2.1.

Spin Coating and Pre-Exposure Bake Parameters

HSQ thin films are obtained by spin coating 4% HSQ (Dow-Corning XP1541-4) in methyl isobutyl ketone onto Si substrates and silicon on insulator (SOI, Soitec) substrates with a top Si thickness of 250 nm and buried oxide (BOX) thickness of 1 μm. Si substrate samples are used to study the effects of fabrication parameters on HSQ mask LER, while SOI samples are used for optical propagation loss measurements. Comparisons of the etch profiles of Si and SOI samples show no discernible difference in dimensions or SWR. Spin coating the samples at 3000 rpm results in approximately 80 nm of HSQ. Immediately after spin coating, samples are baked for 40 min in a ${\mathrm{N}}_2$ purged oven at temperatures ranging from 50 to 200°C. Additional samples are baked for 40 min at room temperature in a YES III HMDS Vapor Prime/Vacuum Oven. After pre-exposure baking, samples are loaded into a Vistec EBPG 5000 e-beam lithography tool for patterning at 100 kV acceleration voltage.

2.2.

E-Beam Writing Parameters and Development

Exposed patterns consist of straight lines, rings with bus waveguides, and tapered width sections. Lines and rings have a designed width of 450 nm. Rings with bus waveguides have coupling gaps between 100 and 500 nm. Tapered width sections vary in width from 60 to 450 nm. We create ring resonator waveguides with radii from 2.5 to 10 μm as a series of 0.5 deg arc segments for 30 deg on either side of the bus coupling section and 1 deg arc segments elsewhere.

The e-beam current, dose, and BSS are related by $I=10(f)(D)({\mathrm{BSS}}^2)$, where $I$ is the current in nA, $f$ is the beam sweep frequency in MHz, $D$ is the per-pass dose in $\mu \text{C}/{\mathrm{cm}}^2$ and BSS is the beam step size in μm. To minimize writing time, the beam sweep frequency is fixed at the machine maximum value of 10 MHz. Similarly, in order to better reproduce curved mask geometries with the rectangular and trapezoidal subsections into which the digital mask file is fractured for e-beam writing, the BSS is fixed at the machine minimum of 5 nm. With these values fixed, there is a linear relationship between the beam current and per-pass dose. In addition, the beam spot size is linked to the beam current. For currents in the range of 625 pA to 3.125 nA, the beam spot size (diameter) varies from 5 to 11 nm, as theoretically determined by the e-beam vendor.

Patterns are written using four-pass writing. In this writing scheme, the beam is scanned over the same region four times, delivering a quarter of the total dose in each pass, before moving to the next writefield. Intentional per-pass offsets are implemented by shifting the pattern file by one BSS between each pass.⁴⁰ Curved structures are rotationally offset by a fraction of the minimum prefractured arc segment length used to make up the curved regions. For ring resonators with 0.5 deg arc segments and four-pass writing with per-pass offsets, each pass used to expose the HSQ is then rotated by 0.125 deg from the previous.

Exposed samples are developed in TMAH. The contrast of TMAH-based HSQ development can be increased through several factors including increasing developer concentration,⁷ increasing developer temperature,^9,10 and by adding salts to the development solution.^11,12,16 Since hot TMAH acts as an effective Si etchant,^10,41 we focus only on the role of developer concentration on the HSQ mask LER. All development is carried out at 25°C. Careful control of development temperature is critical since variations as small as 1°C have been reported to result in significant changes in contrast and onset dose.⁹ Top-side HSQ surface roughness is characterized with an Asylum MFP-3-D AFM.

2.3.

Postdevelopment Processing

After development, samples are hard-baked either on a 350°C hotplate for 2 min, or on a 150°C hotplate for 2 min followed by a 1000°C anneal in an ${\mathrm{O}}_2$ ambient for 1 min in an AGA 410 RTA. The 150°C hotplate bake is used before inserting the samples into the RTA to drive off any remaining solvent and prevent RTA chamber contamination. Si etching is performed in a PlasmaTherm SLR770 Inductively Coupled Plasma-Reactive Ion Etcher (ICP-RIE) with ${\mathrm{Cl}}_2/{\mathrm{O}}_2$ chemistry. For this work, etching conditions are kept constant for all samples in order to isolate the role of the mask LER on the waveguide profile. The HSQ mask is not removed from the samples after etching since it forms a low-loss optical cladding after hardbake.⁵ Half of the samples used for optical propagation loss measurements are treated with a postetch bake at 1000°C in RTA in an ${\mathrm{O}}_2$ ambient for 1 min immediately following etching. We primarily characterize the impact of fabrication parameters through qualitative scanning electron microscope (SEM) analysis of both top-down and cross-sectional images of etched Si patterns on Si substrates using a Zeiss Ultra 55 Plus FE-SEM. Given the established fabrication parameters based on SEM analysis, optical propagation loss measurements are used to confirm the reduced waveguide SWR.

SOI samples used for optical transmission loss measurements are clad with 1 μm plasma enhanced chemical vapor deposition (PECVD) ${\mathrm{SiO}}_2$ in a PlasmaTherm 790 PECVD tool using a ${\mathrm{SiH}}_4/{\mathrm{N}}_2\mathrm{O}$ chemistry. Finally, compact cantilever couplers are fabricated to allow low-loss fiber-to-chip coupling for optical characterization.³⁹

2.4.

Characterization of Etched Waveguides

Although qualitative SEM analysis is a useful tool for rapid prototyping and analysis of recipe parameters, it does not give quantitative information about LER or SWR. Direct quantitative measurement of SWR with conventional techniques such as top-down AFM and SEM is challenging. 3-D AFM is a relatively new technique that allows for direct, nondestructive characterization of the sidewalls of submicron structures.^42–46 Since propagation losses in fully etched silicon strip waveguides are sensitive to variations in SWR that are small compared to the total feature size,³¹ accurate nondestructive analysis with 3-D AFM is a promising technique for the study and optimization of fabrication processes.

Optimized waveguide SWR is characterized both directly through 3-D AFM and indirectly though optical propagation loss measurements. 3-D AFM measurements are performed with a Park Systems NX-3DM which tilts the sample stage by $\pm 38\mathrm{deg}$ to directly measure the sidewalls of the 250-nm tall Si waveguides with a standard high aspect ratio AFM tip.^45,46 The NX-3DM system has a resolution of $<0.2\mathrm{nm}$ and a noise level of $<0.05\mathrm{nm}$. Optical propagation losses are determined through both resonant and nonresonant techniques. Straight waveguide propagation losses are calculated by measuring the total fiber-to-fiber insertion loss for waveguides with lengths ranging from 1 mm to 2 cm. By fitting a line to the increasing loss with waveguide length, propagation losses are calculated by the slope and total coupling losses are calculated from the abscissa intercept. Once fiber-to-chip coupling is established, measured power is stable with $\pm 0.05\mathrm{dB}$ fluctuations in power. Based on this measurement uncertainty, propagation losses fit from multiple waveguide measurements are conservatively estimated to be $\pm 0.1\mathrm{dB}/\mathrm{cm}$. All waveguides with lengths greater than 5 mm include four 30-μm radius 90 deg bends to confirm that light is guided in the Si waveguide rather than in the oxide slab. Losses in ring resonators are determined from critically coupled single-bus rings and from dual-coupled rings. Both the coupling between the bus and ring waveguides and the loss inside the resonator influence the ring resonator transmission response.⁴⁷

3.1.

Pre-Exposure Bake

As the pre-exposure bake temperature is increased from 50 to 200°C, the measured top-side RMS roughness of samples exposed and developed in 2.38% TMAH (Rohm and Haas, MF-CD-26) tends to increase for all tested doses. Samples baked at 50°C have a maximum top-side roughness of 3.1 nm for lower doses ($<600\mu \text{C}/{\mathrm{cm}}^2$) and 1.0 nm for higher doses ($>1000\mu \text{C}/{\mathrm{cm}}^2$). Samples baked at 200°C show roughness values as large as 7.7 nm for lower doses and similarly low roughness of 1.3 nm for higher doses. Samples baked at the intermediate temperatures of 100 and 150°C show a similar decrease in top-side roughness from 3.6 to 1.2 nm and from 3.1 to 1.2 nm, respectively, over the same range of exposure doses. Since using a 50°C pre-exposure bake reduces the maximum top-side roughness for lightly dosed resist and the LER of fully dosed resist is comparable to the top-side roughness of lightly dosed resist,²³ low temperature pre-exposure baking results in minimized LER. Trends in HSQ roughness with exposure dose are expected to be similar for both low and high contrast TMAH development for samples with the same pre-exposure bake temperature.⁷ Pre-exposure bake temperatures lower than 50°C were not further explored since uneven lines were observed on these samples during postetch inspection.

3.2.

E-Beam Writing Parameters and Development

We observe a clear trend in reduced HSQ mask LER with an increased exposure dose from 320 to $5000\mu \text{C}/{\mathrm{cm}}^2$ and the corresponding increase in beam current from 200 pA to 3.125 nA and spot size from 4 to 11 nm. As the dose and beam spot size increase, dimensions of developed patterns tend to increase laterally and thin resist residues form between closely spaced patterns, as shown in the nominally 100 nm gaps in Figs. 1(a) and 1(b). By switching from a low contrast developer, 2.38% TMAH, to a high contrast developer, 25% TMAH, properly sized patterns with clean gaps are produced as shown in Figs. 1(c) and 1(d). Figure 2 provides comparative SEM micrographs showing the impact of increased dose and beam current on waveguide LER. The minimum dose required to produce viable waveguide patterns increases from approximately $800\mu \text{C}/{\mathrm{cm}}^2$ for 2.38% TMAH to approximately $1500\mu \text{C}/{\mathrm{cm}}^2$ for 25% TMAH. For either developer, increasing the total dose to at least three times larger than this minimum value results in a significant decrease of LER. The optimal HSQ LER for samples with a 50°C pre-exposure bake corresponds to a total dose of $5000\mu \text{C}/{\mathrm{cm}}^2$ with an approximately 11-nm beam spot size and development in 25% TMAH.

Fig. 1

Scanning electron microscopy (SEM) images showing the result of developer concentration and beam writing techniques on etched Si patterns on Si substrate: (a) and (b) bus and ring waveguides written with a $3500\mu \text{C}/{\mathrm{cm}}^2$ dose and developed with low-contrast 2.38% tetramethylammonium hydroxide (TMAH) without and with 0.125 deg per-pass rotational offset, respectively; (c) and (d) bus and ring waveguides written with a $4000\mu \text{C}/{\mathrm{cm}}^2$ dose and developed with high-contrast 25% TMAH without and with 0.125 deg per-pass rotational offset in the ring section, respectively. The fabrication process for all four samples includes a 50°C pre-exposure bake and postexposure bake for 1 min at 1000°C in a rapid thermal annealing (RTA) in an ${\mathrm{O}}_2$ ambient. All images are at the same scale and the hydrogen silsesquioxane (HSQ) mask is intact.

Fig. 2

Top-down SEM images of etched Si demonstrating the impact of exposure dose on mask line edge roughness: (a) $1000\mu \text{C}/{\mathrm{cm}}^2$ exposure dose, 0.625 nA beam current, and 5.2 nm spot size; (b) $2000\mu \text{C}/{\mathrm{cm}}^2$ exposure dose, 1.25 nA beam current, and 6.7 nm spot size; (c) $3000\mu \text{C}/{\mathrm{cm}}^2$ exposure dose, 1.875 nA beam current, and 8.0 nm spot size. The beam current is changed along with the dose to maintain the beam sweep frequency near 10 MHz. The fabrication process for all three samples includes 50°C pre-exposure bake, four-pass writing with per-pass offsets, development in 2.38% TMAH, and postexposure bake for 1 min at 1000°C in an RTA in an ${\mathrm{O}}_2$ ambient. All images are at the same scale and the HSQ mask is intact.

Next, we studied the impact of per-pass pattern shifting on both straight and curved structures. For straight structures, the effect of per-pass offsets is mainly to minimize the impact of mechanical vibrational noise and stitches between writefields. Based on SEM analysis, stitch error is reduced for four-pass writing where each pass is shifted by one BSS away from the designed position in each coordinate direction ($\pm x$, $\pm y$). Top-down SEM micrographs of bus-coupled ring waveguides without, Figs. 1(a) and 1(c), and with, Figs. 1(b) and 1(d), per-pass rotational offset are given in Fig. 1. The LER visible on the left edge of the ring structure in Fig. 1(c) has been smoothed out due to per-pass rotational offsets in the corresponding region of Fig. 1(d). The LER of the straight bus waveguides in Figs. 1(c) and 1(d) is similar since rotational offsets are not used for straight structures.

3.3.

Postdevelopment Bake

We next explored the impact of hardbake conditions on mask LER. Cross-sectional SEM micrographs of etched Si samples hotplate baked at 350°C show significantly more roughness than those RTA baked at 1000°C in an ${\mathrm{O}}_2$ ambient, as presented in Fig. 3. Annealing HSQ in an ${\mathrm{O}}_2$ ambient breaks remaining Si-H bonds and densifies the film to form a low-loss optical cladding.⁵ By annealing before etching, the increased density also improves the ICP-RIE etch selectivity. For our etch chemistry, Si:mask selectivity increases from $4.7:1$ with hotplate baking to $7.6:1$ for the RTA process. The increased Si:mask selectivity for samples baked in RTA, Figs. 3(b) and 3(c), results in a thicker HSQ layer after etching compared to hotplate baked samples, Fig. 3(a). During annealing, the HSQ films shrink by approximately 21% in the vertical direction. By comparison, the widths of the HSQ films shrink by approximately 5 nm or $<2\%$ of the designed 450 nm width. Masking from ${\mathrm{SiO}}_2$ that forms during ${\mathrm{O}}_2$ RTA causes the reduced etch depth shown in Figs. 3(b) and 3(c).

Fig. 3

Angled cross-sectional SEM images of etched Si showing the result of postdevelopment baking on the SWR and etching uniformity: (a) 350°C hot plate bake for 2 min; (b) and (c) 1000°C RTA bake in ${\mathrm{O}}_2$ ambient for 1 and 2 min, respectively. The fabrication process for all three samples includes a 50°C pre-exposure bake, $2000\mu \text{C}/{\mathrm{cm}}^2$ total exposure dose with four-pass writing, and development in 2.38% TMAH. All images are at the same scale and the HSQ mask is intact.

Top-down and angled cross-sectional SEM micrographs of etched Si patterned with the optimal parameters found in this study are given in Figs. 4(a) and 4(b), respectively. LER is visible in the top-down image. The striations on the sidewall of Fig. 4(b) correspond to pattern width variations from LER in the HSQ mask. A summary of the fabrication parameters studied and the corresponding optimized recipe values is given in Table 1. Optimized process parameters are 50°C pre-exposure bake, $5000\mu \text{C}/{\mathrm{cm}}^2$ total exposure dose with a 3.125 nA beam and 11-nm beam spot size, four-pass writing with per-pass linear and rotational offsets, development in 25% TMAH, and postexposure baking at 1000°C in RTA in an ${\mathrm{O}}_2$ ambient.

Fig. 4

SEM images of etched Si produced with optimized HSQ mask: (a) top-down and (b) angled cross sectional. Both images show Si patterns on a Si substrate with the HSQ mask intact.

Table 1

Summary of the multiparameter fabrication study and the optimized values for each parameter.

3.4.

Sidewall Roughness

Figure 5(a) shows a 3-D AFM image of etched Si after the HSQ mask has been removed with hydrofluoric acid etching. The SWR in Fig. 5(a) corresponds to the striations of Fig. 4(b). Six total fine scans with a 9-μm length and a 9.7 nm step size are measured along both sidewalls of straight structures at heights of 40, 100, and 160 nm from the Si surface. A 3-D AFM profile scan taken at a height of 100 nm is shown in Fig. 5(b). Except for the abrupt transition seen at approximately 1.5 μm, the scan shows a relatively smooth sidewall variation that stays within $\pm 4.1\mathrm{nm}$ of ideal for the remaining 7.5 μm. All scans show similar roughness profiles from the top to the bottom of the waveguide sidewall, indicating that the LER of the mask is directly transferred into SWR through anisotropic etching. The measured RMS roughness, averaged from the six fine line scans, is 2.1 nm. The correlation length of the roughness is determined by fitting an exponential to the autocorrelation function of the 3-D AFM profile data, as shown in Fig. 6.³¹ The spatial autocorrelation function measures the similarity of the roughness profile with a copy shifted by the lag length. The extracted correlation length is 94 nm, averaged across all line scans. The oscillations in the autocorrelation function originate from oscillations in the measured waveguide width in the roughness scan, Fig. 5(b), with the same period. Both the RMS roughness and the roughness correlation length are required for theoretical or simulated estimation of waveguide propagation loss.³¹

Fig. 5

(a) 3-D AFM image of etched straight waveguide section with the HSQ mask removed. 3-D AFM measurements are performed on etched Si patterns on Si substrate. (b) Measurement of the sidewall position measured by 3-D AFM. The sidewall profile has been normalized to pass along $y=0$ but not smoothed.

Fig. 6

Plot of the autocorrelation function for the 3-D AFM scan of the straight waveguide sidewall given in Fig. 5(b). An exponential fit to the autocorrelation result gives a roughness correlation length of 97.7 nm. The measured RMS roughness for this scan is 1.8 nm.

3.5.

Transmission Loss

A variety of test structures are used to characterize the optical propagation losses of waveguides fabricated with optimized HSQ masks. An etched Si dual-coupled ring resonator structure similar to those used for resonant measurement of propagation losses is shown in Fig. 7. The ring resonator exhibits minimal LER and nominally equal coupling gaps to the two bus waveguides. Figure 8 presents the measured ring resonator spectra for single-bus coupled and dual-bus coupled rings. TM-mode single-bus ring resonators near their critical coupling gaps demonstrate quality factors as high as 122,000. Optical propagation losses calculated from critically coupled single-bus rings and dual-coupled through- and drop-port transmission measurements of dual-bus rings like those shown in Fig. 7 give values of 3.6 to $5.7\mathrm{dB}/\mathrm{cm}$. Extracted propagation losses based on straight waveguide measurements are given in Fig. 9(a). When averaged across the telecommunications C-band, TE-mode propagation losses are 2.4 and $2.8\mathrm{dB}/\mathrm{cm}$ for waveguides with and without the additional postetch RTA, respectively. TM-mode propagation losses are $2.5\mathrm{dB}/\mathrm{cm}$ both with and without the additional postetch RTA bake. Measured losses for ring waveguides are larger than those for straight waveguides due to both increased roughness in curved structures and larger measurement error in determining the loss in resonant structures. Error sources in resonant measurements include uncertainty of the critical coupling condition for single-bus coupled rings and differences in the coupling coefficients between the ring and two bus waveguides for dual-bus coupled rings.⁴⁷ Table 2 provides a summary of the 3-D AFM SWR and optical propagation loss measurements. Fiber-to-chip coupling losses averaged across the C-band for all samples are $<1.3\mathrm{dB}/\text{connection}$.

Fig. 7

SEM images of etched Si waveguides on Si substrate. (a) Top-down view showing dual-bus and single-bus coupled ring resonator patterns. (b) Close-up view of the dual-bus coupled ring resonator shown in the dashed box in (a).

Fig. 8

Measured ring resonator spectra for 10-μm radius TM-mode ring resonator waveguides: (a) Single-bus coupled ring with a quality factor of 113,000. (b) Dual-bus coupled ring response used for estimating curved waveguide propagation losses without the requirement for critical coupling. The extracted propagation loss in this ring is $5.0\mathrm{dB}/\mathrm{cm}$ as determined by the symmetric coupled ring method (Ref. 47).

Fig. 9

Measured optical propagation losses of SOI waveguides: (a) average TE-mode and TM-mode losses are 2.8 and $2.5\mathrm{dB}/\mathrm{cm}$ across the telecommunications C-band, respectively. (b) Comparison of TE-mode losses with the optimized HSQ mask to waveguides fabricated without the optimizations reported in this work.

Table 2

Summary of 3-D AFM SWR and optical propagation loss measurement results. Propagation losses are not given for TE-mode ring resonators due to the observation of split resonances.