to other mixed III-V-V alloy systems[18]. This translates into better yield and far easier scale up to larger wafers for lower cost. Fourth, VCSELs can be straightforwardly fabricated using the well-developed GaAs/AlAs mirror and AlAs oxidation for current and optical aperture confinement technologies.
While all of the above highlight the advantages of the GaInNAs system, there are significant challenges for GaInNAs to produce useful, reliable lasers at 1.3-1.6µm{11-15}. This is an entirely new materials system and there are many unknowns. First and foremost is that of basic crystal structure and materials compatibility: InGaN is a hexagonal crystal, while InGaAs is cubic (zincblende), thus creating a miscibility gap in the alloys{16-17}. Second, these materials are grown at very different temperatures. We have observed many differences in defects, impurity incorporation, annealing, etc. in GaInNAs compared to other III-V alloy systems{16-17}. Improvements to both materials and laser designs will be required to capitalize on the intrinsic advantages of the GaInNAs materials system to realize the exciting device potential for next generation photonic networks.
Our lasers are grown in a combined gas and solid source MBE system using metallic elemental sources of Ga, In and Al, As2, from a solid As cracker and atomic N from a gaseous rf plasma source{11,13}. Edge emitting lasers were initially fabricated to determine the effects of in-situ annealing on radiative recombination and material gain. The properties of the lasers depend strongly on the growth temperature of both the active QW region and the upper waveguide cladding layer. Initial lasers emitted in the range of 1230-1250nm with a threshold current of 450 A/cm2. Oxide confined VCSELs were fabricated for substrate emission at 1230nm[13,17]. The lower Si doped mirror consisted of 22.5 pairs of GaAs/AlAs quarter wave layers, the active region consisted of three 70Å InGaNAs quantum wells separated by 200Å GaAs barriers in a one wavelength long cavity and the upper p-mirror consisted of 20 pairs of AlAs/GaAs quarter wave layers. Ti/Au pads 50µm in diameter were formed by evaporation and lift-off to increase the reflectivity of the top mirror and to make electrical contact to the p-region. The AlAs layers of the lowest three mirror periods of the top mirror also double as an oxide aperture upon selective lateral oxidation to reduce the current and heating. Lateral wet oxidation forms square apertures between 3.6 and 30 µm on a side. The devices were mounted without heat sinking on a glass slide for optical emission through the substrate. The output power and voltage vs. injection current for a 5 X 5 µm device operating cw at room temperature are shown in Figure 2. The threshold current is approximately 1.3 mA, and the slope efficiency is 0.045 W/A.
 |
Figure 2. Emission spectra at 1.3 mA (~Ith) and 3.5 mA (~2.6 Ith) for a GaInNAs oxide confined VCSEL. |
One of the challenges of maintaining long wavelength emission is N outdiffusion during anneal. All of our newer lasers use GaInNAs quantum wells with GaNAs barriers rather than GaAs barriers. The advantages of this new active region design are two fold: (a) the decreased conduction band offset, and (b) decreased blue wavelength shift with anneal. The threshold current density for 0.8mm long devices emitting at 1.34µm was approximately 1.5kA/cm2. Peak pulsed output power was 320mW. This result demonstrates the great potential for long wavelength GaInNAs lasers, but the steep increase in threshold current beyond 1.2µm emphasizes the major challenges of understanding the N incorporation and resulting non-radiative defects in this material.
There are clearly many parameters not yet optimized for these devices, but they demonstrate the exciting potential of GaInNAs on GaAs for low cost, long-wavelength VCSELs and that the technology is currently largely materials and knowledge limited. Extension to longer wavelengths, reducing the threshold current density and improving device performance requires research on the fundamental materials properties and defects in nitride-arsenides. New knowledge on the growth, combined with improved modeling will produce a truly paradigm shifting technology for next generation optical networks.
I am deeply indebted to my former and current students, Mike Larson, Chris Coldren, Sylvia Spruytte, Wonill Ha Vincent Gambin and Mark Wistey. This work was supported by DARPA and ONR through contracts, MDA972-00-1-0024, DAAG55-98-1-0437 and N00014-01-1-0010.