Structure for Indium Phosphide/Indium Gallium Arsenide Phosphide Buried Heterostructure Semiconductor

This application is related to an application entitled "Processing Method for Fabricating Electrical Contacts to Mesa Structures in Semiconductor Devices," Ser. No. 07,583,410, filed simultaneously herewith on Sep. 4, 1990 by R. F. Holmstrom, E. Meland and F. D. Crawford, inventors, and assigned to the same assignee as this application, which issued as U.S. Pat. No. 5,093,225 on Mar. 3, 1992.

This invention pertains to semiconductor lasers and more particularly to high modulation bandwidth single transverse mode semiconductor lasers.

As fiber optic communication systems continue to push to even higher data rates and larger bandwidths, the availability of reproducible and reliable high-speed lasers becomes increasingly important. This is particularly true for microwave modulation schemes like subcarrier multiplexing, where low relative intensity noise (RIN) is very important, or for phase modulated systems, where the avoidance of phase noise is crucial. In both of these modulation schemes it is very important to avoid operating near the resonance peak of the laser, where RIN peaks and a .pi. phase shift occurs. Both of these phenomena have a deleterious effect on the noise performance of the system. The further below the resonance peak the fiber optic system can be designed to operate, the better its noise performance will be. Clearly, demands on the system designer are reduced by the availability of lasers with higher resonance frequencies and modulation bandwidths.

Fabrication of single transverse mode semiconductor lasers with modulation bandwidth in excess of 15 GHz depends heavily upon both the accurate control of the active layer doping, width, and thickness, and upon providing a lateral optical cladding of the active layer which minimizes surface recombination and carrier leakage while not affecting the single transverse modal properties of the laser. In addition, a low capacitance and low series resistance structure is required to minimize electrical parasitics.

High frequency single transverse mode semiconductor lasers have been achieved by constricted mesa (Bowers, et al, "High-speed InGaAsP constricted mesa lasers", IEEE J. Quantum Electron., Vol. QE-22, pp. 833-884, June 1986), mass transport (Liau, et al, "A novel technique for GaInAsP/InP buried heterostructure fabrication", Appl. Phys. Lett., Vol. 40, pp. 568-570, Apr. 1982) and vapor phase regrowth (Su, et al, "Ultra-high frequency modulation of InGaAsP lasers", Tech. Dig. Conf. Optical Fiber Communication, pp. 90-91, Feb. 1986) techniques. These structures have broad mesa tops so as to reduce series resistance and facilitate the formation of ohmic contacts. In order to achieve this structure, the processes reported to date require selective wet chemical etching of the active layer of 4 microns or more to reduce its width and/or the epitaxial growth of thick layers of high quality p-doped, semi-insulating, or pn blocking layer configurations. These regrown materials are subject to unintentional doping to and from the p-type device structure layers during the regrowth process and tend to be electrically leaky which severely limits the achievable output power and bandwidth. (Ohtoshi, et al, "Current leakage mechanism in InGaAsP/InP buried heterostructure lasers", 11th IEEE International Semiconductor Laser Conference, Boston, Mass., 1988). In addition, wet chemical etching through dielectric masks, used to form the initial mesa structure, is inherently non-uniform so that control of active layer width, a necessity for high bandwidth single transverse mode operation, is very difficult resulting in low yield and non-reproducible results. Alternatively, the fabrication of narrow mesa tops leads to the additional problems of high series resistance and difficulty in photolithographically defining and forming ohmic contacts.

Accordingly, it is a primary object of this invention to obviate the above noted and other disadvantages of the prior art.

It is a further object of this invention to provide a high modulation bandwidth single transverse mode semiconductor laser.

It is yet a further object of this invention to provide a controllable and reproducible fabrication process for such a laser.

In a first aspect of the invention, an indium phosphide-based semiconductor laser is produced such that the epitaxial layer configurations (composition, thickness and doping) are appropriate for high frequency laser applications. This is accomplished by sandwiching a thin, narrow active layer of small bandgap material between thicker, larger bandgap material and filling the interstices on each side of the active layer with a semi-insulating in such manner that the sidewalls of the device form a smooth surface, thereby reducing the homojunction capacitance and carrier leakage.

In a second aspect of the invention, the method of fabrication of a 1.3.mu. InGaAsP laser with three dB bandwidth of 24 GHz and intrinsic resonance frequencies uses a dry-etched vapor phase regrowth structure to fabricate lasers from LPE-grown, 1.3.mu. double heterostructure material, or any suitable epitaxial growth technique, i.e., VPE, MOCVD, and MBE) that will produce InGaAsP/InP double heterostructure material. The method comprises the steps of dry-etching channels, followed by selective etching to undercut the active layer, then InP regrowth on the sidewalls in a halide vapor phase system. SiO.sub.2, and the p- and n-contact metallizations are alloyed ZnAu and SnAu, respectively, using a flood exposure technique.

FIG. 1 is a cross-sectional view of a semiconductor wafer from which an embodiment of the laser of the present invention can be formed;

FIGS. 2a through 2f are cross-sectional views of a portion of the wafer during various stages of fabrication of the embodiment of FIG. 1;

FIGS. 3a and 3b are cross-sectional views of a portion of the wafer during various stages of fabrication of the embodiment of FIG. 1;

FIGS. 4a through 4d are cross-sectional views of a portion of the wafer during various stages of fabrication of the embodiment of FIG. 1;

FIG. 5 is a perspective view in cross-section showing the structure of the injection type semiconductor laser embodiment of the present invention;

FIGS. 6, 7, 8, and 9 are diagrams illustrating operating characteristics of an embodiment of a laser in accordance with the present invention.

An injection type semiconductor laser device embodying the present invention will now be described in detail by reference to a laser device of a InGaAsP-InP double heterostructure as an example.

A typical wafer as an initial starting point in the fabrication process is illustrated in FIG. 1. The substrate 10 is typically composed of crystalline InP and is covered by a buffer layer 11 of n-InP of about 1-2 microns thick. The active layer 12 is composed of In.sub.0.73 Ga.sub.0.27 As.sub.0.63 P.sub.0.37 and has a thickness ideally of 0.2 microns. On top of the active layer 12 is a cladding layer 13 of InP approximately 1 micron thick. The fourth layer comprises a cap layer 14 of In.sub.0.80 Ga.sub.0.20 As.sub.0.40 P.sub.0.60 of from 0.2-0.3 microns in thickness. A wafer of such composition and structure can be made by techniques well known in the art by chemical vapor deposition (CVD), liquid phase epitaxy (LPE) or a combination thereof and, and include all compositions of In.sub.1-x Ga.sub.x As.sub.y P.sub.1-y that would produce high frequency lasers emitting at .lambda.=1.25 .mu.m to .lambda.=1.58 .mu.m and all wavelengths between.

The initial step to produce the laser of the present invention is to define the mesa. The mesa width is determined by the desired final active layer width, the desired amount of regrowth on both sides of the active layer and any processing variables encountered in the photolithographic, dry etching and wet etching steps. As an example, if a one-micron.+-.0.1 micron active layer width with 0.1-0.2 microns of regrowth per side was desired, the initial mask for the mesa would be in the order of 2,0 microns. The first step is to deposit a mask material 15 for example, 0.3 microns of SiO.sub.2, on the epitaxial cap layer 14 that is compatible with said layer and with the dry and wet processing chemistries used in defining the mesa. The next step is to apply a layer of photoresist material 16 using standard photolithographic techniques. The next step is to align and delineate stripes in the photoresist 16 along the (110) crystallographic direction of the InP material, as shown in FIG. 2a. The exposed mask material is then plasma etched in accordance with well known plasma etching techniques, for example utilizing CF.sub.4, to expose the underlying InGaAsP cap material. After the photoresist is removed, the wafer is subjected to plasma etching in accordance with well known plasma etching techniques, for example C.sub.2 H.sub.6 /H.sub.2, to etch through the InGaAsP cap layer 14 and stopping in the InP cladding layer 13 in regions not protected by the SiO.sub.2 mask, resulting in a wafer as shown in FIG. 2b. The wafer is then subjected to a wet chemical etch of a 1:9 ratio of HCl and H.sub.3 PO.sub.4 acid mixture to selectively and anisotropically remove the InP cladding layer 13 using the InGaAsP cap layer 14 as a mask, resulting in a wafer having the appearance as shown in FIG. 2c. The wafer is again subjected to a plasma etch with the SiO.sub.2, the cap and the cladding layer now acting as a mask to etch through the InGaAsP active layer 12 stopping in the InP buffer layer 11 resulting in a wafer as shown in FIG. 2d. The wafer is again subjected to a wet chemical etch using the aforementioned HCl and H.sub.3 PO.sub.4 acid mixture for a time sufficient to remove a portion, approximately one-half, of the exposed InP buffer layer, using the InGaAsP active layer as a mask, as shown in FIG. 2e.

The wafer is again exposed to wet chemical etching, this time in a solution of K.sub.3 Fe(CN).sub.6 :KOH:H.sub.2 O, which is well known, to selectively etch the InGaAsP cap 14 and active layers 12 laterally as shown in FIG. 2f. In view of the difference in composition of the active layer 12 and the cap layer 14 the solution etches the active layer at about twice the rate of the cap layer. Since the amount of lateral etching required is only on the order of 0.4 microns or less, the resulting active layer is very uniform across the entire wafer. With the mesa, the active layer width and the cap width now defined, the SiO.sub.2 mask is removed by hydrofluoric acid.

The wafer is then exposed to vapor phase epitaxial growth of an overlayer of semi-insulating InP 30 as shown in FIG. 3A. The amount of regrowth is limited to minimize intentional diffusion of dopants into and out of the various device layers thus degrading the quality of the devices. The exact amount and uniformity of regrowth is critical only to the extent that the under cut regions be completely filled laterally. The wafer is again etched in the 1:9 hydrochloric and phosphoric acid mixture, leaving an unetched amount of regrowth cladding 31 and 32 on the side wall of the active layer as shown in FIG. 3B the lateral dimension of this cladding being determined solely by the relative undercutting of the active layer 12 with respect to the cap layer 14 and in view of the relative etch rate of the lateral etching and the amount of etching of the active layer (i.e. about 0.3-0.4 microns) the width of the cladding will be about 0.1-0.2 micron uniformly along the active layer. This technique automatically controls the width of the regrown InP and symmetrically aligns the regrown material 31 and 32 around the active layer 12. This is in contrast to the other techniques where either the active layer width is poorly controlled due to oversized mesa and/or the amount of regrowth is relatively large (e.g. greater than 1 micron) or highly nonuniform for regrowth of less than 0.5 microns. The advantage of achieving a thin regrown region is that it reduces homojunction capacitance and carrier leakage.

A dielectric material, for example approximately 0.4 microns SiO.sub.2, is deposited on the mesa side of the wafer, conforming to and encapsulating the mesa, the channels and covering the cap layer. This dielectric serves a number of purposes; it defines the p-contact using localized flood exposure, as disclosed in the cross-referenced application; it acts as a diffusion barrier during the diffusion step; it electrically isolates the p-metal overlay from the cap layer and is intrinsic to the reduction of the electrical parasitics of the device.

The contact openings are formed by using the localized flood exposure described in the copending application filed of even date herewith and assigned to the same assignee as the present application and is incorporated herein by reference thereto. In accordance with this method, the enitre mesa top and an appreciable amount of the mesa sidewall is bared as is shown in FIG. 4a. Diffusion of an acceptor dopant, such as zinc, to p-type epitaxial layers can be performed as shown in FIG. 4b so that the entire exposed surface is electrically modified to yield a higher differential gain, lower resistivity and lower contact resistance. By controlling the amount of exposed sidewall, a feature of this technique, and the diffusion time and temperature, the active layer can be heavily doped beyond the levels possible in as-grown epitaxial layers, a necessity for achieving high frequency operation. Ohmic contact metal 44 may then be deposited over the entire exposed surface such that the largest area of contact possible in this structure is achieved as shown in FIG. 4c and thus the lowest possible contact resistance in this structure is achieved without increasing the parasitic junction capacitance. In addition, this ohmic metal contact process results in thermal properties that are superior to the other structures. A variety of other process techniques such as thick dielectrics 46 to reduce parasitic chip capacitance and to permit "epi down" mounting are performed. A resulting structure is shown in FIG. 4d.

The substrate, or bottom, side of the wafer is thinned to facilitate cleaving. An electrode is formed on the bottom of the wafer, opposite to the ohmic contact on the top of the mesa structure, the wafer is divided into chips and made into laser devices as illustrated in FIG. 5 in a manner well known in the art.

The intrinsic resonance of a semiconductor laser is proportional to the product of the total loss, the differential gain .differential.g/.differential.N, and the photon density. If these parameters are optimized, it is then necessary to minimize electrical parasitics (i.e., the RC time constant) to fully exploit the modulation capabilities of the laser. Past work on ultra-high-frequency diode lasers has been plagued by the difficulty of fabricating lasers with very small active areas, low leakage, and low electrical parasitics. This invention utilizes a dry-etched, vapor-phase regrowth structure to fabricate lasers from LPE-grown, 1.3 .mu.m double heterostructure material. This technique is a modification of the wet-etched, vapor-phase-regrowth technique previously used. Basically, it consists of dry etching channels, followed by selective etching to undercut the active layer, then InP regrowth on the sidewalls in a halide vapor phase system. SiO.sub.2 is used as a mask and for electrical isolation, and the p- and n-contact metallizations are alloyed ZnAu and SnAu, respectively. The dry-etched perpendicular sidewalls result in two improvements compared to previously wet-etched vapor-phase-regrown lasers:1) well-controlled regrowth widths of about 0.2 .mu.m per side. Since this regrowth forms the p-n blocking homojunctions, limiting this width reduces homojunction leakage (thus improving the linearity of the power-current curves) and minimizes diffusion capacitance (thus extending the RC limit); 2) the entire top of the mesa was contacted, using a flood exposure technique, thus reducing the contact and series resistance of the device.

A third improvement was aimed at increasing the photon density in the laser cavity. Because of the perpendicular mesa sidewalls and the minimized regrowth, it was possible to accurately control the width of the emitting region to 0.9.+-.0.1 .mu.m. This dimension, coupled with active layer thicknesses of about 0.2 .mu.m and cavity lengths of 100 to 130 .mu.m, resulted in extremely high photon densities in a single spatial mode, even at moderate drive currents.

Lasers made in accordance with this invention have achieved a modulation bandwidth of 24 GHz at room temperature and intrinsic resonance frequencies in excess of 22 GHz. The lasers are also characterized by a linear power-current curve, resulting in excellent modulation efficiencies even at 24 GHz. The swept frequency response (S.sub.21) is shown in FIG. 6 of a laser with a 3-dB bandwidth of 24 GHz. The 3-dB bandwidths for lasers according to this invention ranged from 18 to 24 GHz for lasers with cavity lengths of about 130 microns. FIG. 6 shows that the modulation efficiency at 24 GHz has dropped only 3 dB with a 100 mA bias current compared to its low-frequency, low-bias values.

The intrinsic resonance frequency, f.sub.o, of the lasers was found from the measured relative intensity noise (RIN) spectral density. The RIN spectrum, shown in FIG. 7, indicates that even at bias currents as low as 60 mA the response peak is beyond 18 GHz. These data, together with the S.sub.21 data of FIG. 6, indicate that the intrinsic resonance of this laser is in excess of 22 GHz. By fitting the RIN data to the standard single-mode rate equation solution, the resonance frequency, f.sub.o, and damping rate, .gamma., are accurately determined, free of any electrical parasitics. The maximum achievable bandwidth in the absence of parasitics (i.e. limited only by damping) can then be estimated, as shown by Olshansky, et al, IEEE J. Quan. Electron. QE 23, p. 1410 (1987), using the relationship f.sub.3dB =8.8/K, where K=.gamma./f.sub.o.sup.2. FIG. 8 shows the fitted .gamma. versus f.sub.o.sup.2 for the same laser. From the slope, K is found to be 0.19 ns, while more typical values are approximately 0.3 ns. These values of K suggest that the maximum damping-limited bandwidth could be as high as 44 GHz.

The intrinsic resonance frequency is plotted as a function of the square root of the optical power in FIG. 9. The slope of this line is 8.9 GHz/mW.sup.1/2. We believe this to be the highest value ever reported for a semiconductor laser, more typical values of this slope are 5 to 6 GHz/mW.sup.1/2 for semiconductor lasers. Because of the extremely high value of this slope, large bandwidths are obtained at relatively low operating currents. Specifically, a 3 dB bandwidth of 20 GHz is achieved with as little as 55 mA of drive current. This may result in improved long-term reliability compared to previous high-frequency lasers.

A record modulation bandwidth of 24 GHz at room-temperature is reported for 1.3 .mu.m lasers fabricated using a dry-etched, vapor phase regrowth technique. This and other dynamic characteristics are significantly better than the previous record of 22 GHz set by 1.3 .mu.m VPR-BH lasers, and the 15 to 18 GHz bandwidths which are more typical for high-frequency InGaAsP laser structures. Furthermore, we observe intrinsic resonance frequencies in excess of 22 GHz, the highest yet obtained for either conventional double heterostructure or quantum well InGaAsP lasers. These improvements are accomplished by using a dry-etched, vapor phase regrowth technique. These lasers exhibit reduced leakage currents and electrical parasitics, and are characterized by a linear power-current curve, resulting in excellent modulation efficiencies even at 24 GHz.

We have demonstrated 1.3 .mu.m lasers with intrinsic resonance frequencies in excess of 22 GHz, and 3 dB bandwidths of up to 24 GHz. This is a significant improvement over the previous record of 22 GHz for the 3 dB bandwidth. These advances were achieved primarily by increasing the photon density in the lasing cavity (to increase the intrinsic resonance frequency), and by minimizing electrical parasitics. It is expected that these devices will play an important role in the realization of extremely broadband fiber optic systems in general, and in microwave multiplexed systems such as subcarrier systems in particular.