Introduction
Microdisks have previously been fabricated in GaN-based systems, but performance has been limited by variabilities in the material growth and control of processing techniques. Most reports of GaN microdisks have used the index difference between the GaN layers and the sapphire substrate to confine light in the vertical direction8, 9, 10. These devices have shown modes at low temperature10 and room temperature lasing under pulsed operation8, 9, but the small index difference between sapphire and GaN leads to loss into the substrate, limiting the performance of these devices. Undercut microdisks have been fabricated using photoelectrochemical (PEC) etching11. These 4-
m-diameter microdisks showed a modal structure at room temperature12 and lasing under pulsed operation at 10 K (ref. 11), but the lasing threshold remained high at 6 MW cm-2. These microdisks were limited in performance by their sidewall roughness and deviations from circularity, which led to optical loss.
We have refined the process developed by Haberer et al.11 to reproducibly fabricate GaN microdisks with smaller diameter, smooth sidewalls, good circularity, and uniformity across the sample, leading to high-quality (Q) modes and extremely low thresholds. Using the material structure shown in Fig. 1, we have made changes in the processing that have resulted in dramatically improved device performance. Our microdisk lasers operate at room temperature under CW operation. Their threshold of about 300 W cm-2 is the lowest reported for GaN-based microdisks by several orders of magnitude9, 12. In fact, our threshold is also many orders of magnitude lower than current optically pumped GaN-based vertical-cavity, surface-emitting lasers13, 14. This is also the first report of CW lasing in a GaN-based microdisk at room temperature. Additionally, the disk shape, diameter and photoluminescence (PL) spectra are very uniform across samples—every disk measured showed distinct, high-Q modes at room temperature under CW pumping.
Figure 1: Epitaxial structure of material used to make microdisks.
Light is confined to the disk region, and the post region is selectively etched to form an undercut structure.
Full size image (29 KB) (29 KB)We will discuss optical data from 1.2-m (Fig. 2) and 8-m devices. As shown in Fig. 2a, an optically isolated, undercut structure has been formed with minimal roughness to the optical cavity. Figure 2b shows a closer view of the sidewall, revealing some roughness and a slightly sloped sidewall caused by resist reflow and mask erosion during dry etching. Figure 2c is a top-down view of the microdisk, showing its circular geometry. The 8-m disks had similarly good circularity and low sidewall roughness.
Figure 2: Scanning electron microscope image of a 1.2-
m microdisk.
![Figure 2 : Scanning electron microscope image of a 1.2-|[micro]|m microdisk.](/nphoton/journal/v1/n1/thumbs/nphoton.2006.52-f2.jpg)
a, Side view of disk. b, Close-up view of sidewall roughness. c, Top-down view of disk, showing circularity.
Full size image (26 KB) (26 KB)An example PL spectrum from an 8-m disk is shown in Fig. 3a. As this microdisk was undercut by <500 nm, only first-order whispering gallery modes (WGMs) are observed. Higher order modes are concentrated spatially closer to the centre and are lossy because of coupling to the post. The mode spacing 
of first-order WGMs can be calculated using = 2/
Rneff, where is the emission wavelength, R is the radius of the disk and neff is the group effective index of refraction. Using R = 4.0 m and neff = 3, the predicted mode spacing is about 2.4 nm, which agrees with the observed spacing of 2.2–2.5 nm. Modes are only observed on the low-energy side of the quantum-well spectrum, most likely because high-energy photons are more likely to be reabsorbed before coupling to cavity modes. This trend is typical for GaN-based microdisks with a quantum-well active region8, 12.
Figure 3: Room-temperature photoluminescence spectra.
a, An 8-m disk. b, A 1.2-m disk (the inset showing a high-Q mode, taken above threshold at 2.8 kW cm-2.
Figure 3b shows a PL spectrum from a 1.2-m disk, with the inset showing a higher resolution spectrum of the peak at 428 nm. The 1.2-m disks show fewer modes than the larger disks, as predicted. These modes have both higher intensity and higher quality than those in the spectrum of the 8-m disks. Also notable is the much lower background of broad emission from the quantum well, indicating a much more efficient decoration of the modes by the photons. As these disks are deeply undercut, with almost no post material remaining, there are several higher order modes present. Finite-difference time-domain (FDTD) simulations were performed in two dimensions to identify the modes visible in the 1.2-m disk spectra.
We believe that the peak at 428 nm is a second-order mode, as shown in Fig 4b. There is a first-order mode visible at 418 nm (Fig. 4a), and there should be no other first-order modes that overlap the low-energy side of the quantum-well spectrum. We speculate that sidewall roughness and ion damage from the dry-etch step make the first-order mode a higher-loss mode, because its electric field is concentrated closer to the surface than the second-order mode. The peak at 428 nm has a high quality factor Q = / of about 3,700, measured below threshold at 270 W cm-2, which is close to the resolution limit of our spectrometer.
Figure 4: FDTD simulation data for the 1.2-m disk.
![Figure 4 : FDTD simulation data for the 1.2-|[micro]|m disk.](/nphoton/journal/v1/n1/thumbs/nphoton.2006.52-f4.jpg)
a, First-order whispering gallery mode at 418 nm. b, Second-order whispering gallery mode at 428 nm. Red corresponds to the highest field density and blue is the lowest field density. The black circle indicates the edge of the microdisk.
Full size image (40 KB) (40 KB)Lasing was observed in several 1.2-m disks under CW operation at room temperature. Power-dependent measurements for the 428-nm mode of the disk from Fig. 3b are shown in Fig. 5. The threshold of this disk is only 300 W cm-2, which is many orders of magnitude lower than previous measurements on GaN microdisks9, 12. Our value of the threshold should be considered more as an upper bound on the threshold than an exact value. We have assumed that all incident pump power is absorbed, but some may be reflected, and we have assumed a gaussian spot shape, although in reality the laser spot shape varies over time and can be almost uniform. We have also assumed that about 4% of the power measured at the laser reaches the sample, which was estimated by placing a power meter where the sample would be. No heating effects were observed over the course of these measurements. We believe that this is the first observation of room-temperature CW lasing in GaN-based microdisks8, 9, 11, 12.
Figure 5: Power-dependent data for 1.2-m disk.
![Figure 5 : Power-dependent data for 1.2-|[micro]|m disk.](/nphoton/journal/v1/n1/thumbs/nphoton.2006.52-f5.jpg)
a, Peak intensity versus pump power for two modes: the lasing mode at 428 nm, which exhibits a clear threshold at about 300 W cm-2, and another mode at 432 nm. b, PL spectra taken at different pump powers.
Full size image (22 KB) (22 KB)We attribute the drastic reduction in threshold observed in these microdisks to several factors. A SiO2 hard mask was used instead of a photoresist, leading to decreased mask erosion during the dry-etch step and thus smoother sidewalls. Resist reflow was performed to further smooth the sidewalls, which resulted in less vertical sidewalls, but smoothness appears to be more important than verticality. Because the WGMs are concentrated very close to the surface, especially in small disks, sidewall smoothness is very important in attaining low-loss modes. Electron-beam lithography was used rather than optical lithography, even for large disks, leading to improved circularity of the disks, which is also beneficial for achieving high-Q modes and a low threshold15.
Finally, electron-beam lithography allows fabrication of much smaller disks than those previously studied, resulting in fewer modes overlapping the quantum-well emission than in larger disks. In 8-m disks, there are approximately 25 first-order modes overlapping the quantum-well spectrum, but in the 1.2-m disks, there are only two. Disks on the order of 4 m, such as those studied by Haberer et al.11, 12 should have about 13 first-order modes; therefore, an increased spontaneous-emission factor caused by a reduced number of modes most likely contributed to the dramatic improvement in the threshold we observed. Compared to the PL of the 8-m disk shown in Fig. 3a, the 1.2-m disk in Fig. 3b shows a dramatically reduced PL background, in addition to fewer modes with higher Q and intensity. The PL for the 4.7-m disks described in ref. 12 showed a spectrum very similar to that of Fig. 3a. Although we photo-pumped the entire area of the microdisk, the carriers generated in the periphery of the microdisks have a much higher likelihood of radiative recombination and coupling with WGMs. Carriers generated in the central region of the disk are unlikely to decorate the highest Q modes, particularly for InGaN, which has a short non-radiative recombination time. The smaller diameter disks have a relatively smaller contribution from this central region, as can be seen in the PL spectra of Fig. 3a and b. On the other hand, the surface recombination velocity of InGaN has been reported to be about 103 cm s-1 (ref. 16), an order of magnitude less than that of InGaAsP, and several orders of magnitude less than that of GaAs, further favouring high performance in small-diameter microdisks. Finally, our measurements of Q are carried out at the limit of our instrumental resolution. Higher precision measurements may reveal higher values of Q. Further characterization of our InGaN microdisks is required to better understand the ultimate limitations of lasing performance in these devices.
We have developed a process to fabricate reliably optically isolated, undercut GaN microdisks with small size, smooth sidewalls and good circularity using electron-beam lithography and PEC etching. These microdisks show photoluminescence spectra in good agreement with theoretical predictions made using FDTD simulations. The smallest microdisks fabricated lase at room temperature and under CW operation, with thresholds much lower than any previous report of lasing in a GaN microdisk.
Methods
Fabrication
The epitaxial structure used in this study is shown in Fig. 1. The quantum-well composition was chosen for emission at 420 nm. Microdisks with diameters ranging from 1.2 m to 8 m were patterned using electron-beam lithography and ZEP520 resist. Resist reflow was performed at 160 °C for 1 min. CHF3 reactive-ion etching was used to etch a 50-nm SiO2 hard mask, and then Cl2 reactive-ion etching was used to etch down to the GaN template, resulting in pillars of height 300 nm. Then 50 Å Ti and 300 Å Pt were deposited around the pillars, leaving 20-m concentric circles unmetallized.
The disks were then undercut using bandgap-selective PEC etching, as described previously11, 17. A 1000 W Xe lamp was used to illuminate the sample with light above the bandgap of the post layers, generating electron–hole pairs in this material, but below the bandgap of the GaN and AlGaN layers. The electrons are extracted through the metal, but the holes travel to the surface of the post region, allowing etching to occur in a 0.004 M HCl electrolyte solution. As the metal is placed only around the pillars and not on top, electron–hole pairs generated in the quantum wells recombine before reaching the cathode so that there is no etching of the quantum-well layers. In this way, we form an optically isolated, undercut structure.
Measurements
Photoluminescence measurements were performed at room temperature using a 325-nm CW He–Cd laser. The spot size was approximately 3 m, and power densities were calculated assuming a gaussian spot shape.
Simulations
From two-dimensional (2-D) FDTD simulations of the layer structure of the disk, we calculate an effective index of refraction of 2.33 for our layer structure. Although difficult to measure, the group effective index should be 20–30% higher than the effective index18, so we have chosen neff = 3, in agreement with our results. We also note that the 2-D mode profile simulations for small disks are very sensitive to the choice of n. We have chosen to use 2.30 instead of 2.33 in our simulations because there will be further overlap of the modes with air at the periphery of the disk, lowering the effective index, and the choice of 2.30 matches several modes in the spectrum for 1.2-m disks.
