Growing Smooth Epitaxial Layers on Misoriented Substrates

The invention, together with its various features and advantages, can be readily understood from the following more detailed description, in which:

FIG. 1 depicts a Nomarski contrast micrograph showing terraces on a 3.mu.m thick GaAs layer grown by LPE (growth rate about 0.1.mu.m/min at 780.degree. C) on a nominally (100) GaAs substrate (misoriented by .phi. = 0.07.degree.); and

FIG. 2 is a Talysurf height profile of the layer of FIG. 1 showing the angle .theta..sub.c = 0.91.degree. between the tread and riser of a terrace (note the 200:1 vertical-to-horizontal magnification).

In order to graphically depict the manner in which the invention is implemented, consider the specific example of a GaAs layer grown by LPE at 780.degree. C on a nominally (100) GaAs substrate. As mentioned previously, if the substrate were cut precisely along the (100) plane, then no terraces would be formed. However, even slightly misorienting the substrate by only .phi. = 0.07.degree. results in numerous terraces as depicted in FIG. 1. Using wellknown Talysurf equipment by Rank Cherr-Tumico Inc., the height profile of the layer was plotted as shown in FIG. 2. Then the angle .theta..sub.c between the tread and riser of each of a plurality of terraces was measured. The angles were averaged to yield .theta..sub.c = 0.9.degree. approximately.

In accordance with the invention, however, misorienting the substrate by approximately .theta.c virtually eliminates the terraces and results in smoother epitaxial layers. As mentioned previously, .theta..sub.c is the tread-to-riser angle, where the tread is typically a major crystallographic plane and the riser is a stable surface or crystal facet. A stable surface or riser is defined in terms of the variation of crystal surface energy E.sub.s versus substrate misorientation .theta.; that is, a stable surface has a minimum surface energy at .theta..sub.c approximately. Put another way, dE.sub.s /d.theta. is 0 at about .theta..sub.c for a stable surface, whereas an unstable surface has no such minimum. If the substrate misorientation is less than .theta..sub.c (but not precisely 0), the crystal surface is in tension and terraces are formed during epitaxial growth. On the other hand, if the substrate misorientation is greater than .theta..sub.c, compression sets in and during growth hill-and-valley instabilities are formed (i.e., an undulating surface).

The above procedure was repeated for several compositions of Al.sub.x Ga.sub.1-x As on GaAs substrates nominally oriented (100) or (111)B and for LPE growth temperatures in the range of 730.degree. C to 880.degree. C. The critical angles .theta..sub.c were measured and tabulated below.

Experimental error in the measurement of .theta..sub.c is about ten percent or .+-. 0.05, whichever is greater. Thus, within experimental error it can be seen than when Al.sub.x Ga.sub.1-x As (x=0 to 0.36) is epitaxially deposited from the liquid phase on a nominally (111)B GaAs substrate at growth temperatures in the range of about 730.degree. to 880.degree. C, terraces are formed from which .theta..sub.c has been measured to range from about 1.0.degree. to 0.5.degree.. The corresponding range of .theta..sub.c for (100) GaAs substrates is about 1.1.degree. to 0.2.degree..

In order to demonstrate that misorienting a substrate by .theta..sub.c virtually eliminates terraces, GaAs--AlGaAs double heterostructures (DH) were grown on nominally (100) GaAs:Si substrates and on GaAs:Si substrates misoriented from (100) by .theta..sub.c = 0.9.degree. .+-. 0.1.degree.. Standard boat and slider apparatus was used to grow the layers by LPE at a growth temperature of 780.degree. C and a cooling rate of about 0.1.degree. C/min. In particular, on the substrate were grown the following layers for devices identified as LZ106 (nominally (100) substrate, and LZ115 and LF711 (substrates misoriented by about 0.9.degree. .+-. 0.1.degree.): an n--Al.sub..36 Ga.sub..64 As layer about 2.0 .mu.m thick, a p-GaAs layer (the active region) about 0.15 .mu.m thick, a p-Al.sub..36 Ga.sub..64 As layer about 0.8 .mu.m thick and a p-GaAs layer about 1.2 .mu.m thick (LZ115) or a p-Al.sub..36 Ga.sub..64 As layer about 1.2 .mu.m thick (LF711).

Inasmuch as these devices are useful as DH junction lasers, the smoothness of the active region, where stimulated radiation is generated, is especially important. The presence of terraces in the active region increases losses and hence probably increases lasing thresholds as well. If the top layer of the DH is p-GaAs (LZ115), then terraces in the active region usually can be inferred from the presence of terraces in the top layer. Terraces in the top layer can be detected by standard Nomarski contrast micrographs. On the other hand, if the top layer is p-Al.sub..36 GA.sub..64 As (LF711), then terraces in the active region may not be reflected in the top layer. However, using standard photoluminescence techniques, the Al.sub..36 Ga.sub..64 As layers can be excited by green laser light to generate carriers which pump the active region. Terraces appear as a spatial intensity modulation of emission from the active region corresponding to a modulation of the thickness of the active region.

Nomarski contrast techniques were used to demonstrate that device LZ106 (nominally (100) substrate) exhibited terraces in the top layer and therefore probably had terraces in the active region, whereas in LZ115 (substrate misoriented from (100) by about 0.9.degree. .+-. 0.1.degree.) the terraces were virtually eliminated. Similarly, photoluminescence techniques were used to demonstrate that device LF711 (substrate misoriented from (100) by about 0.9.degree. .+-. 0.1.degree.) had no terraces in the active region.

Note that in multilayer structures including layers of different composition, such as DH junction lasers or waveguides, some compromise between .theta..sub.c for different layer compositions may be called for. If eliminating terraces in a particular layer (e.g., the action region of a DH laser) is more important than others, then .theta.hd c should be chosen so that terraces are reduced or eliminated in that layer, even though they may be created elsewhere.

It is to be understood that the above described arrangements are merely illustrative of the many possible specific embodiments which can be devised to represent application of the principles of our invention. Numerous and varied other arrangements can be devised by those skilled in the art without departing from the spirit and scope of the invention. In particular, pyramids, which are analagous to terraces have been observed in epitaxial silicon layers grown by CVD on improperly misoriented silicon substrates. It is intended, therefore, that the term terraces as used herein includes pyramids and like crystal irregularities. For pyramids the critical angle is defined as the angle between a major crystallographic plane (analagous to a tread) and any stable surface which forms an inclined side of the pyramid (analagous to a riser). The epitaxial layers were grown on nominally (111) Si substrates by the reduction of SiCl.sub.4 with H.sub.2, a standard procedure. For growth temperatures in the range of about 950.degree. C to 1200.degree. C, the critical angle of .theta..sub.c ranges from about 1.10.degree. .+-. 0.05.degree. to 0.20.degree. .+-. 0.05.degree.. However, in contrast with Group III(a)-V(a) compounds, for silicon the direction of misorientation was found to be important; that is, the substrate should be misoriented from the (111) plane (tread) in the direction of the (112) direction because in the (112) direction the crystal facet is a stable surface (riser). In contrast, in the opposite (112) direction the side of the pyramid is an unstable surface on which defects such as hillocks tend to form during epitaxial growth. Of course, it is well known that (112) is crystallographically equivalent to (121) and (211), but is opposite to (112) which is equivalent to (121) and (211). It is expected that misorienting (111) silicon wafers by .theta..sub.c in the (112) direction will virtually eliminate pyramids.

Finally, another type of surface defect known as meniscus-dissolution lines or rake-marks is observed in the LPE growth of GaAs-AlGaAs stripe geometry DH lasers on nominallly (100) GaAs substrates. These lines correspond to voids in the active region. They are roughly parallel and equally spaced and run normal to the direction of translation of the slider. Although the impact of rake-lines can be reduced by orienting the contact stripe parallel to the lines, it has been found that in relatively clean LPE systems rake-lines are eliminated (1) if the layers are grown instead on (111)B GaAs substrates, whether misoriented or not; or (2) if the layers are grown on GaAs substrates misoriented from (100) by about .theta..sub.c ; and the GaAs saturation seed(s) is misoriented a similar or greater amount (e.g., as much as 5.degree. has proved effective). The purpose of misorienting the saturation seed is to produce a relatively large number of atomic steps on the seed. These steps act to relieve the supersaturation of the source solutions (especially ternary solutions such as AlGaAs), which, in turn, reduces the likelihood that rake-lines will occur. In this context, a conventional LPE system was employed; i.e., a solution holder having a plurality of wells in which source solutions are placed and a seed holder in which both a saturation seed(s) and the substrate are inserted. The seed holder is slidably positioned beneath the wells so that the saturation seed precedes the substrate under each well. For more detail see my article in J. Crystal Growth, Vol. 20, pp. 13-23 (1973).