1. Introduction

Widespread applications for (Al,In,Ga)N based laser diodes (LDs) have driven tremendous development of high brightness and efficiency LDs into lighting, laser display, and information storage [1–3]. While blue emitting devices are mature, green LDs have lagged in brightness and efficiency, and broadening the wavelength range past green has proven difficult. There is a need for direct emission yellow LDs at 589 nm targeting a transition in the sodium atom for atomic cooling, where a direct generation yellow beam can be integrated onto photonic systems to study photon entanglement [4,5]. While III-nitride light emitting diodes have proven emission into the red [6], there are few reports of InGaN edge emitting LDs reaching longer than 550 nm by electrical injection [7–8]. Other methods to reach narrow yellow emission depend on low power and inefficient frequency conversion or gas, ion, and dye lasers [9–11] which are not compatible for integration onto a chip.

Eliminating the “green gap” requires a solution to the lattice mismatch between GaN and InGaN. Higher indium incorporation needed for green emission induces strain causing defects and increases piezo-electric fields, reducing material gain. This is exacerbated by the reduced dispersion between GaN and InGaN or AlGaN material at longer wavelengths, weakening the effectiveness of lower cladding needed for modal confinement. Porous GaN has been shown as a replacement for the lower cladding in blue LDs by offering high index contrast with the core while remaining lattice matched [12–16]. It reduces the need for AlGaN cladding, which has been correlated with performance degradation and sudden failures [17]. Porous GaN has also been shown to increase the optical confinement factor more than 350% over GaN or AlGaN cladding, and is comparable to lattice-matched InAlN at longer wavelengths [18]. Previous works have suggested porous GaN allows similar heat dissipation to InAlN [19,20], leading it to be a possible candidate for inclusion in long wavelength III-nitride LDs.

Here we demonstrate porous GaN in laser diodes on a semi-polar orientation of GaN emitting in the green spectrum. The porous cladding has proven comparable performance to traditional green LDs from this group, though further optimization of the p-side, active region, and porous material will be needed to reach Watt level performance achieved by state-of-the-art green lasers.

2. Experimental

Metalorganic chemical vapor deposition (MOCVD) was used to grow the structure shown in Fig. 1. The growth substrate was the $20\overline 2 1$ orientation of a bulk grown GaN substrate, and began with a 400 nm low Si-doped 2 × 10¹⁸ cm⁻³ GaN layer used for current spreading in the electrochemical etch, followed by the intended porous cladding layer at 3 × 10¹⁹ cm⁻³ and 180 nm thick, both grown at 1180 °C. The temperature was then dropped to 965 °C and the doping taken low to 1 × 10¹⁷ cm⁻³ to avoid parasitic etch paths during the EC etch, as seen in past studies [12,15]. First 200 nm of n-GaN was grown, then a small amount of indium was incorporated for 80 nm of approximately n-In_0.02Ga_0.98N before dropping the temperature again to grow the multiple quantum wells (QWs). 20 nm of UID GaN was grown at 850 °C followed by two periods of approximately 3 nm In_0.25Ga_0.75N QWs at 760 °C with a 10 nm GaN barrier back at 850 °C. The temperature was ramped back up to 915 °C to grow the p-side, and a 10 nm p-Al_0.28Ga_0.72N electron blocking layer (EBL) was grown doped with 3 × 10¹⁹ cm^–3 of Mg. A 15 nm p-In_0.05Ga_0.95N waveguide was grown on top of that, capped with a two-step pGaN with 245 nm doped at 2 × 10¹⁸ cm^–3 and 20 nm doped at 1 × 10²⁰ cm^–3 for electrical contact to ITO.

 

Fig. 1. (a) Device schematic and (b) optical mode profile.

Download Full Size | PDF

After activating the p-type material at 600 °C in an atmosphere furnace, 200 nm deep ridges were etched via chlorine based reactive ion etching (RIE), and 140 nm of SiO₂ was deposited in the field. Ridges are oriented parallel to the projection of the c-axis to maximize gain on this substrate orientation [21], and consisted of an array including widths of 2.5 µm and 8 µm, as well as lengths of 900 µm, 1200 µm, 1500 µm, and 1800 µm. Similarly, a 2 µm deep trench was etched with RIE parallel to the ridge opening a window to the highly doped material. After soldering an indium dot to the corner, samples were submerged into a 0.6 M oxalic acid bath with conductive tweezers under an applied bias of 5.5V for 150 min creating the porous GaN material for lower cladding. The pores undercut the laser ridge propagating from the trench as seen from the SEM images in Fig. 2, prepared by focused ion beam. The porous area was etched only to undercut the ridge, measured by intermittent pauses in the etch to optically inspect the area with color change. The porosity was measured to be approximately 30% by image analysis of the SEM cross section. The interface between the epitaxial growth and substrate also etches, which we have speculated etches from contaminants at the growth surface. This is far outside the optical mode and expected to have little impact on the laser performance. Fabrication of the LDs was completed according to [12], with an ITO p-contact, backside metal, and CAIBE facets.

 

Fig. 2. (a) Scanning electron micrograph of LD cross section and (b) plane view of porous layer opened by focused ion beam.

Download Full Size | PDF

Pulsed measurements were taken on-wafer under 500 ns pulsed with 0.5% duty cycle. A lens was placed near the sample to collect light from the facet and reflections off the substrate into an integrating sphere. Continuous wave (CW) measurements were taken after soldering the sample to a copper block for heat sinking.

3. Discussion of results

A summary of pulsed electrical injection measurements is recorded in Fig. 3 for representative porous devices. The emission is uniform across the ridge and in the far-field pattern. Devices that were held outside of the acid were not made porous and did not reach laser threshold under electrical testing. At a wavelength of 510 nm, the best devices have a slope efficiency of 0.13 W/A. The maximum total power was measured at 150 mW but limited by the maximum current supplied by the pulse generator. When compared with the best green lasers from this group [22], power is lower, but threshold voltage of the porous devices is much lower at 7.5 V, resulting in a higher peak wall plug efficiency of 0.9%, albeit at high current density. Current runs through the backside of the substrate, but there is an electrical conduction path around the porous layer. Comparing IV measurements of devices with the porous layer and without it show matched electrical resistance.

 

Fig. 3. Pulsed electrical injection measurements including (a) device under operation, (b) far-field pattern, (c) electroluminescent spectrum with inset of full-width at half-maximum for a 1800µm x 2.5 µm ridge under varied current density, (d) light-current density-voltage comparison and (e) wall plug efficiency for a 900 µm x 2.5 µm ridge.

Download Full Size | PDF

Threshold current density is high for all devices, with the best shown to be 14 kA/cm² for the longest 1.8 mm ridges. At these high current densities, it is likely a significant portion of electrons overshoot the quantum wells and EBL resulting in a low injection efficiency. By plotting the inverse of the differential efficiency against varied device lengths in Fig. 4(a), the injection efficiency is confirmed to be a meager 16%. High indium composition quantum wells for green emission have been shown to exhibit alloy fluctuations impacting carrier injection [23], though in this case two quantum wells should be relatively uniform for indium concentration. It is also possible changes in the p-side growth conditions contributed to this issue. Growth temperatures for the EBL and pGaN were decreased as much as 40 °C from traditional blue lasers to mitigate formation of “dark triangle” morphology defects, outlined in [22]. We suspect fluctuations in EBL composition and/or dopant incorporation may also have impaired the injection efficiency, and sample destruction by secondary ion mass spectroscopy could be done for better evidence of this. It is not clear what other factors are leading to low injection efficiency, but the p-side material can likely be optimized as has been shown in [24] and [25].

 

Fig. 4. (a) Inverse differential efficiency plotted against ridges of varying length, (b) material gain of devices compared with material gain calculated from segmented contacts in (c) gain curves and (d) absorption curves

Download Full Size | PDF

High threshold current density also presents a problem for reaching longer wavelengths. Growing on the $20\overline 2 1$ GaN plane mitigates the quantum confined Stark effect [21], but the spectral peak still blue-shifts over 10 nm from J = 4 kA/cm², where a modest green laser might reach threshold [3], to J_th = 14 kA/cm². It is critical to understand why the threshold current density is so high. The optical confinement factor Γ is calculated to be 2.1% [26], reasonable for a GaN green laser [18], pointing to problems with the material gain or profuse loss.

Estimates for the gain and loss were calculated by two independent methods to ensure accurate values. The variable stripe method mentioned above is where the efficiency of multiple devices is plotted inversely against the length and fit to a line. Assuming perfectly etched facets (R = 0.18) gives the mirror loss, and the internal loss was calculated to be <α_i>_var-str = 36 /cm. The material gain at threshold is a sum of the internal loss and mirror losses over the calculated confinement factor, and is plotted in Fig. 4(b). Similarly, the gain and loss were found using the segmented contacts method. Described in detail by Blood et al. [27], two contacts in series on a ridge were alternately pumped measuring intensity and related to create gain and absorption curves, shown in Fig. 4(c) and Fig. 4(d), respectively. The absorption curve is asymptotic to the internal loss at long wavelengths and comes out to <α_i>_seg-ct = 45 /cm, reasonably close to previous calculations using the variable stripe. The peaks of the gain curve were also plotted against current density in Fig. 4(b) and fits well within the points calculated from the variable stripe method.

Sources for the internal loss need to be found to understand the performance limitations. Absorption due to free carriers was estimated using [28] and tabulated in Table 1. While the higher doping in the porous cladding layer might be expected to increase absorption losses, it has a very small effect due to low modal overlap. The highest losses are instead seen in the pGaN cladding and ITO as typically found for GaN lasers, amounting to <α>_abs = 23 /cm. The best lasers show losses less than 10 /cm [29], and better mode placement is needed to reduce absorption losses. The porous layer was designed to maximize overlap of the optical mode over the active region, but there is a tradeoff with absorption in the p-type material. A deeper investigation is needed to optimize the tradeoff between confinement factor and absorption losses in the p-side for our porous devices.

Table 1. Values used to estimate free carrier absorption

View Table

Previous works have shown high loss due to scattering at pores in parasitically etched layers [12,15]. Inspection of the facet cross section in Fig. 2 shows the etch was highly selective to the target layer in this case, but scattering in the porous cladding may still explain the excess loss. Scattering was estimated using Rayleigh spheres from Eqs. (3) and (4) in [30]. The pore density and radii were averaged from lateral, vertical, and longitudinal measurements in the 2D cross sections of the porous layer in Fig. 2, approximately 1.5×10¹⁴ cm^–3 and 40 nm, respectively. This, coupled with a simulated 0.38% overlap with the cladding layer shows scattering loss to be <α>_sc = 21 /cm. This is lower than seen on blue lasers with porous GaN [12,13], but still inhibiting device performance. It has been noted that the Rayleigh spheres may be an inaccurate model for scattering centers [12], as evidenced by the micro-pipe nature of the porous material in Fig. 2(b). The pores are clearly non-spherical and elongate perpendicular to the optical field. Our estimate reasonably accounts for the remaining excess loss, but a full finite-domain time dependent model would need to be performed to assess this value more accurately.

Designing the porous layer for minimal excess loss comes with several tradeoffs. It is beneficial to increase the porosity to maximize index contrast and minimize modal overlap. This can be done by increasing the EC etch voltage, though this reduces EC etch selectivity as well as increases the pore size. Scattering loss rises with pore size to the power of four [30] while modal overlap has been shown to decrease at a low rate with porosity above 30% [16], making this a non-viable path. As has been discussed in previous works, the best way to increase porosity is through high doping and minimizing pore size. Germanium can replace silicon as a dopant source to reach higher concentrations, and thus lower pore size, while avoiding defect formation during growth [31]. If pore size can be decreased to insignificant scattering levels, there may be benefits to utilizing a graded porous layer similar to graded refractive index separate confinement heterostructures as seen in other material systems [32].

The calculated material gain from both methods is plotted together in Fig. 4(b) and logarithmically fit to g = g_onln(J) + C. The gain factor g_on was found to be 940 /cm. This value is proportional to the momentum matrix element. Table 4.1 in [33] shows for green emitting In_0.24Ga_0.76N material to be approximately a third of the corresponding blue In_0.15Ga_0.85N active region. Comparison to [24] where g_on = 3200 /cm for a semipolar two quantum well blue emitting LD shows the porous cladding green devices are reasonable and may need only minor changes to the active material to improve the gain.

Devices were tested under CW injection. The spectrum under constant CW operation is shown in Fig. 5. The full-width at half maximum (FWHM) in Fig. 5(b) shows narrowing up to 9 kA/cm² tracking well with the pulsed behavior. Many changes occur as the current density rises above 10 kA/cm²: the FWHM increases again, the peak intensity saturates, then red-shifts and falls. Spectral ripple also becomes apparent. This evidence suggests a contribution of stimulated emission [34], but self-heating prevents laser action.

 

Fig. 5. Continuous wave electrical injection measurements for 1800µm x 2.5 µm bar soldered to copper block including (a) electroluminescent spectra and (b) full-width at half-maximum.

Download Full Size | PDF

A separate experiment was performed to understand the temperature dependence in which an 1800µm x 2.5 µm bar was instantly driven to high current density under CW injection. Tested at J = 20 kA/cm², far-field laser emission was visibly observed and recorded for 32 ms before falling back below threshold. Since threshold current density is known to be 15 kA/cm² at room temperature, temperature dependence can be estimated based on the rise in threshold current. Thermal conductivities for bulk GaN and 30% porous GaN are approximately 130 W/m-K and 7 W/m-K, respectively [19]. With a 330 µm bulk GaN substrate, thermal impedance through the substrate can be estimated as 2-D heat flow from Eq. (2).67 in [33] giving Z_Tsub = 0.9 K/W. The porous layer is only 180 nm, less than 0.05% of the total substrate thickness, and heat flow can be estimated as 1-dimensional using Eq. (2).66 in [33] to give Z_Tporous = 0.6 K/W and total thermal impedance of Z_T = 1.5 K/W. The result is that for the given current and V = 7.2 V, the temperature rise is ΔT = 9.7 K higher than room temperature. Using I_th = I₀exp(T/T₀) for both conditions, the characteristic temperature T₀ is found to be 34 K. This value indicates the dependence of threshold current on temperature, and values less than 100 tend to need thermoelectric cooling [33].

While the lower thermal conductivity of porous GaN is undoubtedly a problem, there are ways to improve the thermal performance. The porous cladding contributes a third of the thermal impedance but heat dissipation can be significantly improved by removing the substrate. Lifting off the substrate with a technique such as photo-electrochemical etching [35] would allow soldering a heat sink closer to the porous layer. Decreasing the substrate thickness from 330 µm to 5 µm, total thermal impedance Z_T drops nearly in half to 0.85 W/K. The accompanying temperature increase is only 5.5 K, with the threshold current and characteristic temperature expected to drop. Another effective way to improve CW performance is to decrease the losses from absorption and scattering. Reducing the excess losses would allow lower threshold conditions. If J_th can be improved to a reasonable 4 kA/cm², the power dissipated as heat will drop and the temperature rise at threshold would decrease to a manageable 1.1 K after removing the substrate. Finally, reducing the series resistance with doping optimization or contact can also be done to enhance performance at CW.

4. Conclusions

Porous GaN can replace traditional cladding to make green lasers, but the threshold condition needs to be improved by fixing the injection efficiency and reducing loss. The loss due to scattering was lower than shown on blue LDs with porous cladding, but too large to match industry leading green laser performance. Laser behavior was not seen under CW operation, though there are ways to improve heat dissipation. A fundamental solution to porous GaN heat dissipation is lacking. Though porous GaN cladding is an exciting technology to broaden the applicable wavelength of III-nitride LDs, there are many challenges to overcome.