Is XeF2/Ge Etch Process The Future For Epitaxial Liftoff?
Researchers in Austria and the U.S.A have demonstrated a novel room temperature lateral etching process which can be applied to AlGaAs devices with arbitrary aluminum content.
The scientists, led by Garrett Cole at the University of Vienna, used a high quality III-V/Ge/III-V epitaxial heterostructure and a unique gas-phase etching technique to demonstrate Epitaxial Lift-Off (ELO) of arbitrary aluminum content AlGaAs films.
Xenon difluoride (XeF2) was used to spontaneously etch an embedded Ge sacrificial layer in the epitaxial heterostructure, freeing the overlying layers.
Micromachining processes for monocrystalline GaAs, InGaAs, InGaP and low aluminum containing AlGaAs compounds have previously been used to produce high-quality free-standing heterostructures. Typically, a sacrificial AlAs layer is selectively removed through immersion in hydrofluoric (HF) acid.
This process is commonly used to fabricate devices transferred to foreign substrates. Such devices include HEMTs, LEDs, laser diodes, photodetectors and solar cells. According to Cole, the drawback of HF is that it is not selective to AlGaAs films with aluminum contents over ~40%; this makes it difficult to release arbitrary epi-structures with high aluminum content layers. Also, most relevant dielectrics (used in AR coatings, passivation layers, etc.) are rapidly etched in HF and thus must be protected or added subsequent to release, further complicating the process. Finally, for undercutting compounds containing greater than 40% aluminum, selective etching of GaAs in a buffered citric acid solution is typically employed. However, this process is not selective enough to use in ELO.
Another limitation to the standard HF-based ELO process—the selective removal of a sacrificial film for the release of epi layers from their original handle substrate—is its slow lateral etch rate (on the order of 0.1 mm/h) when attempting to lift-off feature sizes larger then a few millimeters. This reduced rate arises due to the slow liquid phase diffusion of the acid to the etch front through the narrow gap left by the etched sacrificial layer. Techniques have been developed to enhance the etching speed with maximum rates on the order of 10 mm/h realized by mechanically opening the gap during the etching process. Unfortunately, this requires at least one of the substrates (handle or donor) to be flexible and a complex apparatus may also be necessary to mechanically deform the compliant substrate.
There are a number of interesting aspects in regards to the gas-phase etching process discussed in this new publication. Unlike the wet etching processes previously described, this technique is a completely dry process, with both etchant and etching product (mainly Xe and GeF4) in the gas phase, resulting inextremely fast etch rates, even without the use of mechanical rate-enhancing techniques. According to Cole, he has recorded etch rates exceeding 150 um/min (>9 mm/h) with amorphous germanium films, while collaborators at MIT have recently achieved similar results, with lateral etching rates up to 9 mm/h for thin epitaxial Ge layers. With the incorporation of rate-enhancing techniques in this new process, the etch rate has potential to be improved further.
Besides the high etch rate, this technique does not require a plasma, so no ion damage is generated in sensitive devices. High selectivity etching occurs at room temperature with the elimination of surface tension forces in the release of suspended structures.
The gas phase process also eliminates the need to submerge the sample in caustic liquid solutions that may damage sensitive layers, or lead to the collapse of fragile thin-film structures; in this case all that is necessary for etching is a simple vacuum chamber and valving to control the volume of XeF2 introduced to the sample chamber.
Cole points out that the neat thing about XeF2 etching is that the system requirements are minimal. In this case, the large viewports on the vacuum chamber allow for real-time viewing of the etching process. Adding a stereo microscope also makes it possible to actually watch the rapid progression of the etch front under transparent films in real time, further simplifying device processing.
Finally, the process is highly selective to polymers (e.g. photoresist), dielectrics (including SiO2, TiO2, SiNx, Ta2O5, Al2O3), and most relevant metals - allowing for the lift-off of fully processed structures in addition to high aluminum content AlGaAs.
Cole says that he initially accidentally encountered this rapid etching process in 2006 while developing micromechanically tunable VCSELs at Lawrence Livermore National Laboratory using solid, rather than polymer, sacrificial layers for the MEMS actuator.
According to Cole the procedure turned out to be very useful for tunable VCSEL development as it greatly simplified the process flow by eliminating the need for critical point drying. He continues, “It also allowed me to use a more robust inorganic sacrificial layer, enabling high temperature deposition steps for the actuator layers and dielectric DBR, which was not possible with the limited glass transition temperature of resist-based sacrificial films."
He had originally tried etching sacrificial Ge thin films with H2O2, but the slow rate and violent bubbling made the yield very poor. The chemical similarity between Si and Ge inspired him to then try etching the Ge in a new XeF2 system. He says, “I was amazed at the etch rates I saw - in fact the first samples were basically obliterated in the chamber. “
Cole and his collaborators at MIT have identified a number of immediate applications, generally centered on transferred thin-film optoelectronic structures. They suggest that XeF2 etching of Ge would be interesting for AlGaAs-based micromechanical devices, while the XeF2/Ge ELO process would be applicable to VCSELs, LEDs, and micro-electronic devices that could then be transferred to a foreign handle wafer.
Furthermore, this process may be an interesting alternative to hydrogen exfoliation processes for generating semiconductor-on-insulator structures at a low cost and over a large area; in this area the Smart Cut process (an implant-based layer separation technique) is currently the most prominent method.
The researchers emphasize that the ability to grow high quality III-V/Ge/III-V epitaxial heterostructures is of great importance in developing this technique. The valence discontinuity at the III-V/Ge interface tends to introduce defects during epitaxy and has long been a challenge for film growers. To create a high quality III-V/Ge/III-V heterostructure, both interfaces need to be carefully treated for correct epitaxial initiation. Recently, Cole’s collaborator, Yu Bai, a researcher in Professor Eugene Fitzgerald's group at MIT in the U.S.A., has achieved high quality Ge epitaxy on a III/V underlayer with low pressure MOCVD.
On the other hand, the technology of GaAs-based heterostructures on Ge has previously been demonstrated for very high performance photovoltaics—utilizing III-V layers on Ge wafers—as commercialized by Spectrolab and Emcore. Emcore's work on inverted solar cell designs from the National Renewable Energy Lab (NREL) is particularly interesting as they do not rely on a Ge junction, which would obviously be destroyed in this process.
To test the viability of the technique, the researchers tested two processes; the first focused on the development of a suspended GaAs/AlAs Distributed Bragg Reflector (DBR) grown on a Ge substrate. Secondly,they explored ELO of GaAs films via removal of an embedded Ge sacrificial layer in a GaAs/Ge/GaAs epitaxial heterostructure. As mentioned above, the work required high quality III-V/Ge heterostructures and the epi-wafers were supplied by collaborators from Professor Eugene Fitzgerald's group at MIT.
For the first demonstration process the researchers fabricated low-dissipation optomechanical resonators. These devices were created using a bulk micromachining method involving photolithography and reactive ion etching through the mirror stack. Finally, the devices were undercut through exposure to XeF2. For this process anetch rate of approximately 12 µm/min was recorded; comparable with the maximum achievable rates found in sacrificial silicon etching(see Figure 1).
Fig 1. Microfabricated monocrystalline optomechanical resonator undercut using the XeF2/Ge gas-phase lateral etching process. The device consists of a high-reflectivity 40.5-period GaAs/AlAs distributed Bragg reflector (DBR) grown on an off-cut Ge substrate. The use of the epitaxial mirror stack results in both high reflectivity and low mechanical loss, while the simple bulk micromachining process enabled through XeF2 etching allows for high yield and excellent geometric control.
For the ELO experiment, single-crystal epitaxial GaAs films were released through removal of an embedded epitaxial Ge sacrificial layer. The samples used in this study were 1 cm2 in size and lateral etch rates of 3 mm/h and a conservative selectivity of ~ 106 were achieved. Compared with early results for HF-based ELO etch rate of 0.3 mm/h (best results to date of 30 mm/h, using a mechanical etch-rate-enhancing apparatus), the lateral etch rate for the demonstration gas-phase XeF2 process shows similar selectivity, is an order of magnitude faster, and additionally enables the release of epitaxial films with arbitrary aluminum content (se Figure 2).
Fig 2. Epitaxial liftoff using XeF2 etching of a GaAs/Ge/GaAs heterostructure. In this case a nominally 150-nm thick GaAs film is released from a cleaved 9 mm × 9 mm chip. The plot shows the measured reflectance curve (along with a theoretical fit) for the GaAs layer transferred to an adhesive backing, while the inset includes (left) a simple schematic of the layer structure before and during etching, plus (right) a photograph of the free-standing film and growth substrate following the etch procedure. Etch rates up to 9 mm/h have recently been achieved.
Further details of this work are reported in ‘Applied Physics Letters’, 96, 261102 (2010); doi:10.1063/1.3455104.
The next step for the scientists is to push the limits of the existing process and scale this up to relevant wafer sizes. Beyond that, they hope to explore the release of optoelectronic structures, particularly photovoltaics at MIT, and intend to identify any Ge lattice-matched compound semiconductor alloys that exhibit a spontaneous chemical reaction with XeF2.
The work described here was funded by European Commission, the Austrian Science Fund and the European Research Council. Microfabrication was carried out at the Zentrum für Mikro- und Nanostrukturen (ZMNS) of the Technische Universität Wien. For the US-part of collaboration, the work was funded by Chesonis Family Foundation, Army Research Office (ARO) and the Singapore-MIT Alliance for Research and Technology (SMART) Innovation program.