Summary of the Lateral Epitaxial Overgrowth Workshop, Juneau, Alaska, Aug 2-5, 1999

Summary of the Lateral Epitaxial Overgrowth Workshop, Juneau, Alaska, Aug 2-5, 1999.

This workshop was organized by Prof U Mishra, Univ Santa Barbara, and sponsored by Office of Naval Research, and TMS. The practical arrangements were handled by Shari Allwood, Allwood Inc. Fortunately the weather was excellent during the week, clear skies and warm, so the nonscientific part of the program including various excursions to the wilderness were very much enjoyed.

The main idea of the workshop was to explore in detail the status of understanding of the LEO process, and its relevance for various device applications. Although most of the contributions were concentrated on III-nitrides, LEO-related work on SiC, II-VI materials as well as Si and SiGe were reported and discussed.

The basic idea behind the LEO process is to substantially reduce the threading dislocation density, which is known to interfere drastically with device performance. It was pointed out that there is generally not a strong reduction of the total dislocation density. Most dislocations remain in the material, although they are bent over and moved to a new location, allowing a quite low threading dislocation density in parts of the overgrown wings of the LEO pattern. It was pointed out that most of the claimed dislocation densities in recent literature are probably too optimistic. It is important in TEM studies to consider the volume observed properly, to correctly give a dislocation density. In most cases the total dislocation density in the resulting LEO layer after coalescence is believed to remain in the order of 10⁸ cm^-2, although the threading dislocation (TD) density may well be reduced below 10⁶ cm^-2 in certain areas.

A problem typically occurring is a structural inhomogeneity of the overgrown material, so that a tilt is observed with a different direction on both sides of the coalescence boundary. This is presumably due to the strain on the growing surface in the initial lateral growth phase. This problem appears to be common to LEO and PENDEO structures. Another problem is the often observed n-type donor doping in the LEO layer, apparently irrespective of the mask material. It is not clear whether the origin of these donors is impurities originating from the mask material, or if the high strain around the mask openings perhaps injects native donors in the material during the initial phase of regrowth.

An interesting development is the use of nanostructured templates as seeds for the LEO process. Several different schemes were proposed. One suggestion was to use dense nanosize patterns (with islands of size 20 - 40 nm) on SOI silicon as a template. Growth of GaN in a single growth step is suggested to lead to defect-free material via strain relief before coalescence. Practical demonstration was so far only partially successful, due to problems in preparation of a high quality nanostructured template. A similar process with a weak nanostructured link to a foreign substrate might be possible on several compound material nanostructured templates. A related idea of using porous GaN as a large area substrate for LEO growth of GaN was presented, but no details were given about the material quality obtained.

Much attention was given to growth on silicon. The cracking was recognized as a major problem, still not controlled. An interesting observation is that MBE growth of GaN at low temperatures (about 70 C) seems to be one way around this problem, cracking was reported not to occur for GaN layers as thick as 1.5 um grown on silicon with MBE. The reason for this is still not clear. Otherwise MBE still has problems to produce lateral overgrowth, some progress was reported, but it is still not clear if that will be a feasible technique. There is simply no efficient feeding of growth at the sidewalls in MBE. Growth of suitable low temperature buffer layers by MBE is also not easy. Quite good MBE layers can be produced on GaN templates though.

LEO is also of substantial interest for SiC technology. In this case bulk substrates are available. They are not yet perfect concerning defect density, however. For high power bipolar devices µ-pipes cannot be tolerated, and dislocations (at present 10⁴ - 10⁵ cm^-2) are certainly harmful. Further there is an interest in having larger areas available. Processes are being developed based on the use of large area polycrystalline SiC wafers, which are available in very large sizes, and with an excellent surface finish. Overgrowth with a defect free single crystalline epilayer would be of substantial interest, if successful. Work towards this goal was reported, but there are still early days in this development. A sidetrack from this development is the possibility of gluing two large size crystalline wafers together with an LPE epilayer grown on top, which was demonstrated. LEO on masked areas of single crystalline SiC wafers was also discussed. It was shown that overgrowth is readily achieved, but the work was not taken to or beyond coalescence. The process will be difficult on off axis wafers, since the overgrowth follows the c plane. No detailed defect studies of the overgrown material was reported. An obvious driving force for this work is the possible elimination of µ-pipes and dislocations in the overgrown material.

Another area where LEO potentially would be important is for ir detector arrays based on HgCdTe. It is desirable to produce these arrays integrated on a Si substrate, but the present heteroepitaxial techniques produce too many dislocations in the material. It is desirable to use a CdTe buffer layer on Si, of low dislocation density. LEO of CdTe might be the solution. At present the selective growth of CdTe on exposed CdTe areas of masked substrates has been successfully demonstrated in MBE as well as MOCVD at a reasonably elevated growth temperature (>300 C). The lateral overgrowth still poses a problem, however, possibly more easily performed at the higher growth temperatures available in MOCVD growth.

The use of LEO technology for III-nitride devices was discussed. There is now a common belief that dislocations do have levels in the bandgap, and are active nonradiative recombination sources in devices. The need for dislocation reduction in lasers is well demonstrated. The threshold current of lasers is substantially reduced by using LEO substrates, and the increase in reliability (life time of the laser) is even more dramatic. For LEDs the current density is lower, and dislocations can be tolerated in InGaN-based devices, however in UV LEDs a remarkable improvement with LEO material was demonstrated (weaker localization). Clearly for bipolar vertical devices the total dislocation density is still relevant, and the benefit of LEO may be marginal in some cases. However, much improved reverse bias characteristics is observed for such devices due to the reduction in TD density. For horizontal devices, such as unipolar transistor structures an important improvement in device characteristics, such as reverse bias current and breakdown voltages, has been demonstrated in several cases. Dislocation reduction might not be enough for many devices, however, point defects are also harmful, causing carrier trapping in unipolar transistors, as well as a short minority carrier lifetime in the base of HBTs. This area has not yet been addressed. An important experimental result is that the thermal conductivity of GaN appears to be significantly larger in good LEO material (of the order 2W/cmK), compared to what was reported 25 years ago (1.3 W/cmK).

In summary it is clear that LEO is still a technique that is far from ideal, but very useful at present as no GaN substrates are available. The density of dislocations needs to be reduced even further for many applications, the inherent n-doping could be a problem, as well as cracking in thick structures. The present status of bulk GaN material was discussed at the meeting, there are still no high quality wafers in sight. Judging from the rather slow development witnessed for SiC bulk wafers during the last decade we probably have to wait for a while (several years) before the transition to growth on bulk GaN can be made. But there is a cry out there for bulk GaN wafers, in spite of the strong development of the LEO technique.