Effects of Post-Etch Microstructures on the Optical Transmittance of Silica Ridge Waveguides

Abstract

Silica waveguides are often etched by reactive ion etch (RIE) processes. These processes can leave residual topography that can increase optical loss. We investigated the relation between optical loss and various RIE etch. A wet etch step meant to remove microstructures was also considered and compared. Ridge waveguides were fabricated in plasma enhanced chemical vapor deposited films by three different RIE processes, each with a different gas composition, pressure setting, and applied power setting. Half of each set of waveguides were also subjected to a hydrofluoric acid (HF) solution. The waveguides were tested for optical transmission via the cutback method. The transmission vs waveguide length measurements were plotted to fit an exponential curve and the optical loss and measurement uncertainty for each waveguide set was calculated. Clear distinctions in optical loss were found between the different RIE processes. The HF treatment also has an effect, significantly reducing optical loss for two processes and increasing it for the third. Of the tested RIE processes, one can be suggested for silica waveguides. It results in the lowest optical loss and coincidently has the fastest etch rate.

I. Introduction

IN integrated photonics, MEMS, and MOEMS, silicon dioxide (silica) and silicon nitride films are often the basis for optical waveguides, especially in applications that require visible wavelengths. Such applications include lab-on-a-chip biophotonics sensors [1–3], photonic integrated circuits (PICs) [4], and Light Detection and Ranging (LIDAR) functions [5].

The fabrication of silica and silicon nitride waveguides usually requires the layered deposition of these materials on a substrate. Possible methods for film deposition include sputtering and chemical vapor deposition (CVD). Plasma-enhanced chemical vapor deposition (PECVD) is a specific type of CVD that uses plasma to reduce the temperature threshold for the required deposition reaction. It offers dense layers with great flexibility in refractive index and internal stress control [6]. This paper will concentrate on waveguides fabricated using PECVD films.

Once material layers are deposited on a substrate, desired waveguide structures are formed by etching away portions of the layer and leaving a designed-for cross-section. One option for etching is to submerge a masked film into a liquid etchant solution. In the case of PECVD films, this will result in cross-section profiles with concave walls due to isotropic etching. Dry etching methods, such as reactive ion etching (RIE) and deep reactive ion etching (DRIE), use a plasma to both chemically react with and physically bombard the material. This allows for greater control in etch rate and results in an etch profile with straight walls, which is typically closer to the intended cross-section profile. However, RIE can risk damage to the structures from the ion bombardment. As shall be shown, it also can leave behind microstructures in the etched field, an effect that has ramifications for waveguide performance.

Scattering loss is a major contributor to total waveguide loss. Random geometric variations in a waveguide’s sidewalls increase scattering losses along the waveguide’s length [7]. Ridge waveguide is also aggravated by residue topography adjacent to the ridge. When a cladding oxide layer is placed on top of the waveguides and field, the randomly-changing effective index has an increased effect on the overall loss.

Post-etch microstructures and their effect on the optical behavior of silica waveguides is the focus of this work. Ridge waveguides, made by various RIE processes and representing three major types of residual microstructures, were measured to determine how different microstructure formations affect optical loss. Further, a wet-etch supplement meant to remove microstructures was applied to some waveguides during the fabrication process to observe its effects on optical loss. While there are RIE and DRIE processes that mitigate the formation of microstructure defects [8], the purpose of this paper is to examine the effect that each type of defect has on a waveguide’s optical properties.

II. Waveguide Fabrication

The silica ridge waveguide used in this study were fabricated using a process outlined in Fig. 1. Bottom guiding was provided by anti-resonant reflecting optical waveguide (ARROW) layers deposited through sputtering directly on the silicon substrates [9,10]. Waveguides were then formed by etching PECVD silicon dioxide deposited over the ARROW layers. This silica layer had a refractive index of 1.51. To pattern the ridge waveguides, a metal mask was deposited onto the silica layer by a photolithographic process. The metal mask was composed of a chromium/aluminum stack, but we have found that nickel may also be used with similar results.

Fig. 1.

Overview of the waveguide fabrication process. (a) The process begins with a substrate of silicon with reflective dielectric films on top. (b) A layer of PECVD silica with a refractive index of 1.51 is grown over the reflective films. (c) A metal mask is put on top of the silica by photolithographic patterning. (d) The silica is etched down by RIE process, leaving the ridge as protected by the mask. (e) The metal mask is stripped away. (f) After a dehydration bake, the waveguides are cladded with a layer of silicon dioxide with a refractive index of 1.45.

Using a Trion Technology® Minilock Phantom III RIE/ICP, the silica layer for each wafer was etched to form ridge waveguides of 4μm in height. The RIE process parameters are outlined in Table 1. Wafers that also underwent a wet etch treatment were dipped and agitated in a 5% HF solution for four seconds and afterwards were rinsed and sonicated for two minutes; further details about this wet etch treatment are discussed in the section below. The metal etch mask was then stripped away and the silica ridges were placed in a long dehydration bake, up to 300°C, to clear out any water that may have absorbed into the layer. Immediately after the bake, a cladding layer was deposited over the waveguides. This cladding layer was also PECVD silicon dioxide, but of a lower refractive index than the prior layer, measured at 1.45.

TABLE I.

RIE ETCH PROCESSES WITH RESPECTIVE RESIDUE MICROSTRUCTURES

Etch Parameters	O2/CHF3 (nanotubes)	O2/CHF3 Half Power (“grass”)	CF4 (“stalks”)
Chamber Pressure	19mT	19mT	12mT
Gas 1 (flow rate)	O2 (9sccm)	O2 (9sccm)	CF4 (50sccm)
Gas 2 (flow rate)	CHF3 (125sccm)	CHF3 (125sccm)	-
ICP Power Setting	175W	88W	270W
RIE Power Setting	70W	35W	75W

III. Topographical Microstructure Residue Origins

The topographical microstructures formed in a plasma dry-etch take different forms depending on the specific etch parameters used. When etching a silicon dioxide PECVD film, these microstructures consist of silicon dioxide that did not etch away during the etch process. Some structure types are the result of a subtractive micro-masking effect or an additive self-organizing effect. There are applications where these residue microstructures are intended and useful, such as bonding a glass substrate to another material [11,12]. However, as explored here, they can have an adverse effect on waveguide transmittance.

This paper will describe three residue microstructure types, each formed by differing RIE parameters. These microstructure types have been observed by a variety of research groups [11–13] and we have adjusted our specific inductively coupled plasma (ICP) RIE system (Trion Technology® Minilock Phantom III) in order to produce each type with the etch parameters summarized in Table 1. Scanning electron microscope (SEM) images of the different types of etch residue are shown in Fig. 2.

Fig. 2.

SEM images of structures etched by various processes. (a) Nanotube residue topography from the O₂/CHF₃ process; cross section of ridge also shown. Brightness has been increased by 40% and contrast increased by 20% from original image. (b) Silica grass residue topography from the “Half Power” process; cross section of ridge also shown. Brightness has been increased by 30% and contrast increased by 15% from original image. (c) Silica stalk residue topography from the CF₄ process. Brightness has been increased by 15% and contrast increased by 5% from original image.

The first type of residue is produced by an O₂/CHF₃ process and results in the formation of silica nanotubes, less than a micron in diameter (Fig. 2(a)). Utilizing a plasma from a mixture of oxygen and CHF₃ gases, the nanotubes are the result of a redeposition of silica material around a nucleation site. As the etch progresses, more material adds to the nucleation site, eventually forming either tubes or pillars with a deep hole in the center.

A second type of residue microstructure can be formed by maintaining the same process parameters that caused the nanotubes, except the ICP and RIE power settings are halved (referred to here as the “Half Power” process). This reduction in power changes the micro-structuring process from an additive process of etched material recombining to a subtractive process. In this case, when a metal mask is used, some of the metal sputters off and into the field. This creates many new micro-masks that prevents the silica underneath from etching away. The result is a silica “grass” formation that is very dense in some areas (Fig. 2(b)). The change in microstructure formation based around a change in power is consistent with previous findings [11,12].

The third type of residue microstructure is found using completely different etch gases and power settings from the first two cases. Using only CF₄ and a high ICP/RIE power setting, the etch process forms silica stalks (Fig. 2(c)). These stalks appear taller and are much less dense than the silica grass formation. They have diameters of approximately 0.3μm. Each stalk appears to grow out of a hole in the material, so it is possible that it is the result of an additive process from material displaced from a high-energy ion bombardment in the plasma. Based on the relative sparseness of silica stalks, we conjectured that they will have a smaller effect on waveguide loss than the previous formations as they have smaller diameters than the nanotubes and occur less frequently along the waveguide length than the grass.

Additionally, the RIE process can be supplemented with a wet etch treatment to reduce the number and size of the residue microstructures. After the dry etch is complete, waveguide structures can briefly be dipped into an etchant solution and agitated before being removed and rinsed. An option for the wet etch is a 5% hydrofluoric acid (HF) solution used for slow-etching silica. The dilute concentration should be sufficient to etch small microstructures while leaving the waveguide features largely intact. With only four seconds of wet etching, thicker microstructures such as nanotubes are not completely removed, but the grass and stalk formations are.

The result of this treatment is shown in Fig. 3. The nanotube structures from the O₂/CHF₃ process partially etch away, but they are thick enough to still remain after the HF treatment, as shown in Fig. 3(a). The grass microstructures from the “Half Power” for the most part clear away, but some of their remains clump together and stick to the sides ridge structure, as shown in Fig. 3(b). The stalk microstructures from the CF₄ process etch away completely, as seen in Fig. 3(c). As will be shown, when the waveguide features are subsequently coated with a cladding layer, the result is a cleaner cross-section profile that is more uniform across the ridge’s length.

Fig. 3.

These SEM images show the ridge waveguides of each dry etch process under study after the four-second HF treatment is applied. They are (a) the O₂/CHF₃ process, (b) the “Half Power” process, and (c) the CF₄ process.

Fig. 4 shows the cross section of the ridge waveguides when the cladding layer is grown over the devices. Fig. 4(b) is an image taken by SEM to show a ridge waveguide with the cladding layer deposited on top of grass microstructures, leaving a cross-section cluttered near the guiding ridge. Fig. 4(c) alternatively shows the cross-section of a waveguide that has been treated with a 5% HF solution prior to cladding. The grass and stalk microstructures result in a cleaner post-clad profile. However, nanotube structures did not completely etch away, and so they remained when the cladding layer was deposited.

Fig. 4.

(a) Illustration of a waveguide cross section with the cladding layer grown over residue silica grass. (b) SEM image of a cross section of a cladded ridge waveguide with grass residue in close proximity to the waveguide core. Brightness has been increased by 40% from original image. (c) SEM image of a cross section of a ridge waveguide that has been treated with a 5% HF solution before being cladded. Brightness has been increased by 40% and contrast by 20% from original image.

IV. Optical Characterization

To compare the residue microstructures’ effect on optical transmittance, a variety of ridges were etched by the different processes and optically characterized. The same type of etch mask was used for waveguides constructed on different wafers, but each wafer was subjected to a different plasma dry-etch process. The wafers were then cleaved in half. One half of each wafer was given the HF treatment described above. Both halves of each wafer were then given a cladding layer.

A. Optical Transmission Measurement

Each wafer was cleaved into chips that contain several waveguides of various core widths. Each chip had a set of five 3-μm wide ridges, a single 6-μm wide ridge, and a set of five 10-μm wide ridges. The chips were cleaved so that each waveguide is approximately two centimeters long, with some variation. Each chip was placed on an optical bench setup. One end of the chip was placed so that the edge was at the focal point of an objective lens that transmits the image of the chip to an optical power photodetector. A fiber waveguide was placed on the other end of the chip. This fiber guided light from a 635nm laser source (Thor Labs® Model S1FC635) to the ridge facets on the chip. The chip and fiber were aligned with the objective and photodetector.

When alignment was completed, the fiber was coupled with each ridge waveguide and the laser was engaged. When the optimal fiber coupling placement was found, the laser power was turned up to 1mW. The optical power was measured by the photodetector and recorded as a P(x) value. Then the chip was removed and the fiber waveguide was placed so its facet was in the objective’s focal point. At the 1mW setting on the source, the power was measured by the photodetector and recorded as P₀. The chip was measured to record the waveguide lengths. Utilizing the cutback method, we cleaved the chip and repeated the alignment and optical power measuring process at several different lengths.

The above process was repeated until we measured at least five different 6μm ridges and fifteen different 3μm and 10μm ridges. Each P(x) value was divided by its corresponding P₀ value in order to produce a transmission value (T). On a scatter plot, each T value was plotted by its corresponding waveguide length. Because of an inherent inconsistency in how clean a facet is after cleaving, the highest transmission values were selected for each waveguide length. These values were used to create a curve fit to calculate the waveguide loss (cm⁻¹).

B. Propagation of Uncertainty

To account for error in the measured loss values, a sample ridge waveguide of each width was measured ten times for its transmission value. The P(x), P₀, and waveguide length (x) values were compiled. The means, standard deviations, and variances of each figure were used to calculate the expected error for the corresponding loss value.

We can relate the optical power transmission, waveguide length, and loss (α) by an exponential equation,

$$T=\frac{P_{\mathit{\text{mean}}}\left(x\right)}{P_{0,\mathit{\text{mean}}}}=A{e}^{-\alpha x}.$$ (1)

The factor A is calculated in the curve fitting and accounts for the amount of light that is coupled into the ridge waveguide and the facet transmissions, leaving the loss value free for analysis. The equation can be rewritten to bring the loss value to the left-hand side.

With this equation, we can calculate the partial differentials for each factor and write the following equation for the propagation of uncertainty in the loss.

$$\begin{gathered} {\sigma }_{\alpha }^2={\left(\frac{\partial \alpha }{\partial x}\right)}^2{\sigma }_{\alpha }^2+{\left(\frac{\partial \alpha }{\partial P_{\mathit{\text{mean}}}\left(x\right)}\right)}^2{\sigma }_{P\left(x\right)}^2+ \\ 2\left(\frac{\partial \alpha }{\partial x}\right)\left(\frac{\partial \alpha }{\partial P_{\mathit{\text{mean}}}\left(x\right)}\right){\sigma }_x{\sigma }_{P\left(x\right)}+{\left(\frac{\partial \alpha }{\partial P_{0,\mathit{\text{mean}}}\left(x\right)}\right)}^2{\sigma }_{P_0}^2+ \\ 2\left(\frac{\partial \alpha }{\partial P_{\mathit{\text{mean}}}\left(x\right)}\right)\left(\frac{\partial \alpha }{\partial P_{0,\mathit{\text{mean}}}\left(x\right)}\right){\sigma }_x{\sigma }_{P_0} \end{gathered}$$ (2)

The respective partial differentials of (1) can be substituted into (2). By substituting the mean, measurement error, variance, length, and A values, we found the variance and expected error for the calculated loss of a given waveguide [14]. As shown in Fig. 5 below, the expected error was added and subtracted from the loss value gained from the transmission fit line to predict highest and lowest transmittance expected for a given waveguide.

Fig. 5.

Transmission plots of 10-μm ridges ridges etched with the CF4 process. The x markers indicate the transmission of a waveguide measured at a given length in the cutback method. The solid line is the fit line to those markers from which a loss value can be calculated. The dashed lines represent the adjusted transmission line if the expected loss error is added and subtracted from the calculated loss. (a) Waveguides not treated with HF. (b) Waveguides that are not treated with HF.

V. Comparison Of Waveguide Losses

After testing several waveguides of various widths and various etch processes, we can compare the losses to determine an optimal process. For each set of waveguide widths with the same etch process, the 10μm ridges exhibit the lowest losses and the 3μm ridges exhibit the highest losses. This is consistent with expectations. The complete set of found losses and errors is listed in Table 2.

TABLE II.

SUMMARY OF OPTICAL LOSSES AND EXPECTED ERRORS BY RIDGE WIDTH AND ETCH PROCESS

Ridge Width	Optical Loss (cm-1)	O2/CHF3		O2/CHF3 Half Power		CF4
		Without 5% HF	With 5% HF	Without 5% HF	With 5% HF	Without 5% HF	With 5% HF
3μm	Loss	1.585	1.411	1.9	2.925	1.465	0.705
	Standard Deviation	0.0625	0.0166	0.0226	0.1986	0.2958	0.062
6μm	Loss	1.448	1.02	0.886	1.696	0.807	0.619
	Standard Deviation	0.0394	0.0594	0.0455	0.3192	0.0571	0.1178
10μm	Loss	0.722	0.685	0.735	0.913	0.443	0.356
	Standard Deviation	0.0688	0.0284	0.0964	0.23	0.1069	0.0121

The residue microstructures left by the different dry etch processes had an observable effect on the transmission and loss of each waveguide as well. For each waveguide width, the CF₄ process with the stalk microstructure had the lowest loss, the “Half Power” process with the grass residue had the highest loss, and the O₂/CHF₃ process with its nanotube surface roughness was in between.

The application of an HF treatment to the waveguides prior to the cladding oxide layer growth also had an effect on the loss and transmission of the waveguides. The treatment lowered the loss and raises the transmission of the waveguides etched by the O₂/CHF₃ and CF₄ processes, but it raised the loss and lowered the transmission of the “Half Power” waveguides. Fig. 5 gives a comparison of the 10μm ridges as measured for transmission as a function of waveguide length. Fig. 5 also gives the transmission lines when the loss error is factored into the fit line. Fig. 6 gives a comparison of all measured loss values across all waveguide widths and etch conditions studied in this work. For comparison, we are aware of a silica waveguide with a loss as low as 0.02 dB/cm (0.005 cm⁻¹) [15]. It is a similar structure, 5μm high, 5–6μm wide, made of phosphorus-doped silica, and cladded with undoped silica.

Fig. 6.

Summary of loss values by process and ridge waveguide width.

In Fig. 7 we can see the mode profile of the “Half Power” waveguides not treated with HF was much more confined than the mode profile of the corresponding waveguides that were treated with HF. This corresponds with the higher losses exhibited by this set of “Half Power” waveguides with the HF treatment compared to those “Half Power” structures not treated. We offer conjecture, based on SEM imaging, that when the “grass” microstructures were cleared away by the HF treatment, some of their debris clumped together on the sidewalls of the ridges, as can be seen in Fig. 3(b). It is possible this contributed to sidewall scattering, resulting in a worse outcome for loss than had the treatment not been applied. However, further examination is required to determine the exact cause of the increase in loss in this case and is beyond the scope of this comparative study.

Fig. 7.

Mode profiles of 3-um ridges etched by the “Half Power” process, viewing cleaved facets. (a) Mode profile of a 3-μm “Half Power” ridge not treated with HF. Brightness has been increased by 20% from original image. (b) Mode profile of a 3-μm “Half Power” ridge that has been treated with HF. Brightness has been increased by 20% and contrast increased by 40% from original image.

VI. Conclusion

Surface roughness from the etch residue topography causes increased loss by inserting changes to the waveguide’s effective index, and the repeated changes increases waveguide loss. This greater loss decreases the transmittance that can be utilized by a variety of applications. Exploring different etch processes can provide alternatives with different optical losses in the waveguides. Also, applying a hydrofluoric acid treatment has an effect on the optical loss, with an improvement in the waveguides that featured nanotube and stalk microstructures.

Overall, the best process found in this study for fabricating waveguides was the CF₄ RIE process that utilized the 5% HF treatment. The lowest loss and the fastest etch time make this method the overall best choice in this sample set for fabricating such waveguides.

Biographies

Joel G. Wright, Jr. (S’15) received the B.S.E. degree in electrical engineering from Arizona State University, Tempe, AZ, USA in 2015. He is a Ph.D. candidate at Brigham Young University in the Electrical and Computer Engineering Department. His current research includes the fabrication of optics-based biosensors and computer modeling such devices.

Holger Schmidt (F’17) received his Ph.D. in electrical and computer engineering from the University of California, Santa Barbara. He served as a postdoctoral fellow with MIT. Currently, he is the Narinder Kapany Chair of optoelectronics, a professor of electrical and computer engineering and the associate dean for research with the Baskin School of Engineering, University of California, Santa Cruz. He has authored or coauthored over 400 publications and is a Fellow of the OSA, IEEE, and NAI.

Aaron R. Hawkins (F’16) received a B.S. degree from Caltech and a Ph.D. degree from the University of California, Santa Barbara. He was a Co-founder of Terabit Technology and an engineer at CIENA and Intel. He is currently a Professor with the Electrical and Computer Engineering Department, Brigham Young University, doing research in optofluidics, integrated optics, and MEMs. He has authored or coauthored over 400 technical publications and is a Fellow of the IEEE and the OSA. He has served as the Editor-in-Chief for the IEEE Journal of Quantum Electronics and currently serves as the IEEE Photonic Society’s VP of Publications.