Breakdown-induced conductive channel for III-nitride light-emitting devices

III-nitride semiconductor-based light-emitting diodes (LEDs) have superior physical properties, such as high thermal stability and brightness, for application to solid-state lighting sources. With the commercialization of GaN-based LEDs, improving LED reliability is important because they can be affected by electrostatic discharge, reverse leakage, and breakdown. However, research on the reverse bias characteristics of GaN-based LEDs is insufficient. We studied the reverse breakdown mechanism and demonstrated that a local breakdown can form a conductive channel in GaN-based LEDs, which can be expanded to a novel planar-type LED structure without an n-contact electrode. Furthermore, we found that this approach can be applied to AC-controllable light-emitting devices without any AC–DC converter.


Full breakdown and local breakdown in the III-nitride LEDs
,b show the current vs. voltage (I-V) and current vs. light output power (I-L) curves, respectively, of a conventional n-p LED under different breakdown conditions. The typical electrical behaviour of the n-p LED has been observed from I-V measurements to consist of a turn-on voltage (~2.54 V) and very low reverse leakage current (I r = 0.45 nA at −5.0 V). After forming the breakdown at a reverse voltage (~100 V) much higher than the breakdown voltage of ~60 V, the breakdown LED indicates linear I-V characteristics and no emission properties. This is called the full breakdown n*-p* LED, where the asterisk represents the reverse breakdown. This is the very general breakdown phenomena in GaN-based LEDs 30,36,37 . However, after a local breakdown (p*) is formed in the only p-layer under the reverse breakdown voltage of ~60 V, the I-V curve of the n-p* LED represents almost linear properties-like a conductive material-below +3.4 V and then the normal I-V characteristics of an n-p LED above 3.4 V. As a result, the electrical properties of the n-p* LED were modified from an n-p diode to an n-n-or n-i-like structure in the low-voltage region. This led to almost linear I-V properties like a full breakdown n*-p* LED below a turn-on voltage of 3.4 V. Based on these results, we believe that the limited linear I-V characteristics of the n-p* LED can be attributed to the formation of a BCC in the local breakdown p*-layer, as shown in Fig. S1b. However, as the forward voltage was increased above 3.4 V, the I-V properties of the n-p* LED represented the same I-V characteristics as an n-p LED. We surmised that the injection current through the BCC became saturated within the limited BCC around 3.4 V (Fig. S1c) and overflowed to other non-breakdown regions above 3.4 V (Fig. S1d). This means that the I-V characteristics of the n-p* LED recovers from the breakdown n*-p* to an n-p LED above a critical voltage (>3.4 V) because the BCC can bypass the current in the n-p* LED like a parallel resistance (R p ), as shown by the inset of Fig. 1a.
The electroluminescence (EL) of an n-p* LED would not be observed at a low injection current (<40 mA), which was the saturation point of the carrier flow through the BCC and the starting position of the carrier overflow from the BCC to non-breakdown regions at ~3.4 V. This led to radiative recombination from the active layer, , I-L curves of conventional n-p, n-p*, and n*-p* LEDs. The insets of (a) show a schematic of the n-p* LED structure with BCC as a parallel resistance and the log I-V curve of n-p LED, and the inset of (b) shows the V-L curves of the n-p and n-p* LEDs. as shown in Fig. S1d. During carrier injection through the BCC (<3.4 V), radiative electron-hole recombination did not happen at the region of InGaN/GaN quantum wells (QWs), which is consistent with the large forward leakage current below 3.4 V in the n-p* LED. This indicates that the BCC acting as an R p in the current bypass was formed at the p-type layer and InGaN active region. However, the EL intensity of the n-p* LED exhibited almost the same emission intensity behaviour as that of the n-p LED at different applied forward voltages, as shown by the inset of Fig. 1b. This indicates that the applied voltage can control the carrier flow and emission performance of nitride-based n-p* LEDs. Based on these results, we suggest that the reverse breakdown phenomenon forms a localized BCC in nitride-based LEDs, which can play a role in a limited current-leakage path under a low forward bias (<3.4 V).
Formation of BCC and giant dot-like luminescence from surface V-defect Figure 2a,b show the as-fabricated LED image and its reverse dot-like luminescence under the reverse voltage of 25 V, respectively. There are two kinds of dot-like luminescence: the general small dot-like luminescence (SDL) at position B and the giant dot-like luminescence (GDL) at position A. As the reverse breakdown voltage was increased to ~60 V, we observed the local breakdown phenomena to be a BCC at position A, as shown in Fig. 2d. Based on these results, we believed that the GDL region induced by the reverse bias can form a local breakdown to be form a BCC in an n-p LED. In addition, we initially could not clearly observe any surface defect at position A using an optical microscope because of its very small size (<3.0 µm), as shown in Fig. 2a. However, after we measured the GDL region, we accidentally found a surface V-defect at position A, as shown in Fig. 2c. It implied that the GDL is generated by the surface V-defect region as the reverse bias is increased, as shown in Fig. S2. Therefore, we suggest that the surface V-defect of a GaN-based LED can generate a GDL at a reverse bias and that the BCC can be formed at surface V-defect as the reverse bias reaches the breakdown voltage. Figure 2e,f show schematic diagrams of the reverse tunnelling-assisted transport leakage path and the emitting transport paths of the conventional SDL and GDL in the InGaN QW region, respectively. We observed that the GDL among the reverse dot-like luminescences originated from the surface V-defect, which is the main source of the reverse breakdown phenomenon for forming a BCC. In general, defect-related leakage carriers are injected to the QW through preferential paths with a lower potential barrier, which leads to the SDL 38 . The surface V-defect that formed at the threading dislocation represents a small valley edge region with a higher In content than the sidewall regions like a quantum dot 39 ; this leads to the GDL transiting from the deeper localized state of the V-defect to the lowest energy level of the valence band. Because the deeper localized state formed in the valley end of the surface V-defect can easily be filled by the injection carrier, we suggest that this deep level transition first emits at the valley end, and the emission of the V-defect expands from the valley end to the sidewalls with six facets of {10-11}. This results in the GDL, as shown in Fig. S2.

Relationship between reverse leakage current and BCC
In order to understand the local breakdown mechanism to form a BCC, we measured the temperature-dependent breakdown voltage of the GaN-based LED. In general, the breakdown voltage V B at a temperature T is given by the relation where V B0 is the breakdown voltage at room temperature, T 0 is the room temperature, and β is the temperature coefficient of the breakdown voltage 31,32 . Figure 3a shows that the temperature coefficient (β) of the breakdown voltage had a negative value as the measured temperature increased. It indicates that the BCC formation mechanism is a Zener-type breakdown phenomenon 31,32 . In particular, because the BCC can be generated at a surface V-defect, we suggest that the BCC is formed by the defect-assisted Zener breakdown mechanism. We also obtained the thermal activation energy (E a = 18.6 meV) to form a BCC at the temperature-dependent breakdown voltage of GaN-based LEDs (Fig. 3a), which is lower than the thermal energy (kT = 25.9 meV) at room temperature. This implies that the reverse-bias-induced Zener breakdown can occur more easily than the escape of a thermally assisted carrier. Figure 3b shows the activation energies of the reverse leakage current and breakdown voltage in the GaN-based LED as a function of the reverse voltage. Two slopes of activation energies were found for the reverse voltage, which indicates two reverse leakage mechanisms: (1) a low reverse bias region below 30 V and (2) high reverse bias region above 30 V. The activation energy of the temperature-dependent reverse leakage current for the carriers trapped in deep centres to escape can be represented by an Arrhenius plot (Eq. 2), which can be expressed as follows 29,40 : where E a is the thermal activation energy, k is the Boltzmann constant, Φ PF is the barrier height of the carrier trapped in the deep centre without an external electric field, β PF is the Poole-Frenkel constant, and F is the local electrical strength applied to the deep centres shown in inset of Fig. 3c. This indicates that the activation energy of the reverse leakage current decreases with an increasing reverse bias because of the increase in the reverse bias-induced electric field strength (F) at deep levels. Figure 3c shows  leakage mechanism. In particular, the reverse leakage current is reported to be strongly related to the space-charge limited current above the region of the Poole-Frenkel emission 40,41 . Moreover, the V R2 (~42 V) at which the fitting curve contacted the thermal energy (kT = 25.9 meV) is almost consistent with the soft breakdown. Based on these results, we suggest that a BCC can be formed by the defect-assisted Zener breakdown, which can be generated from a soft breakdown related to the space-charge limited current after the Poole-Frenkel leakage model is closely followed at the low-voltage region 1.
p-n-p*LEDv without n-type electrode Figure 4a depicts the schematic of a p 1 -n-p 2 LEDv consisting of two n-p LEDs with n-type (n 1 and n 2 ) and p-type electrodes (p 1 and p 2 ). This p 1 -n-p 2 LEDv was systematically evaluated by comparing with the n 1 -p 1 and n 2 -p 2 LEDs because they share the same epi-structure and chip design. In an n-p LED, the n-type (n 1 ) and p-type (p 1 ) electrodes are injected with cathode and anode currents, respectively. However, the p 1 *and p 2 -electrodes of the p 1 *-n-p 2 LEDv are introduced to the cathode and anode currents, respectively, without the use of an n-type electrode as the cathode, as shown in Fig. 4b. In terms of the carrier path, therefore, the conventional LED is an n 1 -p 1 or n 2 -p 2 structure, whereas the BCC-embedded LEDv is a p 1 -n-p 2 structure whose p-electrodes are separated by the etched n-type region to form the isolation between two n 1 -p 1 and n 2 -p 2 LEDs. Figure 4c shows the I-V characteristics of an n-p LED, the p 1 -n-p 2 and p 1 *-n-p 2 LEDv structures. The reverse breakdown of the n-p LED with a turn-on voltage of 2.94 V did not occur until −20 V, which indicates the excellent crystal properties. On the other hand, when the anode and cathode currents were applied to the p 1 -and p 2 -electrodes, respectively, in the p 1 -n-p 2 LEDv without a BCC, there were no forward and reverse currents as we expected. However, the I-V curve of the p 1 *-n-p 2 LEDv with the BCC in the p 1 region showed characteristics similar to the I-V curve of the n-p LED because the injection current could be transferred to the p 2 region through the BCC in the p 1 * region. In addition, the reverse leakage currents of n-p LED and p 1 *-n-p 2 LEDv are 0.45 nA and 1.34 nA at the reverse bias of − 5.0 V, respectively. It indicates that the leakage current of p 1 *-n-p 2 is not significantly affected by BCC formation. However, above the turn-on voltage, the series resistance of the p 1 *-n-p 2 LEDv was higher than that of the n-p LED owing to the increase in carrier paths through the p*-region and n-type region between the p 1 * and p 2 electrodes. Figure 4d shows the light output power of the n-p LED, the p 1 -n-p 2 and p 1 *-n-p 2 LEDv's as a function of the injection current. The EL emission of the n-p LED exhibited a significant increase above the turn-on bias of 2.94 V, whereas that of the p 1 -n-p 2 LEDv was not observed within our measurement range. However, the p 1 *-n-p 2 LEDv showed I-L behavior similar to that of the n-p LED. From these results, we achieved that the external quantum efficiencies of conventional n-p and p 1 *-n-p 2 LEDs were 49.5% and 48.8% at the injection current of 20 mA, respectively, as shown in inset of Fig. 4d. It indicates that the p 1 *-layer plays a role in the supplemental path of electrons as a parallel resistance under cathode-injection conditions, as shown in the inset of Fig. 4a. For these reasons, we believe that the injected electrons move from the p 1 *-layer to the p 2 -layer through the n-layer in the p 1 *-n-p 2 LEDv. This leads to similar L-I-V characteristics to an n-p LED like a low turn-on voltage and strong EL emissions from the p 2 region of the p 1 *-n-p 2 LEDv. Figure 5a shows the I-V curves of the p 1 *-n-p 2 , p 1 -n-p 2 *, and p 1 *-n-p 2 * LEDv's, which were obtained by using only p-type electrodes as the anode and cathode currents without the n-type electrode. In addition, we measured the L-I curves of the p 1 *-n-p 2 , p 1 -n-p 2 *, and p 1 *-n-p 2 * LEDv's by changing the current direction from the p*-layers, as shown in Fig. 5b. Before forming the BCCs in the p-layers, we did not observe the turn-on voltage and emission at applied voltages of -20 to 20 V because of a huge Schottky barrier for negative bias and the formation of depletion regions between the p 1 -n and n-p 2 junctions in the p 1 -n-p 2 LEDv. However, after BCCs were formed in the p 1 -n-p 2 LEDv by using the current-induced breakdown methods shown in Fig. S3a, we found that the p 1 *-n-p 2 and p 1 -n-p 2 * LEDv's exhibited I-V curves similar to that of a typical p-n LED consisting of a turn-on voltage and series resistance. In the p 1 -n-p 2 * LEDv, when we applied the anode and cathode currents to the p 1 and p 2 * electrodes (Fig. S3c), respectively, the turn-on voltage was ~2.75 V, and the reverse breakdown was not observed up to -20 V. Similarly, when we applied the cathode and anode currents to the p 1 * and p 2 electrodes in the p 1 *-n-p 2 LEDv, the turn-on voltage was ~2.67 V, and the reverse breakdown was not observed up to +20 V. We clearly observed the EL emission from p-layer of the p*-n-p LEDv's by applying the anode and cathode currents to p-and p*-layers, respectively. Because the p 1 *-n-p 2 and p 1 -n-p 2 * LEDv's operated as n-n-p 2 and p 1 -n-n LEDv's, respectively, both EL emissions were only observed from the p 2 and p 1 regions, respectively, as shown in the inset of Fig. 5b. However, when the BCCs were formed in both p-layers (Fig. S3d,e), there was no turn-on voltage for the p 1 *-n-p 2 * LEDv. In the applied voltage range of -6.0 to +5.9 V, the I-V curve showed linear resistance properties similar to those of the resistance of a conductive material. However, as the positive bias (>|6.0| V) to the p 1 * and p 2 * electrodes increased, the p 1 * and p 2 * layers recovered to the original properties of p 1 and p 2 layers as the source of positive carriers because of the saturation of the current flow through the limited BCCs. In particular, the p 1 *-n-p 2 * LEDv exhibited both side emissions of the p 1 * and p 2 * regions at the injection currents of >42 mA and <-50 mA, respectively. However, there was no EL emission between -50 and 42 mA, which is consistent with the linear region (−6.0 to +5.9 V) of the I-V curve. These results indicate that the EL emission of the p 1 *-n-p 2 * LEDv is alternatively generated from both the p 1 * and p 2 * layers by the directions of the injection bias. As a result, we observed the EL emissions of the p 1 *-n-p 2 * LEDv under the pulse conditions (pulse width of 1 s and 50% duty cycle) of ±100 mA operating current. As the p 1 *-n-p 2 * LEDv was injected with the pulsed current, the EL emission was observed similar to the pulse injection condition, as shown in Fig. 5c. This indicates that the p 1 *-n-p 2 * LEDv can be consistently operated by an alternating bias for continuous EL emissions under the pulsed injection conditions. However, the positive-bias-induced EL intensity of the p 1 *-n-p 2 * LEDv was slightly higher than the reverse-bias-induced EL intensity shown in blue line of Fig. 5c. This may be due to the difference in the BCCs of p 1 * and p 2 *, such as the size and position. This is consistent with the slightly asymmetric breakdown-induced voltage-drop phenomenon and I-L characteristics (blue line) shown in Figs S3a and 5b, respectively. In addition, we observed the AC performance of the p 1 *-n-p 2 * LEDv as the AC frequency was increased. Figure 5d show the EL intensity of the p 1 *-n-p 2 * LEDv under different AC conditions (the peak voltage, V p = 7.0 V) from 1 Hz to 1 kHz as a function of time. The EL intensity of the p 1 *-n-p 2 * LEDv matched the operating voltage of the AC frequencies well. Despite the AC frequency being increased to 1000 Hz, the EL emission of the p 1 *-n-p 2 * LEDv operated well, as shown in inset of Fig. 5d. Furthermore, we clearly observed that the light was alternately emitted from p 1 * and p 2 * regions at 1.0 and 1000 Hz AC frequencies shown in Fig. S4. This indicates that the p 1 *-n-p 2 * LEDv without an n-electrode can be used for AC lighting sources without an AC-DC converter.

Conclusions
We studied the local breakdown phenomenon and its applications for an n-p and p-n-p GaN-based LEDv's by applying the critical bias to the p-type layers. When a reverse bias was applied to the n-p LED, it found that BCCs were formed in the surface V-defect to transport carriers in the breakdown region (p*) of the n-p LED. The temperature-dependent breakdown voltages of the n-p LEDs showed that the BCC formation mechanism was the defect-assisted Zener breakdown phenomenon. In addition, the p 1 *-n-p 2 * LEDv exhibited both side emissions of the p 1 * and p 2 * regions at the alternative injection currents. Based on the results, we suggest that this approach promises a novel n-type electrode free p-n-p* LEDv and an AC-controllable p*-n-p* LEDv without an AC-DC converter, leading to the extension to additional new applications.

Methods
We prepared a 2.0-µm-thick (0001) GaN template grown on a c-plane sapphire substrate by using a conventional two-step growth method employing metal-organic chemical vapor deposition (MOCVD) developed in-house 2 . Trimethylgallium (TMGa), trimethylindium (TMIn), and ammonia (NH 3 ) were employed as the Ga, In, and N sources, respectively. Silane (SiH 4 ) and biscyclopentadienylmagnesium (Cp 2 Mg) were used as the n-and p-type dopants, respectively. After growing the c-plane GaN template, we grew a conventional n-p heterojunction LED structure consisting of a 3-µm-thick Si-doped n-type GaN layer (n e = 2.0 × 10 18 /cm 3 ), five-period InGaN/GaN quantum wells (QWs), and a 0.1-µm-thick Mg-doped p-type GaN layer (n h = 1.0 × 10 18 /cm 3 ). After the LED wafers were grown, two types of LED chips were produced in a standard fabrication process: a conventional n-p LED with lateral electrodes and a novel p-n-p LEDv with two p-electrodes separated by a mesa structure consisting two conventional lateral-electrode-type LEDs (see Fig. 4a). In the structures of the conventional n-p LED and novel p-n-p LEDv, the n-type and p-type metals were Ti/Al and Ni/Au, respectively, and deposited by an electron-beam evaporator. Before depositing the p-type metal, we deposited a 100-nm-thick indium-tin oxide layer on p-GaN as a transparent conductive electrode to increase the current-spreading effect. To measure the local breakdown phenomenon in the p-layer of the n-p LED and p-n-p LEDv, we evaluated the voltage and current of both the LED and LEDv by using a Keithley 2400 source meter. The L-I-V measurements were performed with an HP-4155 parameter analyser (Hewlett-Packard, now Agilent Technologies). In particular, we observed SDL and GDL of the n-p LED using a high-magnification objective lens under reverse bias. The surface V-defect of GaN-based LED was analysed by a scanning electron microscopy.