Long-wavelength vertical-cavity lasers

Long-wavelength vertical-cavity lasersDubravko I. Babi´cHewlett-Packard Laboratories, Palo Alto, CaliforniaJoachim PiprekMaterials Science Program, University of Delaware, Newark, DelawareJohn E. BowersDepartment of Electrical and Computer Engineering, University of California, Santa Barbara1§1. IntroductionVertical-cavity surface-emitting lasers (VCSELs) operating at 1300 nm and 1550 nm are potentially low-cost sources for optical communications. The market for multi-mode 1300 nm VCSEL lies in short-distance data-communication links which presently utilize 1300 nm light-emitting diodes and 850 nm vertical-cavity lasers as sources (Fibre Channel). Owing to the larger distance-bandwidth product achievable in silica-based fiber at 1300 nm, the maximum data-transmission rate and point-to-point distance in these links is expected to increase more than two times with the introduction of multi-mode vertical-cavity lasers operating at 1300 nm. Single-mode vertical-cavity lasers operating at both 1300 nm and 1550 nm are also viable choice for sources in telecommunications and wavelength-division multiplexing (WDM) applications due to the expected low cost of VCSELs, the possibility of fabricating arrays [1], and the integrability of the VCSEL structure.GaAs-based VCSELs operating in the 780 - 980 nm range have exhibited tremendous progress in their performance over the last few years: Threshold currents of several tens of microamperes and wall-plug efficiencies greater than 50% have been demonstrated [2] as well as the introduction of the first 850 nm VCL products for High-Speed Asynchronous Transfer Mode (ATM) [3], and Gigabit Local-Area Networks (LAN) by a number of U. S. manufacturers. Even though the first VCSEL operated at 1300 nm [4], the development of long-wavelength VCSELs has been slower over the past few years in comparison to GaAs-based VCSELs owing to several technological difficulties. The key issues have been the difficult realization of high-reflectivity mirrors in the 1300/1550 nm wavelength range and the realization of active layers with sufficient gain at elevated temperatures. The InGaAsP and AlInGaAs material systems lattice-matched to InP (Figure 1) have been the natural choices for the fabrication of long-wavelength lasers [5], [6]. However, neither of these systems offers a large enough range of refractive index values that can produce high-reflectivity quarter-wave mirrors of both p- and n-type at 1300 and 1550 nm. This has prompted a more intense investigation of amorphous-dielectric mirrors and other compound semiconductor-material systems for fabrication of long-wavelength VCSELs. The2reports on long-wavelength VCSELs over the past years have been dealing with very few devices at a time and of a variety of different structures. It has been, hence, a difficult task to develop consistent models for these devices and reach a clear understanding on the mechanisms limiting improved device performance. For this reason, the effort was concentrated on improving all aspects of the laser structures: developing active layers with higher gain, fabricating quarter-wave mirrors with higher reflectivity, and device structures with lower thermal resistance. The device structures presently investigated for long-wavelength applications can be divided into three groups: 1) Etched-well VCSEL structures which use amorphous-dielectric mirrors on both sides of active layer [7], 2) VCSEL structure with one semiconductor and one amorphous-dielectric mirror that utilize ring contacts [8], and 3) All-epitaxial devices which mimic the single-growth GaAs-VCSEL structures [9]. The simplified schematics of these device structures are shown in Figure 2. In recent years, significant progress has been reported in all three device structures, owing to improvements in the epitaxial growth techniques, and the application of fusion-bonding technology to surface-normal optoelectronic devices.This chapter reviews the current status of long-wavelength VCSEL development. We start by discussing the active layers and quarter-wave mirrors for long-wavelength applications, and then describe their practical implementation in various VCSEL structures. We conclude the chapter with a discussion of design issues for vertical-cavity lasers fabricated by fusion bonding, since this technology has been producing the best performing long-wavelength VCSELs at present time.layers§2. ActiveBecause of the shorter gain region, higher threshold gain, and typically higher thermal resistance, the requirements on active regions in VCSELs are more stringent than that of edge-emitting lasers. In addition, at long wavelengths (1300 nm and 1550 nm), the performance of semiconductor lasers is plagued by increased Auger recombination, carrier leakage and carrier-related optical losses [5], [10]. The desired temperature range for operation of long-wavelength3VCSELs is 0 C to 70 C for multi-mode data-communications and up to 85 C for single-mode optical-communications. These requirements place high demands on the design and material development, and leave a narrow margin for fabrication error.Bulk InGaAsP active layers were used during most of the initial development of long-wavelength VCSELs, with quantum wells coming to prevalent use only in recent years [11]. It is now clear that lower transparency and higher differential gain achievable with strained quantum wells is necessary to produce above-room-temperature operating long-wavelength VCSELs. Due to still high (and largely uncertain) values of optical cavity loss and the quantum-well gain saturation, the number of quantum wells used in these devices has always been large (up to 15 wells have been reported). The reason for using a large number of wells is illustrated in Figure 3. This figure shows the dependence of VCSEL threshold current on the round-trip loss and the number of quantum wells positioned optimally at the peak of the optical field. The example threshold-current density is calculated for a typical logarithmic gain-current relationship (constants taken from Ref. [12]). It is evident from this relationship that for every cavity loss value there is a number of wells that produces lowest threshold current density, and this optimum number of wells increases with the round-trip cavity loss. If the value of loss is uncertain, it is always advantageous to use a larger number of wells because of the smaller sensitivity of threshold current to the variations in the round-trip loss.Strained-quantum-well active a layers with large number of wells often require epitaxial growths of epilayers that exceed the critical thickness for stable epitaxial multilayers with no dislocations [13]. In this case the net strain can be reduced or entirely eliminated by utilizing strain compensation, originally employed to increase the differential gain in edge-emitting lasers [14]. The first demonstration of VCSELs with strain-compensated multi-quantum-well active layers were reported in Refs. [15], [16], [17]. It was the application of strain-compensated and near-strain-compensated quantum-wells to long-wavelength VCSELs that has resulted in above-room-temperature continuous-wave operating devices [9] , [18], [19].4The use of separate confinement regions is of key importance in the transverse-mode definition in edge-emitting lasers because it provides index-guiding. In vertical-cavity lasers this requirement does not exist and the separate confinement (SCH) regions may substantially be reduced in size [19], step-graded [20] or completely eliminated [12]. In this way, carrier-leakage-related optical losses (free-carrier and inter-valence-band absorption) in the SCH layers are greatly reduced. The elimination of wide SCH regions improves the high temperature laser performance, but high carrier-related absorption still remains in the multi-quantum-well regions. The effect of the carrier-related losses on laser performance is further intensified by, so called, loss multiplication: The presence of carriers in the active region increases the cavity losses and hence more carriers are required to sustain threshold gain. The additional carriers then raise the loss even further, requiring even more carriers. The positive feedback realized in this way results in a very strong temperature dependence of the cavity optical losses [21], [22]. Carrier-related optical losses [22], [23], [24] have especially been pronounced in long-wavelength InGaAsP active layers which exhibit high carrier leakage due to poor electron confinement (in respect to GaAs system). It has been demonstrated that the electron confinement can be improved by usinga different material system with higher conduction band offset, such as, AlInGaAs grown on InP[6] or on InGaAs substrates [25], and InGaNAs grown on GaAs substrates [26].§3.Quarter-wave mirrors for long-wavelength applicationsThe realization of high-reflectivity mirrors in a particular material system critically depends on the availability of two materials: one of high and one of low refractive index that can be grown or deposited on a specified substrate to form a multilayer structure of alternating quarter-wavelength-thick layers. The number of layers necessary to achieve required reflectivity and the bandwidth of such a mirror primarily depends on the refractive index ratio between the high and the low index material. The larger the refractive index ratio, the smaller the number of layers necessary for a given reflectivity and the broader the mirror stopband. In addition, in the presence of material absorption, the material combination with a larger refractive index ratio produces quarter-wave mirrors with higher maximum possible reflectivity. Beside the optical5properties, the choice of the material system determines the thermal and electrical resistances of the mirrors. Electrical conduction is possible with epitaxially grown mirrors, but not with amorphous-dielectric mirrors. The practicality of a certain material system depends profoundly on all of these parameters, and ultimately determines the performance and the structure of these lasers. Even at present time, the choice for the long-wavelength VCSEL mirror materials has not been settled, as it is with the GaAs-based VCSELs.The peak reflectivity of a quarter-wave mirror depends on the number of quarter-wave layers, the refractive indexes of the incident and the exit media, the high and low index materials, and the material absorption. Figure 4 shows a comparison between reflectivity dependence on number of layers for three quarter-wave mirrors used in 1550-nm vertical-cavity lasers: a Si SiO,2 AlAs/GaAs, and InGaAsP/InP. It is clear that because of a smaller refractive index ratio between InGaAsP and InP, this mirror requires twice as many layers to reach a specified reflectivity as thematerial combination, on the other hand, AlAs/GaAs at the same wavelength. The a Si SiO2requires even fewer number of layers because the refractive index ratio is much larger between the two materials. In real quarter-wave mirrors the reflectivity is reduced by the presence of material absorption and light scattering, which manifests itself as the saturation in the value of the reflectivity in Figure 4. Material absorption has typically been the dominant mechanism determining the peak value of VCSEL mirror reflectivity, due to the small roughness of state-of-the-art semiconductor and amorphous-dielectric mirrors used in vertical-cavity lasers [27]. We consider the effect of material absorption in more detail, while we refer the reader to more complete treatments of optical scattering, for example [28]. The presence of weak scattering in quarter-wave mirrors can be treated as an effective absorption in a simplified treatment described in Ref. [29]).In order to illustrate the mutual dependence of loss in the layers and the refractive index difference we consider the maximum achievable reflectivity of a quarter-wave mirror in the presence of absorption loss. The maximum reflectivity is a hypothetical value of the mirror67reflectivity that would be achieved with an infinite number of layers. It is given analytically for the case of small absorption (always the case in VCSEL mirrors) [30] by:R n n n n n IL H H L H 11222 ()() R n n n n n n n L H H L I H L L 1122222 ()(1)where is the free-space center wavelength. The incident medium refractive index is denoted with n I and the absorption coefficient in the high and low index materials are denoted with H and L . The equations differ depending on whether the first layer index is high (n H ) or low (n L ). In order to observe the effect of loss for all mirrors on equal footing we introduce a normalized relationship for the maximum reflectivity by setting the refractive index of the incident medium equal to that of one of the quarter-wavelength layers. The maximum reflectivity is then written as R max = 1 ()()212n n n , where n (n L n H )2 and n n H n L are the average and the difference between the refractive indexes n H and n L .Here we define normalized absorption coefficient as n (dimensionless), and the fractional refractive-index difference as n n . There are two definitions of n : When the refractive index of the first mirror layer is of high index, we have ( L H )2. If the first layer in the mirror is of low index (in which case the incident medium has a high refractive index), we have L n H 2n L H n L 2n H . The fractional refractive index difference is a figure-of-merit for quarter-wave mirrors and can be used to compare quarter-wave mirrors of different materials in terms of efficiency in realizing high reflection coefficients. In Figure 5 we show a family of constant reflectivity curves in the n n – n coordinate system. The fractional refractive index differences for four epitaxial quarter-wave mirrors and one amorphous-dielectric mirror (a Si SiO 2) are also given for comparison between the mirrors. The mirrors are tuned to 1550nm and 1300 nm (as indicated) and use the corresponding values of the refractive indexes. This representation clearly illustrates that the maximum reflectivity of a quarter-wave mirror is more susceptible to the presence of absorption if the refractive index difference is small.Epitaxial mirrors with small refractive index difference require relatively long growth times, and may be difficult to perform repeatedly. The random variation or drift in the layer thicknesses always results in reduced peak reflectivity. However, it is important to note that even with perfect tuning and an infinite number of layers, the material absorption places the ultimate limit on the maximum reflectivity value. The interplay between the refractive indexes, the absorption/scattering, and the peak reflectivity of the quarter-wave mirrors represents the strongest argument in favor of mirrors with large refractive index difference. This is particularly important to consider in extrinsic epitaxial mirrors. A useful rule-of-thumb deciding whether a certain epitaxial material combination can be used to achieve a peak reflectivity R is given byfor epitaxial mirrors. The constant 1, where we have assumed that n nn Rreflectivity curves in Figure 5 graphically illustrate this relationship for all mirrors. The peak reflection coefficient is achievable if the material combination falls to the left of a specified constant-reflectivity curve.§4.Etched-well VCSELs and amorphous-dielectric mirrorsThe invention of the VCSEL and the subsequent key development of these lasers has been taking place at the Tokyo Institute of Technology, Japan [7]. The first VCSEL, demonstrated by Soda, et al [4], operated at 1300 nm with 77K pulsed threshold current of 900 mA. This structure used two AuZn metal mirrors deposited on both sides of a 90 m thick InP wafer. The high threshold current density was a result of relatively low reflectivity of the AuZn alloyed mirrors (R~80%) and the cavity absorption and diffraction losses. The threshold current of this prototype vertical-cavity laser was subsequently reduced by implementing buried heterostructure active layers, shortening the cavity length to several micrometers by etching vias in the substrates, and finally by using amorphous-dielectric mirrors combined with ring electrodes [7]. The resulting structure has been modified and investigated to present date and is commonly known as the etched-well vertical-cavity laser because of the necessity of etching vias (wells) in the substrate to access the back of the active layer (shown in Figure 2).89In this VCSEL structure, both cavity mirrors are fabricated using amorphous semiconductor and insulator materials which are deposited by low-temperature deposition techniques, such as,electron-beam evaporation, sputtering, and plasma-enhanced chemical-vapor-deposition (PECVD). All of these materials are electrically insulating, but exhibit a relatively large range of thermal conductivities (00125.. W cmK ). Thermal conductivities and refractive index values of select amorphous materials used for long-wavelength VCSELs is given in Table 1along with the fractional refractive index ratio. For long-wavelength vertical-cavity-lasers, the most common material combination has been (amorphous) SiO 2 as the low index and amorphous silicon (a Si ) as high index material [31]. Owing to the large refractive index ratio these mirrors only require a few periods to achieve reflectivity above 99.5%. However, in amorphous semiconductors the absorption tail extends deep into the forbidden gap [32], and hence the reflectivity of these mirrors is often limited by high material absorption (typical absorption coefficients for a Si are 5001cm at 1550 nm and 10001cm at 1300 nm[33]). In addition to the material absorption, the thermal conductivity of both of these materials is quite poor ( 0012.W Kcm for a -SiO 2 [34] and 0026.W Kcm for a Si [35]). For a long time, this has limited the etched-well devices to pulsed operation at room temperature.The recent implementation of a -Si MgO [36], [37] and a -Si Al O 23[18] mirrors in which the low-index material has higher thermal conductance has produced a dramatic improvement of device performance, despite of the fact that the refractive indexes of both MgO and Al O 23 are higher than that of SiO 2, and therefore the maximum reflectivity of mirrors employing these two materials (a -Si MgO and a -Si Al O 23) is lower than that of the a Si SiO 2 combination.Other materials may be used to improve the thermal conductivity of amorphous mirrors without substantial reduction in maximum reflectivity [38]. It is important to note that in order to minimize the thermal resistance between the active layer and the heatsink, etched-well devices are almost always mounted epi-down.The state-of the art in etched-well long-wavelength VCSELs is the structure of fabricated by Uchiyama, et al. [18] (device schematic shown in Figure 6A ). This device employs a partly strain-compensated InGaAsP multi-quantum-well active layer and a current constriction scheme10with two InP/InGaAsP re-growths. Figure 6B shows the SEM micrograph of the cross-section through the re-grown active layer. The 12-quantum-well active layer was formed using Metal-Organic Vapor-Phase Epitaxy (MOVPE). Diamond shaped mesa [39] with sides aligned to (100)planes were defined by silicon nitride patterning and wet chemical etching. The silicon nitride coating served as a mask for the growth of alternating p -InP and n -InP blocking layers around the mesa. It was subsequently removed and the cavity was completed by adding optical thickness to the p -side and the contact layers during the second re-growth. These devices have been mounted epi-down on diamond heatsinks. The heat was taken out through the a -Si Al O 23 mirror which has higher thermal conductivity (as shown in Figure 6a). This device operated continuous-wave up to 36 C with lasing wavelength of 1310 nm. The characteristic temperature T dT d J T th T K 0300 ln () was approximately 50 K.In summary, the etched-well structure, originally plagued with poor mirror reflectivity, current constriction scheme, and thermal resistance, has kept its place as the present day state-of the-art1.3- m VCSEL. This structure has been predominantly used for 1300 nm lasers by many authors [7], [18], [31], [37], [40], [41], [42], [43], but also near 1550 nm [44], [45], [46].§5.VCSELs with epitaxial mirrorsA VCSEL structure with one-epitaxial and one amorphous-dielectric mirror, shown in Figure 2,is attractive because of the possibility of utilizing a planar process with a uniform current injection through a conductive mirror. It necessarily involves ring contacts and a current/mode confinement scheme. We first discuss the properties of epitaxial quarter-wave mirrors relevant to long-wavelength VCSEL applications and then discuss the investigated device structures.§5.1.Properties of epitaxial mirrorsEpitaxial mirrors realized with compound III-V semiconductor alloys of interest for long-wavelength applications exhibit a relatively narrow range of refractive index values..n) [47], [48] and a moderate range of thermal conductivities (00209(2935..W cmK) [49]. Tables 2A and 2B show select material combinations that can be used for long-wavelength applications tuned to 1300 nm and 1550 nm wavelengths. The fractional refractive index difference is listed for every possible combinations as a figure-of-merit. The main advantage of epitaxial mirrors is that they can be made conductive by doping, and, in some cases, they can be realized in-situ at the same time as the active layer.The design of any epitaxial quarter-wave mirror that is also intended for current supply involves the optimization between the reflectivity and the electrical resistance. These two parameters are connected through the doping which effects the absorption and the mirror conduction. The voltage drop across these mirrors depends on the ohmic resistance of the bulk doped layers and on the potential drop across the large number of heterojunction barriers that are incorporated in a typical quarter-wave mirror. Due to lower mobility, larger valence-band offset, and larger effective mass of holes, p-type mirrors exhibit higher resistances at given doping profile. The reduction of heterojunction barriers involves grading and doping of the interfaces to reduce and/or remove the hole/electron barriers [50], [51]. Most grading schemes involve relatively high levels of doping or charge accumulation at the interfaces [51], [52], [53]. However, the interface and bulk doping level has to be carefully selected, because the presence of carriers produces carrier-related material absorption.The thermal properties of epitaxial mirrors vary with the alloy composition. Due to alloy scattering [54], ternary and quaternary alloys exhibit substantially lower thermal conductivities than the binary alloys (even in binary-alloy layers, a reduction of the thermal conductivity is observed due to interface scattering [55]). This somewhat diminishes the attractiveness of epitaxial mirrors that employ ternary and quaternary alloys, but some of the difficulties may be alleviated using top-down mounting and re-growth to provide the necessary thermal conductance.11There are several material choices for the fabrication of quarter-wave mirrors for the realization of long-wavelength vertical-cavity lasers. The deciding factor on what material combinations can be used is the wavelength of the fundamental absorption and the range of the refractive index that can be realized in the transparent regime. The primary compound semiconductor alloys that are used for 1300-nm and 1550-nm quarter-wave mirrors are grown lattice matched to GaAs or InP. Two important properties of the semiconductor refractive index are that all semiconductors exhibit a refractive index decrease below the fundamental absorption edge, and that the refractive index at a given wavelength in the transparent regime generally reduces with the larger energy gap of the semiconductor. Therefore, in the epitaxial mirrors, the high-index material has a the smaller bandgap of the two.§5.2.VCSELs employing InGaAsP and AlInGaAs materialsThe most common choice for epitaxial quarter-wave mirrors for long-wavelength applications is the InGaAsP/InP system lattice matched to InP. A large number of researchers have investigated this system for fabrication of mirrors [56], [57], [58], [59], and vertical-cavity lasers [8], [45], [60], [61], [62], [63], [64], [65], [66], [67], [68], [69]. The main difficulty in realizing these mirrors is their thickness (often more than 10 m ) and their susceptibility to reduction in reflectivity due to the presence of optical loss. Nevertheless, with the progress in MOVPE growth techniques, devices with improved performance are being fabricated. The lowest pulsed threshold-current density measured for a device with one InGaAsP/InP and one a Si SiO2 mirror is 13 kA cm2 reported by Streubel, et al. [68]. The structure and the SEM micrograph of this device are shown in Figures 7A and 7B. The structure consist of an OMVPE grown 50-period InGaAsP/InP quarter-wave mirror topped with a multi-quantum-well active layer. The active layer consists of 9 strain-compensated InGaAsP quantum-well regions grown under a constant As/P ratio [70]. The light output is taken from the 5-period a Si SiOquarter-wave2mirror. Probably the most significant fact exhibited by this laser is that pulsed operation has been demonstrated over a 200 C range (between 160 C and 42 C). This is a very encouraging result that implies that with suitable mode-to-gain-peak offset and low thermal resistance, the full12temperature range required for optical communications (<85 C) may be realized using this type of device. The improvements to thermal and diffraction properties of devices with bottom InGaAsP/InP mirrors can be realized by creating a vertical waveguide surrounded by thermally conductive (lower index) InP. This requires etching tall and vertical pillars in InGaAsP/InP mirrors [57], [71] and then re-growing InP selectively around these mesas [72]. A cross-sectional SEM micrograph of such regrown VCSEL structure is shown in Figure 8 [73]. The mirror is used as the re-growth mask, while two micrographs show that the a Si SiO2planarization is slightly dependent on the crystallographic orientation.The AlInGaAs/AlInAs and AlInGaAs/InP systems have very similar refractive index differences [74] (the latter having a slightly larger refractive index ratio because the refractive index of InP is ~ 0.05 lower than that of AlInAs at 1550 nm). To date a number of researchers have reported such mirrors (AlInAs) [75], [76] and (AlInGaAs/InP) [77], [78], but there have not been any reports of vertical-cavity lasers fabricated using this material combination.In all aforementioned quaternary materials, the bandgap is a continuous function of the composition ending with InGaAs at the long-wavelength side (Figure 1). An optimum composition exists for the narrow gap choice: One desires the highest refractive index (smaller gap) with lowest absorption coefficient (larger gap). The smallest bandgap of the InGaAsP alloy used for mirrors at 1550 nm is in the neighborhood of 0.87 eV [8], [57], [63], [67]. The normalized refractive index difference between this composition and InP is n n 85.% and the mirrors require over 80 layers to achieve reflectivities over 99.5%. At 1300 nm, the smallest bandgap of the quaternary alloy should be in the neighborhood of 1.07 eV which results in an even smaller refractive index difference ( n n 5%). There have been very few reports of using InGaAsP/InP mirrors at 1.3 µm [58].Practically all mirrors employing InGaAsP and AlInGaAs materials reported are of the n-type, because of lower optical losses and better conductance: N-type InGaAsP/InP mirrors exhibit low resistance and high reflectivity [57], [68] sufficient for fabrication of VCSELs, while there are13。