We report low-loss deep-etch AlGaAs optical waveguides fabricated with nitrogen plasma-assisted photoresist reflow. The simultaneous application of a nitrogen plasma and heat is used to reduce the line edge roughness of patterned photoresist and limit the lateral spread of the photoresist patterns of submicron-scale waveguides. Comparison of the edge roughness of the etched sidewalls between the as-developed and smoothed photoresist etch samples show a reduction of the RMS roughness from 3.39±0.17 nm to 1.39±0.03 nm. The reduction in propagation loss is verified by measured waveguide loss as a function of waveguide widths. A 0.65-μm wide waveguide with a modal area of 0.4 μm2 is fabricated with a propagation loss as low as 1.20±0.13 dB/cm for the transverse-electric mode.
© 2014 Optical Society of America
AlGaAs waveguides are a promising platform for nonlinear optical processing applications, with advantages such as a large optical Kerr coefficient, the flexibility to engineer the bandgap energy to achieve broadband transparency throughout the telecommunication band, and the potential for integration with high-speed electronics. Several groups have reported on the third-order nonlinear effects in AlGaAs waveguides [1–4], but the performance of these waveguides is typically limited by propagation loss, which can be as high as 80 dB/cm for sub-square-micron waveguide mode areas. This is technologically important for sub-micron scale high-index contrast waveguides required for high-efficiency nonlinear process and group velocity dispersion (GVD) control .
The main source of propagation loss in high-index contrast waveguides is often the scattering loss due to the sidewall roughness. Several techniques have been proposed to reduce the side-wall roughness in AlGaAs waveguides such as using wet thermal oxidation  or using silicon index loading . Although the sub-micron scale AlGaAs waveguides with silicon index loading exhibits low loss, it is not suitable for nonlinear optical processing application due to the high two-photon absorption loss in silicon. Ultimately, the rough texture and vertical striations in the etched waveguide sidewalls are caused by texture in the developed photoresist, which is used to define the waveguide as an etch mask. The resulting edge roughness of the photoresist is an inherent feature of the specific photoresist chemistry. For example, photoresists that are photo-active in the deep-UV for resolving extremely small features for the CMOS industry have a very small edge roughness. These tools have been used to create low-loss features in silicon by CMOS manufacturers interested in silicon photonics. In contrast to these deep-UV photoresists, Novolac based photoresists that are used with i-line mercury vapor line tools have been available for years. The minimum resolvable features are sufficient for the fabrication of sub-micron scale waveguides, but the line edge roughness (LER) of Novolac-based photoresists is a limiting factor of waveguide performance.
Photoresist reflow is a technique that has been proposed for making low-loss waveguides. When patterned photoresist is heated above its glass transition temperature, Tg, it softens. While hot, the photoresist can reflow, that is the surface of the photoresist becomes much smoother but the starting rectangular profile simultaneously spreads laterally to form a dome-shaped profile. The tapered dome-shaped profiles have been used for shallow etches [8–10] and forming micro lens arrays . Unfortunately, they are not useful as an etch mask for deep-etch sub-micron scale waveguides because the tapered edges of the spread photoresist causes a change in the masked area as the photoresist is consumed during the dry etch processes.
The plasma-assisted process described here is a post-lithographic process that reduces the LER of deep trench features and limits the lateral spreading of the photoresist during smoothing. This plasma assisted reflow process produces a photoresist pattern that is greatly smoothed and allows for sub-micron-sized features that are useful as etch masks for trench etches created with inductively coupled plasma reactive ion etching (ICP-RIE) in semiconductor AlGaAs heterostructures. Significant improvement in smoothness for the plasma-assisted reflow compared to the as-developed photoresist sample is demonstrated from the measured root mean square (RMS) roughness, σ, of the waveguide sidewall and the measured propagation loss of waveguides.
2. Waveguide design
The waveguide epitaxial structure is grown using solid-source molecular beam epitaxy (MBE) on a GaAs substrate and consists of a 0.8-μm-thick Al0.18Ga0.82As core layer surrounding by 0.2-μm-thick Al0.7Ga0.3As upper cladding and 2-μm-thick Al0.7Ga0.3As lower cladding. Waveguides mask features ranging from 0.4-to-2 μm-wide are projected onto the photoresist by a 5x i-line stepper. Every waveguide has a 2.2-μm-wide flare at both facets to improve coupling efficiency. Processing bias produces deep-etch waveguides with a minimum width of 0.65-μm. Figure 1 depicts a 0.65-μm-wide AlGaAs ridge waveguide superposed by the calculated transverse electric (TE) mode contours. The effective area of this waveguide is estimated to be 0.4 μm2, using a full-vector finite-difference method .
3. Waveguide fabrication
Two conventional surface preparations for the photoresist stack are studied here. The first surface preparation uses developable bottom anti-reflection coating (BARC), WiDE-C from Brewer Science. BARC is applied on top of the sample to minimize the standing wave patterns inside the photoresist. The second surface preparation uses the adhesion promoter hexamethyl-disilazane (HMDS) from J. T. Baker. The i-line photoresist, OIR 908-10 from Fujifilm, is then spun on to samples of either the BARC or the HMDS surface prepared chips. Both preparations are completed with a water soluble contrast enhancement material (CEM), CEM 365i from ShinEtsuMicroSi, which is spun onto the pre-exposure baked photoresist samples. The pattern exposure is then performed on a projection aligner. After the photoresist is developed, the waveguide samples are etched to the depth of 1.7 μm using a BCl3 and N2 gas mixture in an inductively coupled plasma reactive ion etch (ICP-RIE) system.
Figure 2 shows scanning electron micrograph (SEM) images of the as-developed photoresist on (a) BARC and (b) HMDS along with the etched ridge waveguides created with (c) the BARC process and (d) the HMDS process. There are several features apparent in Fig. 2 such as the standing wave pattern in the HMDS prepared sample and the vertical striations in the etched sidewalls of both of the AlGaAs heterostructures. For both surface preparations, the etched sidewalls exhibit similar features and noticeable detail with respect to the top of the photoresist and cleaved surfaces. There appears to be no visually obvious significant difference in sidewall roughness texture of the developed photoresist that might be attributed to the different surface preparations. The subsequent ICP-RIE etched features of the AlGaAs ridge produce similar ridges that are 1.1-μm-wide from mask features that are 1.0-μm-wide. The similarities of the features produced with BARC and HMDS confirm that the vertical striations are characteristic signatures caused by the inherent sidewall roughness texture of the photoresist itself and that the BARC does not significantly affect the line edge roughness for the process used here.
4. Photoresist smoothing
Two different techniques are performed to smooth the patterned photoresist before the waveguide samples are etched in the ICP-RIE. First, the conventional reflow of photoresist by heating the sample on a standard hot-plate is demonstrated. Heating the sample with patterned photoresist above its glass transition temperature in air is accomplished on a hot-plate set at 170°C for 2 minutes. SEM images in Fig. 3 exhibit the characteristics of the photoresist following this process on (a) BARC and (b) HMDS, as well as the etched waveguides created with samples prepared with (c) the BARC and (d) HMDS reflow processes. A regular hemisphere of photoresist is observed for the BARC prepared samples as shown in Fig. 3(a). The mounding appears to be attributable to the base of the photoresist staying on the BARC. However, it is difficult to make this consistently repeatable or consistent across the chip field and often results in undulation on the micrometer scale in the photoresist as shown in Fig. 4. This patterned photoresist is not useful as an etch mask for waveguide etching because it transfers the undulation onto etched waveguide and results in significantly increasing the optical loss in the waveguide.
The lateral spread of the HMDS prepared samples to a width of approximately 3.5 μm as shown in Fig. 3(b) has two negative consequences. First, there is loss of the critical dimension (CD) and second, as shown in Fig. 3(d), the tapered ramp edges of the reflowed dome shape performs poorly as trench etch masks. The profile change of the photoresist during the conventional reflow is a function of the length of the polymer molecules, the molecular weight of the photoresist, the surface tension of the photoresist, the interfacial energy between the photoresist and the substrate, and the cross sectional area of the patterned photoresist. In general, the reflowed shape has been observed to range from a regular hemisphere to a flattened mound with tapered edges depending on the staring lateral dimensions of the feature. A regular hemisphere may still be useful as an etch mask in some situations such as shallow etches where little photoresist is consumed during the etch process.
The second reflow technique is plasma-assisted reflow. Using a random search approach, a coarse exploration of parameter space on an existing Oxford PlasmaLab System 100 tool was conducted. During the searching phase, the waveguide mask is used to pattern photoresist on numerous samples, which are then treated with different plasma-assisted reflow processes and inspected with a scanning electron microscope. These images are not presented here, but in brief summary the following trends were observed. Oxygen plasma-assisted reflow tends to etch and roughen the photoresist, Silane plasma-assisted reflow result in nodules of material deposited on the photoresist, Argon plasma-assisted reflow also roughen the photoresist surface.
Only nitrogen plasma-assisted reflow produced photoresist profiles that appear smoother than the as-developed features. With nitrogen, it was found that increasing the ICP power warps and wrinkles the photoresist; pressure (in the range of 10 – 60 mTorr) and DC self-bias voltage (in the range of 0–120 V) have little effect. Time and temperature are the most important parameters. In one extreme, the photoresist is visibly discolored appearing burnt while decreasing temperature and time does not successfully reflow the photoresist.
The samples discussed and shown here are exposed to nitrogen plasma for 8 minutes, which is created with a nitrogen gas flow rate of 20 sccm, in a chamber at 60 mTorr, with an ICP-RF power of 600 Watts, zero DC self-bias voltage and the heated stage set to 210°C.
The SEM images in Fig. 5 show the photoresist following the plasma-assisted reflow on (a) BARC and (b) HMDS, in addition to the etched ridge features created with samples prepared with the (c) BARC and (d) HMDS plasma-assisted processes. The nitrogen plasma-assisted reflow limits the lateral spread of the photoresist during heating from below. For example, photoresist ridges with as-developed widths of 1.1 μm spread to 1.35 μm following this process. This CD penalty has been observed for photoresist lateral widths from 2 μm down to 0.4 μm. As shown in Figs. 5(c) and 5(d), the shape of the plasma-assisted reflowed photoresist is different between the BARC and HMDS samples, but this does not seem to impact the AlGaAs waveguides patterned by either plasma-assisted reflowed photoresist processes as they both have much smoother sidewalls than that of waveguides patterned by the as-developed photoresist.
There are at least two possible interactions of the nitrogen plasma with the photoresist that limit the lateral spread and allow for the observed reduction in the LER of etched trenches; nitrogen may be incorporated into the photoresist or deep-UV emission of the nitrogen plasma could alter the chemical bonds in the photoresist. The degree of cross-linking was not explored, but incorporation of nitrogen into the photoresist during the plasma-assisted reflow process is verified with x-ray photoelectron spectroscopy (XPS). Figures 6(a) and 6(b) illustrate the oxygen and nitrogen concentration of the photoresist in the as-developed state and following the nitrogen plasma-assisted reflow process. Integration of the signal peaks allow compositional analysis to be completed and determines that the as-developed photoresist has an elemental composition of 81% carbon, 18% oxygen, 0% nitrogen and trace amounts of silicon and sulfur, while the nitrogen plasma-assisted reflow sample has an elemental composition of 56% carbon, 27% oxygen, 11% nitrogen, 5% silicon and trace amounts of sulfur. The formation of a nitrogen-rich skin, silicon oxynitride skin, or uniform incorporation is yet-to-be determined. Because chamber conditioning was not performed before reflow, it is possible that the increased silicon composition of the plasma-assisted sample is the result of contamination from the chamber, which is typically used to grow silicon containing dielectric films.
The bulk of the literature documenting plasma interactions with Novolac resins focuses on stabilized resist chemistries in oxygen plasmas. However, insight can be gained with broad comparisons to literature concerning surface reactions between deep-UV resists and plasmas. The role of broad-band deep-UV emission of the nitrogen plasma cannot be discounted [11, 13, 14]. Novolac resins are not suitable for newer lithographic technology nodes such as ArF because of its high absorption coefficient of UV radiation, therefore, any modification resulting from the plasma emission is expected to be in the near surface region. Additionally, the lack of dependence on self-bias voltage indicates that ion-bombardment is most likely not the mechanisms causing the composition change in the resists. Rather, the reactive species of the plasma incorporates into the softened resist in a diffusion limited surface region . For these reasons, it is most likely that the nitrogen restricts the lateral spread during reflow through a skin effect.
5. Line edge roughness
From the perspective of optical loss, both the LER and the correlation length are important parameters [15, 16]. The fabricated AlGaAs waveguides are evaluated with plan-view SEM with the slight tilt of 2° to minimize the imaged extent of the edge profile. Figure 7 shows representative profiles of the sidewall from two AlGaAs waveguides fabricated with the as-developed photoresist and plasma-assisted reflowed photoresist. The sidewall roughness associated with the vertical striations observed in Fig. 2 is evident in the image of the waveguide fabricated with the conventional as-developed photoresist shown in Fig. 7(a). Figure 7(b) shows an Al-GaAs waveguide patterned by the reflowed photoresist process which produces a smoother waveguide sidewall.
The roughness parameters are quantified from these plan-view SEM images that were taken with the InLens detector using an accelerating voltage of 10 kV, and a magnification of 50,000. Each SEM image has a field of view sufficient to contain 2-microns of sidewall length. To insure sufficient resolution of the edge profile, images were digitized with 6.75 megapixels providing a pixel scaling of 0.375 nm/pixel. The line edge profile is extracted from the plan-view SEM images by thresholding each column of the image and using the image scale to convert vertical pixel location to edge position. A linear slope is subtracted from the data to remove the slight angular misalignment of the waveguide edge that exists in the SEM images. Example line edge profiles are shown for waveguides patterned with the as-developed photoresist in Fig. 7(c) and for waveguides patterned with the plasma-assisted reflowed photoresist in Fig. 7(d).
The LER is calculated as the rms-roughness of the line edge profile for each sample. It has been shown that sidewall roughness is best described by height variations that follow exponential statistics . Therefore, the correlation length is determined by calculating the auto-covariance of the line edge profile and performing an exponential fit to the central peak. The correlation length corresponds to the 1/e point of this exponential. The high frequency noise apparent in Figs. 7(c) and 7(d) is pixilation noise from the SEM images. A balance was made to have high resolution for accurate determination of the LER, and a long waveguide length for improved determination of the correlation length. The high frequency noise does contribute to a narrow DC spike in the auto-covariance calculation that does produce errors in the determination of the correlation length. To mitigate this error, the LER and correlation length are calculated for a dozen line edge profiles that are extracted from a dozen SEM images from unique regions from both types of waveguides. The reported LER is the root-mean-square of the LER from the dozen profiles, while the correlation length is the mean value. The reported uncertainty for both parameters is the standard deviation from the twelve values. The reduction in the LER from 3.39±0.17 nm of the as-developed photoresist waveguide to 1.39±0.03 nm for the plasma-assisted reflow photoresist waveguide is evident.
The correlation length of the as-deposited photoresist waveguides is determined to be 57±10 nm. The correlation length of the plasma-assisted reflow process is reduced to 39±7 nm. These results are counter-intuitive, as we expected the smoothing process, that reduces high frequency content to smear the surface as the peaks and valleys coalesce during reflow, to also result in a longer correlation length. To address this observation more thoroughly, additional research is needed to understand the microscopic mechanisms occurring during the plasma-assisted reflow process of i-line photoresist.
6. Linear loss characterization
The propagation loss is measured for a large number of waveguides of different widths using a Fabry-Pérot technique described in . The technique uses the cleaved waveguide facets to form a Fabry-Pérot cavity. To measure loss in the waveguide, a tunable continuous-wave (CW) laser is coupled into the waveguide with a lensed fiber. The transmitted power out of the waveguide is collected using an objective lens and directed to an optical power meter. A polarization controller and a polarizer are used to insure that the launch state is along the transverse-electric (TE) polarization mode of the waveguide. While the transverse-magnetic (TM) mode typically has a lower loss than the TE mode for sub-micron scale AlGaAs waveguides [2,19] because the modes sample the surface roughness differently, the TE mode is selected for plotting in Fig. 8 because it has a lower group-velocity dispersion than the TM mode .
As shown in the inset of Fig. 8, the measured transmitted power exhibits a periodic sequence of peaks associated with the longitudinal modes of the Fabry-Pérot cavity while the input wavelength is scanned between 1550 nm and 1551 nm. The fringe contrast or visibility, κ, is determined from the maximum and minimum transmitted power, Pmax and Pmin, using;20]. The waveguide facet reflectivity is estimated separately to be 32.2±0.5 % for the 2.2 μm-wide flare on every waveguide facet as determined by measuring the propagation loss of waveguides with the same width but different lengths.
Figure 8 compares the TE-polarized propagation loss of these waveguides patterned by the as-developed and plasma-assisted reflowed photoresists. All waveguides measured here are made with BARC. Only the lowest waveguide losses are measured for each width as defects and contamination can cause additional losses that impact yield but not increase the ultimate performance of the waveguide. The primary uncertainty in these measurements originate in the facet reflectivity, not random measurement errors associated with analysis of the optical transmission such as the example shown in the inset of Fig. 8. For wide waveguides, the linear loss is dominated by the material loss. Both photoresist processes produce waveguides with similar characteristics for waveguide widths that are wider that 2-microns and indicate that the material loss is less than 0.2 dB/cm. As the waveguides get narrower, the scattering loss caused by sidewall roughness becomes dominant due to a stronger field intensity interaction with waveguide sidewalls. The solid curves in Fig. 8 show the propagation loss calculated using the model reported in . The waveguide dimensions, material index profile, numerically calculated optical mode, measured RMS roughness and measured correlation length are used as the physical parameters of this model. The propagation loss of a 0.65-μm-wide waveguide patterned by the plasma-assisted reflow process is measured to be 1.20±0.13 dB/cm for TE mode and 1.59±0.17 dB/cm for TM mode, both of which are much lower than loss produced with the as-developed photoresist process. The smoothed photoresist process presented here produces waveguides with measured losses that are among the lowest reported values for AlGaAs waveguides with similar sizes. Table 1 compares the propagation loss of the AlGaAs waveguide studied in this work to various nonlinear AlGaAs waveguides studied in other recently published works.
The nitrogen plasma-assisted reflow smoothing process presented here has practical significance for the fabrication of sub-micron scale deep-etch semiconductor waveguides as shown here with AlGaAs materials. The simultaneous exposure of patterned photoresist to a nitrogen plasma while the sample is heated has been shown to significantly reduce the sidewall roughness and the optical propagation loss at telecom wavelengths.
We thank Dr. Karen Gaskell of the University of Maryland Surface Analysis Center for the XPS data.
References and links
1. J. S. Aitchison, D. C. Hutchings, J. U. Kang, G. I. Stegeman, and A. Villeneuve, “The nonlinear optical properties of AlGaAs at the half band gap,” J. QuantumElectron. 33, 341–348 (1997).
2. G. A. Siviloglou, S. Suntsov, R. El-Ganainy, R. Iwanow, G. I. Stegeman, and D. N. Christodoulides, “Enhanced third-order nonlinear effects in optical AlGaAs nanowires,” Opt. Express 14, 9377–9384 (2006). [CrossRef]
4. W. Astar, P. Apiratikul, B. M. Cannon, B. M. Mahmood, J. J. Wathen, J. V. Hryniewicz, S. Kanakaraju, C. J. K. Richardson, T. E. Murphy, and G. M. Carter, “Conversion of RZ-OOK to RZ-BPSK by XPM in a passive AlGaAs waveguide,” IEEE Photonics Technol. Lett. 23, 1397–1399 (2011). [CrossRef]
6. Y. Lou and D. C. Hall, “Low-loss nonselectively oxidized Alx Ga1−x As heterostructure waveguides,” Appl. Phys. Lett. 93, 261111 (2008). [CrossRef]
8. H. E. G. Arnot, H. P. Zappe, J. E. Epler, B. Graf, R. Widmer, and H. W. Lehmann, “Extremely smooth sidewalls for GaAs/AlGaAs ridge waveguides,” Electron. Lett. 29, 1131–1133 (1993). [CrossRef]
10. T. Ling, S.-L. Chen, and L. J. Guo, “Fabrication and characterization of high Q polymer micro-ring resonator and its application as a sensitive ultrasonic detector,” Opt. Express 19, 861–869 (2011). [CrossRef]
12. A. B. Fallahkhair, K. S. Li, and T. E. Murphy, “Vector finite difference modesolver for anisotropic dielectric waveguides,” J. Lightwave Technol. 26, 1423–1431 (2008). [CrossRef]
13. E. Pargon, M. Martin, K. Menguelti, L. Azarnouche, J. Foucher, and O. Joubert, “Plasma impact on 193 nm photoresist linewidth roughness: Role of plasma vacuum ultraviolet light,” Appl. Phys. Lett. 94, 103111 (2009). [CrossRef]
14. G. S. Oehrlein, R. J. Phaneuf, and D. B. Graves, “Plasma-polymer interactions: A review of progress in understanding polymer resist mask durability during plasma etching for nanoscale fabrication,” J. Vac. Sci. Technol. B 29, 010801 (2011). [CrossRef]
15. T. Barwicz and H. A. Haus, “Three-dimensional analysis of scattering losses due to sidewall roughness in microphotonic waveguides,” J. Lightwave Technol. 23, 2719–2732 (2005). [CrossRef]
16. F. Ladouceur, J. D. Love, and T. J. Senden, “Effect of side wall roughness in buried channel waveguides,” IEE Proc. Optoelectron. 141, 242–248 (1994). [CrossRef]
17. J. A. Ogilvy and J. R. Foster, “Rough surfaces: gaussian or exponential statistics?” J. Phys. D Appl. Phys 22, 1243–1251 (1989). [CrossRef]
18. E. Kapon and R. Bhat, “Lowloss singlemode GaAs/AlGaAs optical waveguides grown by organometallic vapor phase epitaxy,” Appl. Phys. Lett. 50, 1627–1630 (1987). [CrossRef]
19. C. Lacava, V. Pusino, P. Minzioni, M. Sorel, and I. Cristiani, “Nonlinear properties of AlGaAs waveguides in continuous wave operation regime,” Opt. Express 22, 5291–5297 (2014). [CrossRef]
20. C. M. Herzinger, C. C. Lu, and T. A. Detemple, “The semiconductor waveguide facet reflectivity problem,” J. QuantumElectron. 29, 2273–2281 (1993).
21. K. Dolgaleva, W. C. Ng, L. Qian, J. S. Aitchison, M. C. Camasta, and M. Sorel, “Broadband self-phase modulation, cross-phase modulation, and four-wave mixing in 9-mm-long AlGaAs waveguides,” Opt. Lett. 35, 4093–4095 (2010). [CrossRef]