Demonstration of blue semipolar $(20\overline{2}\overline{1})$ GaN-based vertical-cavity surface-emitting lasers

1. Introduction

Group III-Nitride vertical-cavity surface-emitting lasers (VCSELs) have several unique properties that are advantageous over conventional LEDs and edge emitting laser diodes for certain applications, such as lighting, displays, and optical sensors. This is primarily due to their low threshold current, circular beam profile, and arraying capabilities. Since III-N VCSELs were first demonstrated in 2008 [1], multiple groups have shown electrically injected operation with differential efficiencies (η_d) up to 32% [2–11]. Though a wide variety of device designs have been utilized, the crystal orientation of the substrate has been mostly confined to the basal c-plane. Nonpolar and semipolar growth planes provide an interesting alternative to c-plane devices due to the reduced quantum confined Stark effect; higher material gain [12]; lower transparency current density; and inherent polarization [13,14]. Anisotropic gain in lasers grown on nonbasal substrates leads to VCSEL arrays in which each laser is polarization locked in the same direction, an important feature not easily achieved on the more common basal c-plane, as was demonstrated on m-plane [15–17]. However, it is challenging to push m-plane lasers to blue and green wavelengths due to poor indium incorporation and high defect formation [18–20]. On the other hand, certain semipolar orientations have shown enhanced indium incorporation, uniformity, and morphology, enabling long wavelength devices [21,22].

Forman et al. showed CW lasing of an m-plane VCSEL for the first time by optimizing the MBE grown tunnel junction (TJ) intracavity contact, minimizing surface roughness before deposition of the distributed Bragg reflector (DBR), increasing the thickness for heat spreading, and improving the bond interface [15]. The device utilized a flip-chip, dual-dielectric design with an Al ion implanted aperture. Although CW lasing was achieved, the device suffered from a poor η_d of 0.3% near threshold, with an increase to 0.8% at a higher drive current. Mishkat-Ul-Masabih et al. later showed similar device properties for an m-plane device incorporating a porous-GaN based DBR, with a peak pulsed power of 1.5mW and a η_d of 0.3% [17]. Both of these devices operated near 410 nm. Accordingly, we report here the first demonstration of semipolar$(20\overline{2}\overline{1})$VCSELs emitting at 444.5 nm which are polarization locked along the a-direction.

2. Method

The VCSELs, shown in Fig. 1, use a dual-dielectric DBR design with substrate removal through selective photoelectrochemical etching of a sacrificial layer, and aperture definition through Al ion implantation, similar to Ref [15]. The epitaxial structure was grown using atmospheric metal-organic chemical-vapor deposition (MOCVD) on freestanding$(20\overline{2}\overline{1})$GaN substrates and consists of a sacrificial 3x MQW (3 nm wells, 7 nm barriers, 430 nm emission), 20 nm n⁺GaN, 390 nm n-GaN, 45 period superlattice (3 nm In_0.04Ga_0.96N, 3 nm In_0.06Ga_0.94N), 2x MQW active region (3.5 nm wells, 7 nm barriers), 10 nm p-AlGaN electron-blocking layer (EBL), 105 nm p-GaN, and 10 nm p⁺⁺GaN. After activating the p-GaN, mesas were etched with reactive ion etching (RIE) and a Ti/Au hard mask was deposited to define the aperture during ion implantation. Al ions were implanted with a dose of 10¹⁴ cm⁻² and an acceleration voltage of 20kV. The reduction in the implant dose, compared to previous samples, was done to limit the risk of optical absorption loss, since the lower dose was found to still provide adequate electrical isolation. After removing the metal hard mask with heated aqua regia, the sample was cleaned directly prior to the tunnel junction (TJ) growth by dipping in buffered HF for 5 min. A 10 nm n⁺⁺GaN tunneling layer followed by a 52 nm n-GaN current spreading layer were grown by MOCVD and the p-GaN was re-activated through the mesa sidewalls. Ion beam deposition (IBD) was used to deposit the 16 period SiO₂/Ta₂O₅ p-DBR with a 65 nm Ta₂O₅ cavity tuning spacer. A deep RIE etch exposed the sacrificial wells and was followed by Ti/Au (20 nm/ 700 nm) deposition in the etched trenches as the PEC cathode, and on the mesas as the p-side contact. The sample was flip chip bonded to a sapphire submount coated in Ti/Au/In/Au (20 nm/ 700 nm / 1500 nm/ 200 nm) through solid-liquid interdiffusion bonding.

 

Fig. 1 Schematic of the semipolar VCSEL device structure.

Download Full Size | PPT Slide | PDF

It has been shown that on$(20\overline{2}\overline{1})$samples room temperature PEC etching leads to a rough etching surface [23]. However, upon decreasing the temperature of the etch, the resulting surface roughness can be reduced. Etching of$(20\overline{2}\overline{1})$samples was conducted at multiple bath temperatures, and it was found that at −5 °C, the roughness becomes comparable to the epitaxial roughness at 400 pm RMS roughness as measured by AFM on test samples. Therefore, PEC etching of the sacrificial wells was performed in 1M KOH under a 405nm LED array at −5 °C to limit surface roughness for these VCSELs. After removal of a residue present after PEC etching, by swabbing in a dilute Tergitol solution, the Ti/Au (20 nm/ 370 nm) n-contact was deposited by electron-beam deposition. Finally, a 12 period SiO₂/Ta₂O₅ n-DBR with a Ta₂O₅ spacer layer (45 nm) was deposited by IBD. The spacer layers were deposited to align the TJ and the active region with the node and antinode of the expected lasing mode, respectively. The spacers also served to align the cavity resonance with the gain spectral peak, as estimated by the peak of the room temperature spontaneous emission modified with a small red shift to account for self-heating. While the spontaneous emission and the gain spectra do not perfectly align, the spontaneous emission peak is used as a proxy to estimate the position of the gain peak

Increasing cavity length has been shown to increase the threshold, and decrease the power for a given input power [24]. The increase in the confinement factor, as well as the decrease in the round trip loss of the shorter cavity are expected to lower the threshold gain. Having a shorter effective length increases the mirror loss, thereby increasing the expected differential efficiency. Therefore, the devices were designed to have a 5λ cavity length, compared to the previous 23λ m-plane devices to reduce the threshold current and increase the differential efficiency. However, shorter cavities have also been shown to have inferior thermal performance, potentially inhibiting CW operation [2,15,25].

Polarization was measured by inserting an optical polarizer between the device and an optical fiber coupled spectrometer aligned above the lasing mode. The polarization ratio (p) is given by:

3. Results and discussion

As shown in Fig. 2, under pulsed operation with 1000 ns pulse width and a 2.5% duty cycle, the peak power was 1.85 mW and the threshold current and voltage were 4.6 kA/cm² and 7 V. At threshold, the peak of the spontaneous emission was at 435 nm, 10 nm shorter than the wavelength of the lasing mode, suggesting the presence of a misalignment between the gain spectrum and the resonance mode. This misalignment occurred due to the n-side Ta₂O₅ spacer layer being of the wrong thickness and likely resulted in a higher threshold current being required to produce the necessary gain. It is expected that the spontaneous emission initially blue-shifts with increasing current injection due to band filling and screening of the polarization field, though the effect is significantly reduced compared to c-plane devices due to the reduced polarization fields across the quantum well on$(20\overline{2}\overline{1})$. It is then expected to redshift due to self heating, though this effect is also minimal when using low duty cycle pulsed operation. Thus the net misalignment is expected to have remained relatively consistent throughout testing as will be discussed later. The differential efficiency was 2.4% for the mode at 445 nm, which had a spectrometer limited resolution of 2 nm. The differential efficiency from one side of the device can be found with the following equation:

 

Fig. 2 Light-current-voltage results (a) for a 12 µm aperture VCSEL under pulsed operation with a 2.5% duty cycle and a 1 μs pulse width. The inset of (a) depicts the nearfield pattern at 5% above threshold. This pattern is maintained at higher current densities. The spectrum (b) for an adjacent 12 µm aperture device is shown for different stage temperatures when held at 12 mA. This second device was used for thermal characterization due to a catastrophic failure of the other during testing.

Download Full Size | PPT Slide | PDF

The experimental differential efficiency being just under half the simulated value is likely due to a combination of factors. As the lasing mode was shifted from the design wavelength, the tunnel junction was not likely placed directly on the null of the standing wave and therefore increased the absorption in the cavity. Additionally, the inset of Fig. 2(a) shows that the lasing mode was near the edge of the aperture and, therefore, partially overlapped with the implanted region and the contacting metal. Spatial misalignment and complicated mode structure have been seen by many groups with planar cavity VCSELs [1,3,10,26,30,31], and is thought to be due to inhomogeneous current injection across the aperture, though the exact cause is still unknown [26]. While Hamaguchi et al. have shown mode control using a curved mirror [32], this is still a significant issue for planar cavity devices that bears further investigation. The overlap was determined by modeling the mode as a two dimensional Gaussian as a close approximation to the LP₀₁ mode, and determining the percentage of the mode extending into the absorbing region, either the implanted area or the metal. The resulting 1/e² spot diameter was 2.1 µm. As seen in Fig. 3(a), it was difficult to ascertain the exact positions of the absorbing regions relative to the mode; thus, a range of potential losses was found by shifting the center of the mode 200 nm towards, and away, from the center of the non-absorbing region. Due to the nonlinearity of a Gaussian profile, there is a much greater effect on the calculated loss when the absorbing region is closer to the mode. Despite lowering the implant dose to reduce absorption loss, the implant still has a rather high absorption coefficient of 1400 cm⁻¹ as measured from transmission/reflection spectra on an implanted test wafer. The center of the unimplanted area was found to be 3 μm from the center of the mode and introduced ~1 cm⁻¹ of loss (0.6-2 cm⁻¹ depending on the exact position of the edge of the implant). The lasing mode also overlaps with the metalized region, significantly increasing the absorption. The edge of the metal was estimated to be 4.2 µm from the center of the mode, which would suggest 0.0065% of the mode overlaps with the metal. This gives 6 cm⁻¹ of loss (3-12 cm⁻¹ depending on the exact position of the edge of the metal) assuming that the metal overlap only occurs over the effective penetration depth into the top mirror [33,34]. While these losses don’t account for all of the excess loss, they do highlight the importance of centering the mode in the aperture. The polarization dependence of the emission spectrum is shown in Fig. 3(b) and the polarization along the a-direction is clearly seen.

 

Fig. 3 The white circle on the nearfield pattern (a) defines the edge of the metal contact. The green circle represents the unimplanted aperture, while the red circle defines the mode. The light seen to the right of the unimplanted aperture is due to scattering from the edge of the bonding metal. (b) Shows the spectrum of an example 8 um VCSEL at different polarization angles. The polarization of maximum intensity was found to be parallel to the a-direction.

Download Full Size | PPT Slide | PDF

Although the threshold current and differential efficiency show improvement relative to previously reported m-plane VCSELs with a similar structure, the devices were not able to lase under CW operation. Above threshold, the devices had a differential resistivity of 8.9 × 10⁻⁵ Ω cm². The rather high operating voltage increased the dissipated power in the device, increasing the heat generated. The thermal impedance of the devices was estimated from the shift in the spectrum with changing stage temperature (Δλ/ΔT_st) at low duty cycles and while changing duty cycle (Δλ/ΔP) at fixed stage temperature. Lasing was maintained up to 50% duty cycle and to a stage limited temperature of 70 ̊C. Δλ/ΔT_st and Δλ/ΔP were found to be 0.07nm/K and 125 nm/W, respectively, which give a thermal impedance of 1800 K/W, in fair agreement with the value of 1500 K/W found by direct measurement of the operating device with a thermal imaging microscope. This is higher than the 1400 K/W predicted by a COMSOL model of the 5 λ devices [15], and is likely due to a poor bonding interface, as shown in Fig. 4. At 50% duty cycle, the LIV curve began rolling over at 12.6 mA and 7.9 V, which would suggest a roll over temperature of 110 ̊C, in between previously reported values of 98 C [35], and 160 C [2,15]. Investigating the shift in the lasing peak with changing duty cycle gave a Δλ_lasing/ΔP of 25 nm/W, indicating that as the input power increases for a set operating temperature the lasing mode shifts towards the gain peak at 100 nm/W. Additionally, the lasing mode shifted at 0.01 nm/K with changing stage temperature, significantly less than the spontaneous peak. In actual operation it was found that there was a negligible change in the peak separation when changing input current for the 1 μs pulses used in most of the testing. An 8 micron diameter device operated up to 75% duty cycle with a 500 ns pulse width.

 

Fig. 4 The secondary ion image taken with focused ion beam microscopy shows the voids formed in the bonding metal, highlighted by the red rectangle.

Download Full Size | PPT Slide | PDF

4. Conclusions

In conclusion, we demonstrated a blue$(20\overline{2}\overline{1})$semipolar VCSEL with a cavity length of 5λ, an ion implanted aperture, and a dual dielectric DBR design. The peak power under pulsed operation at a 2.5% duty cycle was 1.85 mW for a 31.5 kA/cm² current density. The threshold current was 4.6 kA/cm² and the differential efficiency was 2.4% for the mode at 445 nm of a device with a 12 µm aperture. The lasing emission was found to be 100% plane polarized along the a-direction. Lasing was achieved up to a 50% duty cycle and the thermal impedance was estimated to be 1800 K/W.