Tunnel junctions (TJs) are envisaged as potential solutions to improve the electrical injection efficiency of nitride emitters in the visible as well as in the UV range. Indeed TJs would solve the issues related to the poor contact with the top p type nitride layer, replacing it by an n type one. But if metal-organic chemical vapor deposition (MOCVD) is chosen to grow the n side of the TJ on a LED, one faces the problem of a potential re-passivation by hydrogen of the underlying p type layer. We propose a TJ epitaxial process whereby low growth temperatures, high growth rates and the type of carrier gas will minimize hydrogen incorporation in the underlying layers. In this view, n++/p++ GaN TJs with and without an (Ga,In)N intermediate layer are grown by MOCVD at varying temperatures (800°C and 1080°C), using N2 as a carrier gas under a very high growth rate of 2.5μm/h on top of blue (Ga,In)N/GaN LEDs. The LEDs made under N2 carrier gas and lower temperature growth conditions are operational without the need for further thermal activation of the Mg acceptors. The light emission intensity from the top surface of the TJ-LEDs is improved compared to the reference LED without TJ: besides the more efficient carrier injection this is also attributable to the larger photon extraction efficiency because of the rough surface of the low temperature grown n-GaN contact layer of the TJ-LEDs.
Gallium nitride-based devices have been developed for the last decades to become one of the most important technologies in electronics and optoelectronics, being able to produce in particular efficient laser diodes and light-emitting-diodes (LEDs).1 The much higher resistivity (at least a factor 100 larger) of p-type GaN layers compared to n-type layers imposes specific metallic contact geometries: to have a good current spreading in the LED it is mandatory that the quasi totality of the p-type layer surface is covered by a metal (acting as a reflector) for bottom emission or a transparent conductive electrode such as indium tin oxide (ITO) for top emission.2,3 For standard LEDs, these solutions are perfectly acceptable as shown by the record performance achieved for blue LEDs4 – but for other applications such as vertical cavity surface emitting lasers, which require intracavity metallic contacts,5 the limited current spreading in the p-GaN layer induces a serious limitation to the device performance. Adding a p++/n++ tunnel junction (TJ) on top of this kind of devices can be of great help regarding this issue.6,7 Indeed, using a TJ offers the possibility to take the contact on an n-type layer instead of a p-type layer and, thanks to the much lower resistivity of the n-type layer, the current spreading is largely improved. Lately, polarization-enhanced tunnel junctions have been developed, profiting from a high electric field caused by an (Ga,In)N intermediate layer that results in a large band bending and, consequently, an increase in tunneling probability.8–10 Additionally it has been shown that the efficiency droop could be reduced by using structures including several active regions linked by tunnel junction.11 Such stacked LEDs would of course be much easier to grow if a single growth method is used, as discussed later.
TJ-enhanced devices have already been grown by molecular beam epitaxy (MBE),8,12 metalorganic chemical vapor deposition (MOCVD)13,14 and by a hybrid approach15–17 which consists of growing the p++ GaN of the TJ by MOCVD and the top n++ GaN by MBE. Since MOCVD has the industry preference for LED production, developing efficient TJs by this technique would be directly applicable in the industry. However, p-GaN layers grown by this method have their Mg acceptors passivated by the hydrogen present in the chamber, forming Mg-H complexes14,17 and a post-growth annealing under N2 is then required to break these complexes and activate the acceptors.18 For TJ-enhanced LEDs, the p++ GaN is buried under an n++ GaN layer, and because the diffusion of H in highly doped n layers is much slower than in p layers this activation is less efficient, rendering the escape of H from the structure difficult.13,19 A thermal treatment to outgas the H laterally after etching the mesa structures has been proposed13,20 showing interesting results but still requiring further optimization. Alternatively, one group has recently shown that the re-passivation of Mg acceptors can be minimized using a low MOCVD growth temperature,21 which they have applied to the realization of TJs on top of commercial blue LEDs.
In this work we investigate the effect of the TJ growth temperature during the MOCVD growth of the top n++ part of TJs using an alternative high growth rate and N2 vector gas, since these parameters can reduce the Mg re-passivation by minimizing the thermal budget above the p-type GaN. The chosen growth conditions promote a roughening of the layers which can be beneficial for the light extraction efficiency for top surface emission. The p++ GaN/n++ GaN and p++ GaN/InGaN/n++ GaN TJs were grown on top of standard blue LEDs and had their morphology and electro-optics characteristics compared.
All the different tunnel junctions (Fig. 1) were grown on top of blue-LED structures previously grown on sapphire in a 7x2 in. Aixtron MOCVD reactor with a showerhead geometry. The blue LED structures used in this work (S-LED) are all issued from the same growth run in order to have a direct and reliable comparison between the different samples. Trimethylgallium (TMGa), triethylgallium (TEGa), trimethylindium (TMIn), trimethylaluminium (TMAl), bis(cyclopentadienyl)magnesium (Cp2Mg), silane (SiH4) and ammonia (NH3) were used as the precursors of Ga, In, Al, Mg, Si, and N, respectively. The growth temperatures were measured by pyrometry with a reflectivity corrected system from Laytec. The first part of the structure consisted of a 2 μm non-intentionally-doped GaN template, followed by 2 μm n-doped GaN ([Si]= 6×1018 cm-3) and 5 [In0.15Ga0.85N (2nm)/GaN(12nm)] quantum wells, which were followed by a 20 nm Al0.15Ga0.85N electron-blocking-layer (EBL), 100nm p-GaN and a 10nm p++ cap of GaN ([Mg]= 1×1020 cm-3). At this stage the p-type doping was activated by an in situ annealing under N2 atmosphere during 20 min. at a temperature of 700°C. Then the TJs were grown again by MOCVD at 800°C, consisting of a 20 nm thick n++ GaN doped layer ([Si]= 1×1020 cm-3) followed by a 200 nm n-GaN spreading layer and a 20 nm n++ GaN contact layer (TJ-A). Another TJ structure was grown with the same growth parameters than TJ-A, but with the addition of a 4-nm thick In0.1Ga0.9N intermediate layer (TJ-B). The last TJ (TJ-C) was identical to TJ-A though grown under a standard-temperature-regime (1080°C). The n++ and n-GaN layers of the TJs were grown using N2 as the carrier gas, at a pressure of 10 kPa and a growth rate of 2500 nm/h. These parameters were chosen in order to avoid as much as possible the re-passivation of the Mg acceptors of the p-GaN layers by H. Under these growth conditions, the time needed to grow the n++ and n layers is less than 10 minutes. The LED structures were exposed to air before the growth of the TJs. No chemical treatment was used at this step as it has been shown that impurities at the surface might contribute to the tunneling probability by inserting midgap states into the depletion region.6,14,17 The surface morphology of the samples was evaluated by scanning electron microscopy (SEM) and atomic force microscopy (AFM).
The samples were clean room processed using standard photolithography and reactive ion etching (RIE) steps to fabricate rectangular mesas with a surface of 0.01 mm2 (100x100 μm2), followed by the deposition of the contacts by e-beam evaporation, which consisted of a Ti/Al/Ni/Au stacking (30/180/40/200 nm respectively) for all the “n” contacts and Ni/Au (20/200 nm) for the “p” one (in the case of the S-LED, used as a reference). However, before the p-contact, a Ni/Au (5/5 nm) electrode was deposited on the entire emission surface in order to induce homogeneous current spreading. The contacts were then annealed by rapid thermal annealing (450°C for Ni/Au and 750°C for Ti/Al/Ni/Au).
All the electro-optic properties of the LEDs were measured on wafer at room temperature under CW conditions. The current-voltage characteristics were measured using a Keithley 2104 source-meter. The electroluminescence was detected using a BWTek spectrometer. The output power and external quantum efficiencies were evaluated by approaching a calibrated Si photodiode 30 mm above the LEDs.
RESULTS AND DISCUSSION
Figure 2 shows the SEM images of the different samples incorporating a tunnel junction. For the samples TJ-A and TJ-B, the surface morphology is rough as expected from the large growth rate and the low growth temperature. For the sample TJ-C, a smooth surface without any pits is obtained thanks to the standard GaN growth temperature (1080°C) used for the growth of the n-GaN on top of the base LED. AFM was performed in order to investigate the surface morphologies. Due to their strong roughness, AFM experiments were not performed for samples A and B. The root mean square roughness is 0.8 nm for the S-LED and 2.0 nm for TJ-C for 10x10 μm2 AFM scans (Fig. 3), confirming the SEM observations.
Figure 4 gives images of the top view electroluminescence of the LEDs at an injection current of 1 mA. The peak emission wavelength is 442, 436 and 443 nm for the LEDs TJ-A, TJ-B, and S-LED, respectively. There was no electroluminescence detected for TJ-C, which is interpreted as the consequence of the re-passivation of Mg acceptors by H during the growth at the standard GaN growth temperature of 1080°C. This result is obtained despite the fact that a N2 carrier gas is used for these layers. It is worth noting that H can be provided by the decomposition of the NH3 molecules, and diffuses to the p-GaN, inducing the passivation of the previously activated Mg acceptors. Our result, along with the lack of electroluminescence from one TJ-LED previously grown under a H2-carried 1080°C regime (not shown in this paper), agrees with the fact that the re-passivation of Mg during the MOCVD growth is thermally activated, as already pointed out by P. Sohi et al.21 For the other samples TJ-A, TJ-B, and S-LED the intensity of the light emission is relatively homogeneous. From this point of view, the TJ-LEDs using a low n-GaN growth temperature give similar results to the standard LED S-LED with a semi-transparent electrode.
The J-V characteristics of the LEDs are shown in Fig. 5a. The voltages at a current density of 100 A/cm2 are 12.5V, 6.9V and 5.4V for the LEDs TJ-A, TJ-B, and S-LED, respectively. This reduction of the voltage for the TJ-LED incorporating a (Ga,In)N layer is consistent with previous results from the literature.8–10 The tunneling probability increases due to the band bending induced by the strong polarization fields related to the presence of the (Ga,In)N layer.
The differential resistances dV/dJ (Fig. 5b) calculated for TJ-B are 1.3x10-2 Ω.cm2 at 100 A/cm2 and 3.3x10-3 Ω.cm2 at 1 kA/cm2. The resistance of this structure is lower than the best values obtained for full MOCVD blue LEDs with tunnel junctions using lateral outgassing treatment13 and slightly higher than the ones obtained using lower growth temperatures and rates.21 Nevertheless, the structure of our tunnel junctions can be largely improved by increasing the In content of the InGaN layer of the tunnel junction9 or increasing the Si doping largely above 1x1020 cm-3,15 leading to lower operation voltages.
Figure 6 shows the output power and external quantum efficiency (EQE) versus the current for the S-LED, TJ-A and TJ-B. The light output power of the S-LED at a current of 20 mA is at least doubled when using any of the TJs (0.7 mW for the S-LED against 1.5mW and 1.4mW for TJ-A and B respectively). The peak external quantum efficiencies, which depend on the injection and extraction efficiencies, are 2.4%, 4.8% and 4.3% respectively for S-LED, TJ-A, and TJ-B. The higher output power and EQE of the TJ-A and TJ-B can be explained by the fact that there is no absorbing semi-transparent electrode covering the mesa surface. Additionally, the roughness of the n-GaN surface, caused by the low growth temperature, may increase the light extraction efficiency. TJ-A presented better EQEs than TJ-B due to partial absorption of light by the InGaN interlayer (at least 2% considering a single passage of the emitted light) - however, TJ-A also suffered a significant heating effect at high current densities (for I > 60 mA) due to its stronger resistivity. The same effect is not as strong in TJ-B because of the better electrical behavior caused by the insertion of the InGaN intermediate layer. Finally, the wall plug efficiencies of the LEDs, which rely not only on the operating voltage but also on the optoelectronic properties of the device, are 1.08%, 1.75% and 1.25% for the TJ-A, TJ-B and S LEDs respectively. Overall, the best performance is obtained with the InGaN based TJ LED, which best compromises on the electrical and optical properties.
In summary, p++/n++ tunnel junctions fully-grown by MOCVD on top of blue LEDs could be fabricated without the use of any lateral degassing treatment. The reduced re-passivation of the Mg acceptors from the p++ layer was made possible not only through the use of a low growth temperature and of N2 as a carrier gas, but also by using high growth rates reducing the re-passivation of acceptors and, thus, increasing the injection efficiency. Under these conditions, the LEDs with TJs were found to be more electroluminescent because of the better injection efficiency associated to the TJ. The lack of electroluminescence for the sample grown under a high temperature regime suggests that, in this case, the re-passivation of the acceptors is significant enough and thermally activated; it is thus essential to adapt the growth conditions. Additionally, using a high growth rate induces a rough morphology of the surface of the n-GaN layer, which works as an advantage for the light extraction efficiency of top emitting LEDs.
We would like to thank GANEX (ANR-11-LABX-0014) and the French National Research Agency (ANR) for funding the DUVET project (by which this paper was made). We would also like to thank Drs. Jean Massies, Mathieu Leroux, Julien Brault and Philippe de Mierry for the critical reading of the paper.