2.1 GaN Cap Thickness

Here, R represents a hydrocarbon that combines with sulfur to form ethanethiol. The scanning electron microscope (SEM) images in Figure 2 (middle and bottom) show two different samples where the Au layer forms a very porous structure with coverage ranging from 50% to 65%. This result is consistent with other reports for thin Au films where the porous nature of this thin film is even desirable since the high surface to area ratio could improve sensitivity.^[20]

The shift in peak energy against the different cap layer thickness is plotted in Figure 3 (top) where a general trend of decreasing shift is seen with an increase of the cap layer thickness. Since the step size is very small, that is, 0.5 nm, the measured shift values are close. Cap layer thickness greater than 6 nm resulted in an almost complete loss of the PL signal, so no shift could be measured. The sensor response with a bare GaN surface is also shown, which stays around 1–1.5 meV and generally also trends down with an increase in the cap thickness. The Au functionalization layer however results in a larger shift proving the affinity of gold toward hydrogen sulfide. These results are consistent with previously published work.^{[11, 13]}

In Figure 3, bottom, the root mean square (RMS) surface roughness (ranging from about 6–10 nm) and the covered area percentage of the gold layer is shown. Larger pores on the metal surface increase the surface-to-volume ratio and have shown improved sensitivity. However, in this case, no correlation with the sensitivity could be observed. These values are within the range where they seemingly have little to no effect on sensitivity.

2.2 QW Thickness

For this series of experiments, the QW thickness was varied from 3 to 5 nm in 0.5 nm steps as a thicker QW may experience a larger field resulting in a bigger shift. The rest of the parameters were 3 nm GaN cap and Au layer each and medium high doping concentration (5 × 10¹⁸ cm⁻³). In Figure 4 (top), the energy shift is plotted over the QW thickness for 100 ppb of H₂S. After seemingly a small dip, the shift increases with an increase in the QW thickness. As expected, the 5 nm thick QW produces the highest shift of about 6 meV. GaN surface without functionalization remains fairly flat for the whole series around 1 meV and is also not affected by the QW thickness. The bottom of Figure 4 shows the surface roughness and the coverage percentage of the Au layer. As the roughness decreases, the covered area increases. However, the dip in the 3.5 and 4 nm QW could not be explained by the surface analysis.

2.3 Doping Concentration Variation

The optimal doping level in the GaN buffer layer was also investigated. Five samples were grown where the doping concentration ranges between about 1 × 10¹⁷ and 1 × 10¹⁹ cm⁻³. The GaInN QW and the GaN cap layer thickness were kept constant at 3 nm each. The surface was functionalized with a 3 nm thick Au layer. Simulation results presented in ref. 13 have shown a general trend of increased sensitivity with increasing doping concentration up to a certain optimal value. However, surprisingly, the measured results for 100 ppb of H₂S sensing showed an opposite trend as can be seen in Figure 5 (top). The shift in peak energy drops from about 10–4 meV as we go from lowest to highest doping. Response of GaN surface without metal can also be seen in the graph as it remains flat around 1 meV. An interesting result was the complete loss of the PL signal for the highest doped sample (1 × 10¹⁹ cm⁻³) after the Au functionalization. Figure 5 (bottom) shows the Au layer surface roughness and coverage values being similar to the previous samples, so nothing out of the ordinary at the surface could explain this change in expected behavior.

2.4 Sensing at Higher Temperatures

Some sensing measurements were performed at higher temperature to better understand the behavior of the sensor and to test whether the response and recovery time could be improved. However, the PL intensity decreases significantly with increasing temperature making such measurements extremely difficult. Some results are presented in Figure 6. The top figure shows the sensor response of a sample with a 5 nm QW with 3 nm GaN cap and Au functionalization layer sensing 10 ppm H₂S at different temperatures. The dip from 50 to 100 °C is presumed to be caused by some condensation in the chamber evaporating. A small increase in the sensitivity of about 2 meV compared to room temperature is observed at 150 °C. The rate of adsorption and desorption of the gas molecules at the surface did not seem to improve much with the increase in temperature meaning Au is fairly reactive to H₂S even at room temperature. The sensor response (i.e., energy and intensity) over time with the gas switch from N₂ to H₂S at different temperatures is shown in the bottom two graphs of Figure 6.

2.5 Ag and Other Metal Functionalization

Some other metal functionalization layers were also tested such as Ag, Pt, Ni, and Pd. The results for two samples with Ag are presented in Figure 7 top, the data for bare GaN surface and Au is also shown for comparison. The base sensor structure is a medium high doped (≈5 × 10¹⁸ cm⁻³) GaN buffer layer, 5 nm QW, 3 nm GaN cap, and 3 nm Ag layer and low-doped (≈1 × 10¹⁸ cm⁻³) GaN buffer layer, 3 nm QW, 3 nm GaN cap, and 3 nm Ag layer. Both samples show sensitivity toward H₂S in the 100 ppb range. For bare GaN, after a longer adjustment period, a small blueshift of about 1 meV is observed for both samples. Metal functionalization increases the sensitivity of the sensor considerably as compared to the bare GaN. With Ag, for the 5 nm thick QW sample, the shift is about 11 meV, while for the 3 nm QW sample, it is about 5 meV. These results are comparable to the Au-functionalized samples where the shifts are about 6 and 10 meV, respectively. The better result for the 5 nm thick QW sample with Ag could perhaps be attributed to the more porous surface leading to a higher surface-to-volume ratio. Figure 7 bottom shows the SEM images of the Ag layer for both samples. The surface for the low-doped sample is much smoother with more surface coverage. Figure 8 shows a comparison between these two samples with bare GaN, then with Au and Ag functionalization layers. Based on these results, a clear winner for H₂S detection between Au and Ag could not be determined as for one sample Au shows higher sensitivity, while for the other, Ag was better. However, over multiple measurements, Au has shown more consistent results while Ag-functionalized samples lose their PL intensity and measurements become difficult over time. With Ag and any metal functionalization layer, the barrier height for the QW seems to reduce leading to a tunneling of electrons to the surface and non-radiatively recombining via surface states. This effect might be very strong with Pt, Ni, and Pd, as after the metal deposition, the PL signal disappeared completely. So no measurement could be performed.

2.6 Ammonia Gas Sensing

Ammonia gas sensing was also successfully performed using the same sensor principle. First, the variable doping series samples with Au functionalization (discussed earlier for H₂S sensing) were used. The results shown in Figure 9 (top) confirm the earlier trend as observed with H₂S (cf. Figure 5). The lowest doped sample produced the highest shift for 500 ppm of NH₃. At a high concentration of 500 ppm, GaN surface without any metal also shows sensitivity toward ammonia. It is known that ammonia molecules dissociate at a Pt catalytic layer,^{[21, 22]} which can also facilitate the dissociation of hydrogen molecules into hydrogen atoms on the surface of the metal film. Some of these hydrogen atoms can then diffuse through the thin film and accumulate at the metal–GaN interface. These hydrogen atoms are then polarized by the internal electrical field, creating a dipole and resulting in a reduction in the barrier height. So when NH₃ gas is introduced, a redshift should be observed. However, with Au-metal catalyst a blueshift is recorded. This implies that after the NH₃ molecule adsorption on the Au layer other reaction intermediates are increasing the barrier height thus increasing the near-surface upward band bending. The bottom graph in Figure 9 shows higher-sensitivity measurements where 5–10 ppm of NH₃ in N₂ could be detected, that is, about 1 meV shift. Comparing this result with hydrogen sulfide, Au is much more sensitive toward hydrogen sulfide than ammonia. Such results open up the possibility of these gas sensors to be used in wider industrial applications.