1. Introduction

The III-nitride system has been used to fabricate optoelectronic devices for many applications requiring visible and ultraviolet light. High power (Al, In, Ga)N light emitting diodes (LEDs) are efficient light emitters but show droop at high current densities [1–3]. A large component of droop is caused by Auger recombination which is a non-radiative process involving three charge carriers [4]. Because the Auger rate is proportional to the cube of the carrier density, it limits the efficiency of an LED at high current densities. In addition, droop in today’s commercial devices is exacerbated because of the strong electric field present in the quantum wells (QWs) due to piezoelectric and spontaneous polarization oriented in the growth direction. These fields reduce the electron-hole wave function overlap in the active region and limit the thickness of the QWs, which increases carrier densities. Fabricating devices on semipolar and nonpolar crystal orientations can reduce or eliminate the polarization induced electric fields in the active region and recent results have shown promise for high efficiency devices using thick quantum wells [5,6]. However, these are still limited to low power densities due to efficiency droop. Unlike LEDs, laser diodes can operate at high power densities due to carrier clamping at threshold. This can reduce the chip area required for a device, which reduces substrate and growth related costs. It also opens up applications where the power density and directionality of a laser is beneficial

One of the factors preventing widespread use of laser lighting is that ohmic contacts to p-GaN are difficult to achieve [7,8]. The specific contact resistivity of an optimized semipolar p-GaN contact is often on the order of 10⁻⁴ Ωcm² which is significant for a laser operating at 5-10 kA/cm² [9]. The p-contacts for c-plane devices are somewhat better, and the use of a thin InGaN layer for polarization induced band bending can decrease contact resistivity further [10]. However thick p-GaN, even when highly doped, will still add significant voltage due to its high resistivity which of the order of 1 Ωcm [11]. A 500 nm thick layer of p-GaN would then add about 0.5 Volts for a current density of 10 kA/cm². In addition, the primary source of optical loss in GaN lasers is often the p-type GaN due to the relatively high magnesium doping required for conductivity [12,13]. Tunnel junctions are an alternative to standard p-contacts that have the potential to reduce operating voltages and optical loss in III-nitride lasers. They are formed when the depletion width of a p-n junction is narrow enough that electrons can tunnel from the valence band of the p-type material into the conduction band of the n-type material [14]. Low resistivity tunnel junctions have been difficult to achieve in III-nitride devices due to the large bandgap of GaN and high doping densities required [15,16]. More recently, several groups have utilized polarization induced band bending from an InGaN [17,18] or AlN [19,20] interlayer to reduce the depletion width and increase the tunneling current. Although the AlN interlayer tunnel junctions increase the tunneling current, the voltage drop across the junction is still too high to be used in high efficiency LEDs or LDs. The devices are also sensitive to thickness fluctuations and have strain and cracking limitations. The InGaN tunnel junction devices have better electrical performance but can add significant optical absorption to devices. The InGaN interlayers are also limited by strain and thermal stability which limits the amount of indium that can be incorporated.

A problem with MOCVD grown p-GaN is that high temperature post growth annealing is required to remove hydrogen and achieve p-type conductivity [21,22]. An n-GaN cap on Mg-doped GaN layer provides a strong barrier to hydrogen diffusion for several reasons. First, the internal electric field in the n-GaN/GaN:Mg junction oppose diffusion of H⁺. Secondly, in n-GaN, hydrogen is favored to be negative charged H^-. In n-GaN, H- has much higher migration barrier in comparison with H + in p-GaN [23,24]. To activate buried p-GaN, a sidewall needs to be exposed by etching so that the hydrogen can diffuse out. This also limits the device size due the lateral distance hydrogen must diffuse. In NH₃ molecular beam epitaxy (MBE) however, the partial pressure of ammonia, 2 × 10⁻⁴ Torr, is small enough that p-GaN remains active under standard growth conditions. A recent paper has shown good performance of p-n GaN tunnel junctions grown by ammonia MBE on top of MOCVD micro-LEDs [25]. The hybrid growth technique discussed in this publication was also used to fabricate the first III-nitride vertical cavity surface emitting laser utilizing a tunnel junction intracavity contact [26].

In this work, a laser diode structure was first grown on a bulk $(20\overline{21})$ GaN substrate. The structure was then taken out of the MOCVD and placed in an ammonia MBE growth chamber. The sample was then heated in the chamber which removed the hydrogen from the p-GaN and a highly doped n-GaN layer was grown on top which produced a tunnel junction at the regrowth interface on top of electrically conductive p-GaN. Further discussions on growth and characterization of the tunnel junction structure can be found elsewhere [27].

2. Experimental

Bulk $(20\overline{21})$ GaN substrates were obtained from Mitsubishi Chemical Corporation with a threading dislocation density of approximately 5$\times$10⁶ cm⁻² [28,29] and an epitaxial structure previously optimized for blue lasers was grown by MOCVD [7,12]. The structure consisted of 1 µm of n-GaN, a 65 nm layer n-In_.06Ga_.94N waveguiding layer, a 20 nm GaN barrier, a 4$\times$ multi-quantum well (MQW) with 3.5 nm In_.18Ga_.82N wells and 7 nm GaN barriers, a 20 nm final GaN barrier, a 12 nm p-Al_.18Ga_.82N electron blocking layer (EBL), a 65 nm p-In_.06Ga_.94N waveguiding layer, a 600 nm p-GaN layer, and a 10 nm p⁺-GaN contact. The sample was then taken out of the MOCVD and loaded into an ammonia MBE system. The p-GaN was activated in the chamber at 600 °C and due to the low ammonia overpressure during regrowth the p-GaN remained activated. The epitaxial structure grown in the MBE system was a 20 nm n⁺-GaN layer, a 100 nm n-GaN layer, and a 10 nm n⁺-GaN contact layer.

The sample was processed into a ridge laser diode with the ridge etch stopping at the p-InGaN waveguide (shallow ridge). Before stripping the photoresist, a sputtered SiO₂ layer was deposited on the sides of the ridges and in the field. The SiO₂ limits the injection area to the top of the ridge and prevents additional optical loss from the metal contacts. A 30/1000 nm Ti/Au topside contact was deposited on top of the ridge and a blanket 30/300 nm Ti/Au contact was deposited on the back of the substrate. A second RIE etch, approximately 1.5 µm deep, was used to form the laser facet. A diagram of the processed laser including the epitaxial structure and growth technique for the different III-nitride layers is shown in Fig. 1. A Fimmwave simulation of the optical mode profile is also plotted and the confinement factor was calculated to be 3.4%.

 

Fig. 1 (a). A schematic of the laser diode structure is shown. The regrowth interface between the MBE and MOCVD materials is at the tunnel junction p-n interface. 1(b). An optical mode profile is shown near the active region. The simulated confinement factor was 3.4%.

Download Full Size | PDF

3. Results

The electrical and optical characteristics of the laser were measured at room temperature using a two probe geometry under pulsed conditions with a pulse width of 1 µs and a duty cycle of 1%. The output power was measured out of one uncoated facet and was collected using a lens assembly to focus the laser through a neutral density filter, with an optical density of 1.0, onto a small Si photodetector. All measurements presented in this paper were taken from the same laser which was1800 µm long by 2.5 μm wide.

Figure 2 shows the light-current-voltage (LIV) plot for the laser diode. The threshold current was 284 mA which corresponds to a threshold current density of 6.3 kA/cm². The slope efficiency was 0.33 W/A which gives a differential efficiency of about 12%. If both facets emit the same power this would double the slope efficiency to 0.66 W/A and the differential efficiency to 24%. The laser was measured up to a current of 800 mA which gave an output power of 142 mW at 11.6 V. The differential resistivity of the device was calculated at an injection current above 10 kA/cm² and was found to be 2.3 × 10⁻⁴ Ω cm² which puts a limit on the resistivity of the tunnel junction resistivity at high current densities.

 

Fig. 2 A pulsed LIV plot for an 1800 μm by 2.5 µm laser shows a threshold current of 284 mA (6.3 kA/cm²) and a differential efficiency of 0.33 W/A out of one uncoated facet.

Download Full Size | PDF

The spectra as a function of operating current is shown in Fig. 3 as measured by an Ocean Optics USB2000 spectrometer. Above the threshold current of 284 mA a sharp peak emerges at the lasing wavelength of 444 nm. The dependence of the peak wavelength and FWHM of the spectra on current are also shown in Fig. 3. As the laser reaches threshold the peak wavelength moves towards 444 nm and the FWHM narrows continuously from approximately 17 nm at low injection current density to approximately 3 nm above threshold although this is partially limited by the resolution of the spectrometer.

 

Fig. 3 (a). The FWHM and peak wavelength are plotted as a function of current. 3(b). The measured laser spectra show a sharp increase in light around 444 nm indicative of lasing.

Download Full Size | PDF

4. Discussion

Although past GaN tunnel junction publications have shown high operating voltages without an InGaN or AlN interlayer, this technique was able to produce a low resistivity tunnel junction without an interlayer [19]. Secondary ion mass spectroscopy was done on a similar tunnel junction sample to determine how putting the tunnel junction at the regrowth interface affected the impurity content at the interface, and a large spike of oxygen was found. A Gaussian peak function was fit to the data and the peak area showed an interfacial oxygen concentration of approximately 1 × 10¹⁴ cm⁻². This oxygen spike was also confirmed via atom probe tomography [27]. If the oxygen atoms acted as electrically active donors this would reduce the depletion width dramatically as shown in SiLENSe simulations in Fig. 4(a). A calculation of the minimum distance an electron needs to tunnel to get from the valence band to the conduction band is plotted in Fig. 4(b) for several different delta donor densities at the interface. It can be seen that the delta doping dramatically reduces the bias required to obtain similar tunneling distances and could explain the low operating voltage of the tunnel junctions. Oxygen could also form trap states at the interface due to the high interfacial density which could allow for trap assisted tunneling. A silicon spike was not observed at the regrowth interface but this is likely due to the high intentional silicon doping concentration. Future work will need to be done to investigate how the concentration of oxygen affects the tunnel junction. The oxygen level could likely be increased using thermal annealing in an oxygen containing environment and decreased through chemical treatments.

 

Fig. 4 (a). Calculated band diagrams at a −0.5 V bias is shown with and without the addition of a 10¹⁴ cm⁻² delta donor spike at the interface for a p-n junction with an acceptor concentration of 2 × 10²⁰ cm⁻³ and a donor concentration of 1x10²⁰ cm⁻³. The delta donor spike reduces the minimum tunneling distance at −0.5 V from 5.5 nm to 3.2 nm. 4(b). The minimum tunneling distance is plotted for different biases and delta donor densities for a p-n junction with an acceptor concentration of 2 × 10²⁰ cm⁻³ and a donor concentration of 1x10²⁰ cm⁻³.

Download Full Size | PDF

The epitaxial structure of the laser contained thick p-GaN which placed the tunnel junction far away from the optical mode as was illustrated in Fig. 1(b). This was done to prevent any effect on the optical mode from the tunnel junction which means the demonstrated laser should have similar LI characteristics to a standard p-contact laser although a direct comparison was not made in this paper. However, one of the potential advantages for a tunnel junction laser is the ability to use thinner p-GaN than a standard laser structure which could reduce the series resistance of the structure. The thickness of p-GaN required in a laser is determined by the optical loss from the p-contacts or tunnel junction and future work will need to determine the optical loss of the tunnel junction. The optimal p-GaN thickness will be determined by the tradeoff between optical absorption and series resistance of the laser.

5. Conclusion

We have demonstrated a III-Nitride edge emitting laser diode grown on $(20\overline{21})$ bulk GaN substrates utilizing a tunnel junction as a topside p-contact. The threshold current was 284 mA which corresponded to a current density of 6.3 kA/cm². The slope efficiency from one uncoated facet was 0.33 W/A and gave a differential efficiency of 12%. Future improvements in the tunnel junction could allow for reduced optical loss and a lower operating voltage by reducing the p-GaN thickness required in edge emitting lasers.