*3.1. Separation*

A III-V device structure can be separated from its growth substrate by a technique called epitaxial lift-off. It involves the selective lateral wet etching of a thin sacrificial layer. Once the layer is completely etched away, the III-V structure becomes separated from the underlying layers and growth substrate [21,22]. We tailored this technique to detach our HEMT structures from GaAs substrate. The rate of the heterostructure release was studied with dependence on the composition of an etchant based on hydrofluoric acid (HF). The thickness of the AlAs interlayer was the parameter (three values), and the etching was carried out at room temperature. The results are presented in Figure 1.

−

**Figure 1.** Dependence of the release etching rate of HEMT heterostructure nanomembranes on the composition of an HF-based solution with the thickness of the AlAs sacrificial layer as the parameter.

Thin AlAs layers (<100 nm) were not conducive to the HEMT heterostructure release because the sacrificial etching self-terminated at short distances from the edges of lithographically defined nanomembrane areas (a similar outcome was published previously) [23]. The self-termination of the etching was attributed to a very low wettability of the channel formed as the result of the AlAs etching, as the separation between the underlying GaAs buffer layer and the HEMT heterostructure was too narrow. The situation where the etching of the sacrificial layer (thickness of AlAs is 20 nm) was stopped at a short distance is documented in Figure 2. The wettability problem was alleviated by adding a small amount of a hydrophilic agent to the etching solution.

**Figure 2.** SEM image of a sample where the sacrificial etching of a 20 nm thick AlAs separation layer was self-terminated.

Nanomembranes of the HEMT heterostructure that were processed and studied in this experiment were 1.5 <sup>×</sup> 1.5 cm<sup>2</sup> in size. Their release was achieved by the sacrificial etching

of a 300 nm thick AlAs layer. To assure a successful transfer of the nanomembranes, it was necessary to have all surfaces of the nanomembranes as smooth and clean as possible. This would guarantee that the heterostructure adhered well to the host substrate by means of van der Waals forces, especially the surface that was exposed by the sacrificial etching at the original interface between the GsAs layer; the heterostructure had to be clean from any remnants of the AlAs sacrificial etching. To achieve this, the AlAs layer had to have very sharp interfaces with the overlying HEMT heterostructure and underlying GaAs buffer layer.

To look at the bottom surface of the nanomembranes, some were flipped over and attached upside down to sapphire host substrate, as is shown in the SEM micrographs of Figure 3. The white arrows in the main picture and the inset point to the edge of an HEMT nanomembrane where the heterostructure was delineated. The main SEM image shows the back side of a nanomembrane surface, which is that of the GaAs layer, after the sacrificial AlAs layer was completely etched away. This surface was originally the interface between the GaAs layer and the AlAl sacrificial layer, and it should ideally be completely flat and smooth. However, the SEM image revealed that it was not fully smooth but contained a shallow wavy structure with prolonged shallow depressions. The relatively large shallow depressions may have originated due to etching at crystal defects. The shallow wavy structure may have stemmed from the intermixing of Al and Ga atoms as the epitaxial growth was switched from AlAs to GaAs. As a result, a very narrow inhomogeneous AlGaAs layer was formed. As the HF-based etchant etched the AlAs layer away during the release process, it reached this inhomogeneous intermixed layer, etched it off (although more slowly compared with the AlAs layer), and left the shallow surface imprint behind. We believe that it did not significantly hamper the bonding of the nanomembranes to the sapphire substrate.

**Figure 3.** SEM images of an HEMT nanomembrane that was flipped over to expose the surface of its GaAs back side. It contained a shallow wavy structure with prolonged shallow depressions. The arrows mark an edge of the nanomembrane with the HEMT heterostructure delineated.

Figure 4a shows an SEM image of a GaAs nanomembrane attached to the host sapphire substrate. The nanomembrane evidently adhered well to the substrate. Figure 4b shows a detail of the transferred heterostructure nanomembrane with sharp interfaces between the layers of the heterostructure.

Once the nanomembranes became separated from the growth substrate, the etching solution was carefully diluted with water. The membranes were then taken out of the solution and transferred to sapphire substrate. They were dried and thoroughly cleaned in preparation for HEMT processing. Van der Waals forces between the HEMT nanomembranes and sapphire were strong enough for the subsequent processing steps.

**Figure 4.** (**a**) SEM image of a GaAs nanomembrane attached to the host sapphire substrate by van der Waals forces. (**b**) SEM close-up of the transferred heterostructure nanomembrane with sharp interfaces between the layers.
