1. Introduction

Accurate detection of ultraviolet (UV) is in high demand in various applications such as UV index monitors, UV curers, and flame detectors. The global ozone depletion and growth of the ozone hole have enormously increased concerns about the negative biological effects of UV light and consequently the desire to measure the UV index has increased. On a different note, recent fires in several expensive industrial/residential facilities and heritage sites (e.g., the Sungnyemun gate in Korea and the Notre Dame church in France) have raised the importance of detecting fire and alarming in the early stages of fire. Optical fire detectors to instantaneously recognize the radiation of ultraviolet (UV) or infrared light from a flame [1,2] can constitute a promising technology instead of the widespread fire alarm that detects accumulated heat or smoke [3]. Despite these massive demands, conventional Si-based UV sensors are less accurate due to disruption by visible light. Hence, a completely visible-blind UV detector based on wide-bandgap materials such as Ga(Al)N has been actively investigated [4–8]. A large bandgap (3.5∼6.1 eV for Al_xGa_1-xN with 0 ≤ x ≤ 1 [9]) detects only UV light and precludes malfunctions by visible light. As Ga(Al)N has shown reliable operation in high-voltage transistors [10–12] and high-temperature devices [13,14], AlGaN-based UV photosensors can be considered to precisely measure UV light even in harsh environments. However, due to the lack of a lattice-matched substrate, the heteroepitaxy of a Ga(Al)N layer on sapphire substrates can result in defects such as dislocations [15–17] and consequently the detector is not perfectly insensible to visible light. The dielectric stacked UV pass filter is often integrated on the detector [18,19], but the angle dependence tightly limits the detection angle.

In this study, we have demonstrated the Ag-Al₂O₃ Fabry–Perot (F-P) UV pass filter-integrated AlGaN ultraviolet photodiode (PD) showing a large rejection of visible light irrespective of the detection angle. The inherent optical extinction characteristics of Ag enables the design of a fabrication-tolerant Ag-Al₂O₃ F-P filter with a peak transmittance at ∼325 nm. By simply stacking a tri-layer consisting of Ag/Al₂O₃/Ag on the backside of the AlGaN PD, the visible light rejection ratio (responsivity for 315 nm light to responsivity for 405 nm light) was improved by a factor of 10, reaching a value of ∼10⁶ at angles of up to 80 degrees. Despite the sacrifice of UV transmittance, this will reduce the false detection of UV light in AlGaN UV PD, and such detector-based applications can provide accurate results.

2. Materials and methods

The Ag-Al₂O₃ Fabry–Perot UV pass filter integrated visible-blind AlGaN PD is schematically shown in Fig. 1(a). The heterostructure of AlGaN PD shown in the inset was grown by metal organic chemical vapor deposition (MOCVD). We have designed a simple p-i-n structure and a high Al composition of unintentionally doped AlGaN layer was used as a light-absorbing layer to detect deep ultraviolet light. The n-contact layer was formed by a highly Si-doped AlGaN layer while p-GaN was employed for an efficient p-ohmic contact. The incident light through the sapphire substrate is mostly absorbed in the intrinsic layer so that photogenerated excitons are dissociated by the electric field. The electric field distribution according to the applied reverse bias, doping, and thickness were analyzed by the SILVACO and the final thickness of each layer in the heterostructure were suitably tuned.

 

Fig. 1. (a) Schematic diagram of the Fabry–Perot UV pass filter-integrated AlGaN photodiode. The inset shows the heterostructure of AlGaN photodiode grown by metal organic chemical vapor deposition. (b) Cross-sectional scanning electron microscope image of the fabricated device shows a sloped mesa with an angle of 20 degrees. The inset shows the top-view optical microscope image of the fabricated device. (c) Schematic diagram of the measurement setup for various angles of light incidence. (d) Current–voltage characteristics of the fabricated device showing diode-like rectifying characteristics with a turn-on voltage of ∼9 V. The inset shows the identical current-voltage characteristics at log scale.

Download Full Size | PDF

Fabrication was initiated with patterning alignment keys by standard lithography and a metal lift-off process. The circular mesa was then defined by dry etching down to the n-AlGaN layer. The diameter of the mesa was varied at 50, 100, and 150 µm. Prior to etching, photoresist reflow was performed on a hot plate at 155 °C for 10 min to obtain a sloped mesa sidewall [20]. The same dry etch rate of the photoresist and AlGaN results in the same shape of mesa sidewalls in the AlGaN layer. The resultant slope was approximately 20 degrees, as shown in the scanning electron microscope image in Fig. 1(b). The reason for constructing the mesa in a sloped shape was to spread the electric field in the junction plane. In the mesa with a vertical side wall, an electric field can be gathered at the edge. If a high electric field is applied as a reverse bias increases, a current tends to be attracted to a local portion of AlGaN. This leads to a local breakdown and makes the current–voltage characteristic irregular. On the other hand, when the mesa is sloped, the electric field is concentrated favorably at the center [20–22]. After that, patterns for n- and p-electrodes were formed. Before the metal deposition, the sample was immersed in a 1:1 mixture of hydrochloric acid and deionized water for 10 min to remove the native oxide. Ti/Al/Ni/Au (20/120/55/45 nm) were evaporated by an electron-beam (E-beam) evaporator and electrodes were formed by lift off. The metalized sample was annealed at a temperature of 400/700/900 °C for 3/5/5 min in an N₂ atmosphere to obtain an ohmic contact between the n-AlGaN and the metal. After forming the n-electrode, Ni/Au was deposited by 10/10 nm, respectively, and annealed at 550 °C for 10 min to form the p-electrode. The entire device was passivated by 350-nm-thick SiO₂ to protect the device. The fabricated device is shown in the inset of Fig. 1(b). The designed F-P UV pass filter was fabricated on sapphire (the backside of the sample). Ag and Al₂O₃ comprising the designed structure were successively deposited by an E-beam evaporator and atomic layer deposition, respectively.

The optical properties of the Ag-Al₂O₃ F-P pass filter was calculated by the finite difference time domain method to design a low-loss, highly UV-transmitting filter. A plane wave with a Gaussian frequency spectrum was incident normal to the structure, and the optical fields of the reflected and transmitted light were recorded by the monitors positioned behind the source and after the structure, respectively. The experimentally measured refractive indices of Ag and Al₂O₃ were considered in the simulation. A dip in the low extinction coefficient around a wavelength of 325 nm in Ag film was in particular accounted.

For the electrical and optical characterization of the fabricated device, a jig was constructed; a light-emitting diode (LED) was integrated inside. The device was mounted on a probe station connected to a source meter (Keithley 2614B) in a dark box. Using the jig, the LED light source with a specific wavelength could be positioned below the sample so that light could be incident on the back surface of the sample for photoresponse measurements as shown in Fig. 1(c). The commercially available UV LED has very low efficiency and its continuous operation accumulates heat, resulting in unstable UV emission during measurements. Thus, the UV LED was attached to a heat sink (Avaid's TO-5 and TO-39 321527B00000G) (the inset of Fig. 1(c)) and the long-term stability was confirmed before characterization of the fabricated devices.

3. Result and discussion

Figure 1(d) shows the current–voltage characteristics of the fabricated device. It clearly shows diode-like rectifying characteristics with a turn-on voltage of approximately 9 V. The current exponentially increased with a positive bias to the p-GaN with respect to the n-AlGaN layer. The inset represents the same current–voltage characteristics on a log scale. The reverse bias current stayed at around ∼1 nA and rapidly increased when a reverse bias further increased to more than 20 V. This breakdown behavior can be contributed to the Avalanche mechanism wherein the electrons accelerated by an electric field obtain sufficient energy and subsequent collisions with lattices generate additional electron–hole pairs [23].

The Fabry–Perot UV pass filter was implemented by sandwiching an Al₂O₃ dielectric between two lossy Ag mirrors. In this structure, there are two resonant modes wherein the light resonantly transmits through the filter. One is the conventional cavity mode determined by the thickness and refractive index of Al₂O₃. The other one is determined by the wavelength with minimum loss in the Ag mirrors. Ag has the lowest extinction coefficient of approximately 325 nm, and the light off at this wavelength will suffer significant loss when reflected by the Ag mirror. In other words, light can survive inside the cavity only at a wavelength with minimal loss in Ag, and the rest decays fast owing to absorption by the Ag mirror. Figure 2(a) shows the calculated transmission spectra of the filter with different Al₂O₃ thicknesses. The unpolarized incident light was considered by averaging the results with the TE and TM modes. Two resonant modes are clearly shown for the 60-nm-thick Al₂O₃. Whereas, the conventional cavity mode moves to shorter wavelength and eventually disappear as the Al₂O₃ layer becomes thinner. We chose the thickness of Al₂O₃ to be 20 nm to suppress the conventional cavity mode by locating the fundamental cavity mode at very lossy region in Ag. The thicker the Ag mirror is, the sharper is the resonant peak, but simultaneously the lower is the transmittance. The 30-nm-thick Ag mirror was employed to obtain a decent transmittance of approximately 50% and sufficient blocking of other wavelengths. Figure 2(b) shows the calculated peak transmission wavelength for slightly varied Al₂O₃ thicknesses from nominal value; the structure is schematically shown in the inset. Interestingly, it was found that the peak transmission wavelength hardly varied as the thickness of Al₂O₃ changed. This will relax the requirements for the thickness of the thin film when fabricating an optical filter, consequently reducing fabrication costs. The transmittance of the designed UV pass filter was also calculated for different incident light angles, as shown in Fig. 2(c). It allowed only UV light at around 325 nm with a peak transmittance of 45% and blocked all other lights at normal incidence. No higher-order transmittance peak was noted for any other angle, unlike the one in the filter achieved by stacking dielectrics. As the incident angle was varied, the transmittance decreased but the peak transmittance wavelength and linewidth remained almost unchanged, as shown in the inset. This angle-independent transmittance peak wavelength is highly necessary to minimize unwanted light detection in visible-blind UV sensors.

 

Fig. 2. (a) Calculated transmission spectra of the Fabry–Perot UV filter for various Al₂O₃ thicknesses. The unpolarized incident light was considered by averaging the results with the TE and TM modes. (b) Variation in peak transmission wavelength when the Al₂O₃ thickness is slightly changed from 20 nm. The designed filter structure consisting of Ag-Al₂O₃-Ag is schematically shown in the inset. (c) The transmittance of the designed UV filter calculated for different incident light angles. The inset represents variations of peak transmittance wavelength and linewidth of passband.

Download Full Size | PDF

Figures 3(a) and 3(b) show the current–voltage characteristics of the fabricated AlGaN PD before and after the formation of the designed UV pass filter, respectively. The employment of the filter did not change the dark characteristics as expected. The photocurrents were measured under the deep ultraviolet (DUV) at a wavelength of 315 nm (0.43 mW/cm²) and near ultraviolet (NUV) at a wavelength of 405 nm (6 W/cm²). The device clearly exhibited a large photocurrent when illuminated by DUV light regardless of the UV pass filter. The responsivity was calculated using I_ph/P_light, and plotted alongside in the figures, where I_ph is the difference between the photocurrent and dark current, and P_light is the optical power illuminated on the device [24–26]. The difference between the photocurrent and the dark current increased, and consequently the responsivity increased gradually. The responsivities under DUV light were 77.4 mA/W and 34.1 mA/W at the reverse bias of –17 V for the fabricated PD without and with the UV pass filter, respectively. They correspond to external quantum efficiencies of 30.4% and 13.4%, respectively. A partial transmittance in the filter resulted in a reduction of the responsivity after monolithic integration of the filter. It is worth noting that the device without the filter responded to NUV light, as shown in Fig. 3(a). Despite the energy of the incident NUV light being smaller than the bandgap energy of the AlGaN light-absorbing layer, imperfect material properties of the AlGaN resulted in a small but appreciable photocurrent. However, on integrating the UV pass filter with the device, the current–voltage characteristics under NUV light became similar to that measured in the dark, as shown in Fig. 3(b). Filtering unwanted visible light out in the integrated filter allows the device to generate a photocurrent only for the DUV light and to become perfectly visible-blind. At a reverse bias of –17 V, the NUV light responsivity without the UV pass filter was 2.3×10⁻⁶ A/W, but the filter suppressed the responsivity to an almost 20 times lower value (1.1×10⁻⁷ A/W).

 

Fig. 3. Current–voltage characteristics of the fabricated AlGaN photodiode before (a) and after (b) formation of the designed UV filter. The responsivity was also calculated using I_ph/P_light and is plotted alongside in respective figures calculated. I_ph is the difference between the photocurrent and dark current, and P_light is the optical power illuminated on the device.

Download Full Size | PDF

The effect of the integrated UV pass filter was more clearly observed by plotting the visible light rejection ratio (VLRR), defined as the responsivity for DUV light against the responsivity for NUV light (Fig. 4(a)). The response to visible light with a longer wavelength than NUV should be much less than that for NUV light, and thus we conservatively compared the light response of DUV light with respect to the NUV light. The large VLRR near zero bias may be favorable to suppress visible light response, but a photovoltaic operation at zero bias gave rise to a small photocurrent due to partial depletion of the absorbing layer. As the reverse bias increased, the VLRR declined and increased again above a bias of –15 V in both cases. As the NUV light was filtered out, the photocurrent and successive increment by the Avalanche mechanism under DUV light were particularly distinct for the filter-integrated PD. The maximum VLRR achieved here was about ∼7.6×10⁵ in the UV pass filter-integrated one. By integrating the UV pass filter, the VLRR improved by about 5–10 times on average when comparing the VLRR with and without the filter; the maximum enhancement of VLRR was 12.2 times around a bias of –26 V.

 

Fig. 4. (a) Visible light rejection ratio (VLRR), defined as responsivity for DUV light against responsivity for NUV light. The VLRR conservatively compares the light response of DUV light with respect to the NUV light. Monolithic integration of the filter results in 5–10 times the VLRR by filter integration. (b) Angle dependence of VLRR in the UV filter integrated photodiode. The responsivities for DUV and NUV light were characterized at different angles of 0, 30, 45, 60, and 80 degrees. The VLRR according to the applied bias are very similar for the different incident light angles.

Download Full Size | PDF

Angle dependence of VLRR in the UV pass filter-integrated PD was obtained by characterizing responsivities for DUV and NUV light at the angles of 0, 30, 45, 60, and 80 degrees. Figure 4(b) shows that the VLRR for the different applied bias were very similar for different incident light angles and the maximum VLRR of 2.5×10⁶ was attained at an incident light angle of 80 degrees. The angle-independent transmittance peak wavelength of the integrated filter was very dissimilar to that of the conventional UV filter based on a stack of dielectrics. In the conventional filter, a constructive interference of distributed normal reflections at interfaces of dielectrics determines the transmittance peak wavelength; therefore, an extra in-plane momentum of obliquely incident light led to a resonant transmittance peak energy (wavelength) increase (decrease) [27]. However, the Ag-Al₂O₃ Fabry–Perot-based filter utilizes an optical loss in Ag, and the transmittance peak wavelength becomes less sensitive to the incident light angle. This contributes to sustaining the VLRR in the UV PD for higher incident light angles, which would be indispensable in widespread UV monitoring applications.

4. Summary

In summary, we have fabricated an AlGaN ultraviolet PD with a monolithically integrated Ag-Al₂O₃ Fabry–Perot UV pass filter, and shown improved visible-blindness by comparing responsivities under deep UV and near UV light. By utilizing the inherent dispersion of optical loss in Ag, a fabrication-tolerant and low-cost UV pass filter was designed with 30-nm-thick Al₂O₃ sandwiched by two 20-nm-thick Ag layers. Unlike a conventional dielectric stacked filter, no high-order transmittance peak at all angles was confirmed by the finite difference time domain method. The filter-integrated AlGaN UV PD exhibits identical current–voltage characteristics in dark, while a higher contrast of responsivity when illuminated by 315 nm DUV and 405 nm NUV light. Moreover, such improved visible light rejection ratio was maintained at angles of up to 80 degrees. A largest visible light rejection ratio of 7.6×10⁵ at normal incidence and 2.5×10⁶ at 80 degrees was attained. This will reduce the malfunction of AlGaN-based UV-detecting sensors and promise more reliable and accurate measurement of UV light in any circumstance.