1. AlGaN/GaN Substrates with Different Al Compositions

In order to illustrate selective annealing effects under different Al compositions, we fabricated AlGaN/GaN substrates with Al compositions of 33\% and 39\%. Epitaxial AlGaN/GaN layers and GaN buffer layers were grown by metal-organic chemical vapor deposition (MOCVD) on (0001) sapphire substrates. A 3.5 ${\mathrm{\mu}}$m GaN layer was grown using trimethylgallium (TMGa) and ammonia (NH$_{3}$) as sources, and a bulk 300 nm AlGaN layer was grown using trimethylaluminum (TMAl) as a source.

Table 1 summarizes the important parameters of two epitaxial samples used in this work. Fig. 1 shows PL spectra for each sample at room temperature; the PL peaks of the AlGaN samples were located at 288 nm and 298 nm. The bandgap energy for each Al sample composition was calculated using Eq. (1) ⁽¹³⁾.

2. Fabrication process flow

Fig. 2 shows schematic cross-sectional and three-dimensional views of the proposed asymmetric AlGaN/ GaN MSM UV sensor. The first fabrication step is an initial cleaning using acetone, methyl alcohol, SPM (H$_{2}$SO$_{4}$:H$_{2}$O$_{2}$, 3:1), and SC2 (HCl:H$_{2}$O, 1:1) to remove contaminants from the wafer surface. After initial cleaning, electrodes were patterned using photolithography. Next, Ni (30 nm) and Ti/Al (30 nm/50 nm) were deposited using e-beam evaporation and lift-off processes. An 30 nm thick insulating layer of Al$_{2}$O$_{3}$ was

layered using atomic layer deposition (ALD), and the insulator was selectively removed by dry etching, using plasma to create contact holes. Additionally, plasma ashing was carried out to remove the remaining organic materials, including the negative photoresist. Finally, Ni/Au (30 nm/50 nm) was deposited using e-beam evaporation and lift-off processes.

For selective local annealing, the oxide is layered between the metal electrodes 1 and 2, as shown in Fig. 2(a). When the bias between electrodes 1 and 2 is high enough, a high current may flow through the broken-down insulator and generate heat, which provides an annealing effect.

Table 2 shows the selective annealing conditions, which were achieved using a probe station with an Agilent HP4156C parameter analyzer. In order to optimize the selective annealing conditions, the bias voltage was incremented in steps from 5 V to 50 V to observe the occurrence of local breakdown. The breakdown of an insulator did not occur between 20 V and 30 V, but from 40 V to 50 V, a strong breakdown effect was found. Consequently, a bias voltage was applied 51 times from 39 V to 40 V at intervals of 0.02 V to achieve continuous annealing. The maximum current was set to 50 mA.

1. I-V Characteristics

Fig. 3(a) shows the I-V characteristics of Sample 1, which had an Al composition of 39%. The dark current density at the forward bias before selective annealing was higher than that at the reverse bias, because work function of Ni is larger than that of Ti.

At a forward bias of 1 V, the dark current density was 8 × 10^-4 A/cm² before selective annealing and 0.2 A/cm² after selective annealing, representing a 250-fold increase. As a result of selective annealing, the rectifying contact of the Ti/Al/Ni/Au electrode became a near-ohmic

contact at forward bias, due to the formation of nitrogen vacancies. During annealing, nitrogen easily reacts with Ti to form TiN, and the remaining nitrogen vacancies act as donors on the AlGaN substrate, leading to high electron concentration. As a result of this high electron concentration, the Schottky barrier of the Ti/Al/Ni/Au electrode becomes thinner, changing the contact behavior.

In contrast, at reverse bias, differences in electrical characteristics before and after selective annealing were not clearly observed due to the high leakage current. The dark current density after selective annealing was 2.3 × 10^-6 A/cm² at -0.6 V, representing only a slight decrease as compared with the value of 6.4 × 10^-6 A/cm² before selective annealing.

The observed high leakage currents are not caused only by bulk defects and poor interface quality; they are also due to the surface leakage currents caused by GaO$_{2}$ on the AlGaN surface. On the AlGaN surface, Ga reacts with oxygen to form a thin layer of Ga$_{2}$O$_{3}$ and GaO$_{2}$, and the negatively charged chemical bonds of GaO$_{2}$ act as surface traps, leading to operational instability and performance degradation. After selective annealing, the dark current density was slightly lower than that before selective annealing. This decrease can be attributed to reduction of GaO$_{2}$, which is located on the AlGaN surface due to annealing.

In addition, the Ni/AlGaN interface as well as the Ti/AlGaN interface was expected to have annealing effects. Although the electrodes were isolated, the microscope image confirmed that the large breakdown voltage affected not only the Area 1 and the Area 2 but also a wide range of interdigitated fingers and surrounding insulators. It was expected that the annealing would affect the traps at the Ni/AlGaN interface. Under equilibrium conditions, most of the interface traps are filled with electrons below the Fermi level, and under the application of a reverse bias at the Ni/Au electrode, electrons captured by the interface traps may be emitted by a trap-assisted tunneling, which would contribute to leakage current. Interface traps at the Ni/AlGaN interface were thought to be passivated by the selective annealing, leading to a reduction in the dark current density.

Fig. 3(b) shows the photoresponse characteristics under 365 nm, 288 nm and dark conditions. Due to the high leakage current, there was no significant UV photoresponse at high voltages, which was attributed to poor epitaxial layer quality. In other words, the high leakage current of the proposed UV sensor indicates that the epitaxial layer quality was not optimal, due to incomplete epitaxial growth.

Fig. 4 shows the I-V and photoresponse characteristics of Sample 2, which had an Al composition of 33%, before and after selective annealing. The decrease in Al composition between Samples 1 and 2 substantially affected the epitaxial layer quality. Fig. 4(a) shows that a lower dark current density was measured in Sample 2 than in Sample 1. This occurs because lower Al compositions correspond to fewer defects and interface traps caused by lattice mismatch between Al and Ga atoms.

At a forward bias of 1 V, the dark current density was 9.1 × 10^-6 A/cm² before selective annealing, and significantly increased to 6.0 × 10^-2 A/cm² after selective

annealing. In comparison, at a reverse bias of -0.2 V, the dark current density was 1.9 × 10^-8 A/cm² before selective annealing, and decreased to 7.7 × 10^-10 A/cm² after selective annealing. Consequently, Sample 2 shows better ohmic behavior than Sample 1. Fig. 4(b) shows the photoresponse characteristics of Sample 2 after selective annealing, which are improved as compared with Sample 1, due to the lower dark current density.

2. Spectral Responsivity

We measured the spectral responsivities of Samples 1 and 2 before and after selective annealing. As these samples had high dark currents, we reverse biased them between 0 and -1 V, where optical characteristics were clearly distinguishable. Fig. 5 shows the spectral responsivity of Sample 1. Under a forward bias and after

selective annealing, photo-responsivity was not observed because the electrons supplied from the Ti/Al/Ni/Au electrode easily overcame the lower Schottky barrier height. Under a reverse bias, photoelectrons were the dominant charge carriers and the cut-off wavelength of 288 nm was attributed to the high Schottky barrier height of the Ni/Au electrode.

Before selective annealing, the measured UVRRs were low, with values of 44 and 11 at -0.3 V and -0.5 V, respectively. On the other hand, after selective annealing, the UVRRs increased to as high as 84 and 45 at -0.3 V and -0.5 V, respectively.

Fig. 6 shows the spectral responsivity of Sample 2 under various reverse bias conditions. The cut-off wavelength of Sample 2 was 298 nm. In Fig. 6(a), fluctuations in the curves were caused by interface traps, and high responsivity in the visible region was obtained.

As shown in Fig. 6(b), after selective annealing, the responsivity showed reduced fluctuations and the UVRR improved. The UVRR was 19 at -1.0 V before selective annealing and increased 7-fold, to 135, after selective annealing.

Photoresponse characteristics of the samples improved overall after selective annealing. The improved UVRRs after selective annealing via the Ti/Al/Ni/Au electrode were attributed to reductions in the leakage current, which were due to surface passivation and reduction of the number of traps at the Ni/AlGaN interface. When annealing via the Ti/Al/Ni/Au electrode, the Ni/AlGaN interface is also affected by Joule heat, decreasing the number of interface traps and thereby reducing the dark current density.

3. Transmission Electron Microscopy (TEM) Analysis

To characterize the effects of selective annealing on the interfaces and the surface of the device, we performed TEM EDS analysis. EDS is a device capable of qualitative analysis of samples. After the electron beam collides with the sample, internal electrons in the sample are emitted to the outside by incident electrons, and a vacancy is created and become unstable. The electrons in the upper orbit are shifted to the vacancy to be stabilized, and X-ray energy corresponding to the difference between the two energy orbits is emitted. Since X-ray energy has different values for each atom, the content of elements can be measured by counting X-rays.

Depicted in Fig. 2(a), Area 1 corresponds to an area where annealing effects were weak, and Area 2 corresponds to an area where annealing effects were strong, as dielectric breakdown occurred between Electrodes 1 and 2 of the MIM structure.

The atomic composition was characterized in the depth direction. Fig. 7 shows the atomic content as a function of position in the weakly annealed Area 1 of Sample 2, while Table 3 gives the average values of each element present in each sample region. Fig. 7 and Table 3 illustrate that the atomic content was in line with expectations for the respective layers, except by penetration of atoms such as N and O into the Ti layer.

Fig. 8 and Table 4 show the atomic composition as a function of position in the strongly annealed Area 2 of Sample 2 and the average values of each element present

at each sample region, respectively. Compared to Area 1, Area 2 has unclear boundaries between each layer especially at the interfaces of the Ti layer. The insulating layer of Al$_{2}$O$_{3}$ was not clearly discernable. Also, Ti atoms were present within the Al layer, and N content in the Ti layer was significantly greater than in the weakly annealed Area 1. In addition, the strong annealing effectively formed a new layer, which was clearly observable as an Al hump and a Ga peak in Fig. 8 and Table 4.

The strongly annealed area showed interdiffusion between layers, and the N content of the Ti layer of Area 2 was higher than that of Area 1. It was thought that the local breakdown strongly contributed to modifying the rectifying contact into ohmic contact; the N content in the Ti layer was increased with the annealing strength. In other words, selective annealing provided sufficient energy to break the Ga-N bond, leading to N vacancies. The formation of a Ga layer demonstrates that selective annealing cause diffusion of atoms within the semiconductor, changing the electrical characteristics. But, the changes in contact behavior due to Ga diffusion warrant further study. As seen in Fig. 8, the O content increased in all layers, except the insulating layer of Al$_{2}$O$_{3}$ compared to Area 1. The increase indicates not only that O in the air penetrated into the device during selective annealing, but also that O in the device diffused into other layers. It is expected that dielectric breakdown in an N$_{2}$ atmosphere would reduce the O content.