1. Introduction

Simulation results show that high efficiency compact bends can be implemented in low refractive index waveguide material systems with low refractive index contrast waveguides through the use of air trenches for bends that have bend angles less than 90° [1]. However, the fabrication of such structures is challenging because of the tight tolerance on the sidewall smoothness and verticality, and on the placement accuracy of the trench edge with respect to the waveguide bend. Nonetheless, as reported in Reference 2, we used a conventional contact UV mask aligner in conjunction with a highly anisotropic polymer etch to demonstrate measured bend losses of 0.33 and 0.30 dB/bend (92.7% and 93.2% bend efficiency) for TE (electric field in the plane of the substrate) and TM (electric field out of the plane of the substrate) polarization, respectively, for 45° waveguide bends in a partially fluorinated polymer waveguide material system. This was achieved by fabricating a large set of devices with a range of air trench misalignment and etch undercut compensations such that at least one final device had reasonable fabrication accuracy.

In this paper we report the fabrication and measurement of air trench waveguide bends in which the air trenches are patterned with electron beam lithography (EBL) in a scanning electron microscope. A key advantage of this approach is that an air trench can be placed very accurately with respect to the waveguide bend if the proper design and fabrication strategy is used in conjunction with alignment marks placed in each individual EBL exposure field. The resultant alignment accuracy is consistent with what can be achieved in current steppers, and therefore our measured performance results give some indication of what can be realized if devices that make use of air trench bends are fabricated on a modern microfabrication line. Moreover, the EBL-patterned air trench features produce smoother sidewalls than contact UV lithography, which reduces scattering loss. The net result is that the measured loss is reduced to 0.124 (97.2% bend efficiency) and 0.166 dB/bend (96.2% bend efficiency) for TE and TM polarization, respectively. Note that these values are within a factor of two of three dimensional (3D) finite difference time domain (FDTD) simulation results, which predict 0.066 (98.5% bend efficiency) and 0.088 dB/bend (98.0% bend efficiency) losses for TE and TM polarization [2].

The main motivation to develop air trench bends is to enable the realization of waveguide devices in which the device size is independent of the refractive index contrast between the core and clad materials. For low refractive index contrast materials (i.e. refractive index contrast in the range of ~0.3% to ~1.5%) the minimum bend radius is typically on the order of several millimeters or larger for conventional curved waveguide bends. This fundamentally limits the level of device integration that can be achieved for a given die size compared to the use of air trench waveguide bends. On the other hand, the amount of loss in air trench bends determines the ultimate level of device integration that can be achieved for a desired device insertion loss. Thus, realizing the lowest possible bend loss in air trench bends dramatically affects their practical potential to shrink the size of planar lightwave circuits (PLCs).

While the results reported in this paper are demonstrated in perfluorocyclobutyl (PFCB) [3–5] (Tetramer Technologies, Inc.) waveguides, similar results should be obtainable in other low index, low index contrast (LILIC) material systems. These include silica and other polymers. As discussed in References 1 and 2, the important criteria in air trench bend optical performance are air trench edge placement accuracy and sidewall verticality and smoothness, rather than the particular material system that is used. The placement accuracy is a function of the lithography system and alignment strategy, while the sidewall verticality is determined primarily by the etch tool and the etch parameters and chemistry that are used. The sidewall smoothness depends on the lithography system, etch tool and etch chemistry, and, in many cases, the etch mask material. As long as appropriate choices are made for a given LILIC waveguide material system, we expect that our results can be duplicated for waveguides in other LILIC materials.

In the following sections, we first review the design of air trench bends used in this paper (Section 2). This is followed in Section 3 by a detailed discussion of the fabrication process for 45° air trench bends in PFCB waveguides. In Section 4 we present measurement results.

2. PFCB waveguide single air interface 45± bends

As shown in Fig. 1(a), our PFCB waveguides are designed to be single mode (at λ=1550 nm) channel waveguides with a 3.6 µm square cross section [2]. The refractive index of the PFCB core material is 1.4836 and 1.4816 for TE and TM polarized light, respectively. For the cladding, the TE and TM refractive indices are 1.4644 and 1.4625. The refractive index contrast for both the TE and TM polarization waveguide modes is 1.3%.

 

Fig. 1. PFCB waveguide bend geometry. (a) Cross section of PFCB channel waveguide structure on top of silicon substrate. (b) Top view of air trench and waveguide structure; Z defines the distance from the intersection of the center lines of the input and output waveguides to the air trench interface in a direction normal to the interface. (c) Schematic diagram of bend angle definition.

Download Full Size | PDF

Figure 1(b) shows the waveguide bend geometry with the waveguide bend angle definition given in Fig. 1(c). As discussed in Refs. [1] and [2], the bend angle is selected to insure that all of the angular spectrum components of the waveguide mode undergo total internal reflection (TIR) in order to achieve high optical efficiency for the bend.

Numerical simulation of PFCB single air interface 45 bends is done with the two dimensional finite difference time domain (2D-FDTD) method using a mode overlap integral (MOI) to calculate how much power is in an actual waveguide mode after reflection from an air trench bend. The calculated bend loss as a function of Z [defined in Fig. 1(b)] is shown in Fig. 2. Due to the Goos-Hanchen shift, the maximum bend efficiency occurs at Z=-0.2 µm, where the bend efficiencies for TE and TM polarization are 99.0% and 98.7% respectively. The 2D-FDTD simulation result at Z=-0.2 µm is verified with three dimensional (3D) FDTD, which gives bend efficiencies of 98.5% for TE polarization and 98.0% for TM polarization. Note that the 2D FDTD simulation results are within 1% of these values.

3. Fabrication

To improve the placement accuracy of an air trench interface with respect to a waveguide bend compared to our previous work [2], we chose to use the autoalignment feature of our in-house NPGS version 9 electron beam lithography system (JC Nabity Lithography Systems), which is used with a LEO 1550 field emission scanning electron microscope (SEM). Placement of appropriate alignment marks in each of the four corners of an exposure field enables alignment accuracies of less than 40 nm to be achieved, as illustrated in Fig. 3. The optically patterned vernier scales have a period of 5.00 µm while the EBL patterned vernier scales have a period of 5.04 µm [6]. Figure 3(a) shows a set of verniers in a 120 µm×120 µm exposure field, while Fig. 3(b) shows the verniers from one corner of a 1,000 µm×1,000 µm field. Note that in both cases the EBL patterned scales are positioned nearly perfectly with respect to the optically patterned scales, indicating that the alignment accuracy is significantly better than the difference in periods between the two scales.

 

Fig. 2. The bend efficiency as a function of the parameter Z defined in Fig. 1 (after Ref. 2).

Download Full Size | PDF

To take advantage of the alignment accuracy available in our low-end (i.e., in comparison with a multimillion dollar commercial e-beam lithography tool) EBL system, EBL alignment marks must be patterned in all four corners of each exposure field during the same patterning step that defines the waveguides. This adds significant complexity to our fabrication process, as is detailed below. Nonetheless, the result is fabricated air trench bend structures that have an alignment accuracy comparable to what can be achieved in optical steppers such as would be used in a conventional high volume microfabrication line. Hence the results reported in this paper are intended to be indicative of what should be achievable in volume manufacturing. Note that in such a manufacturing process, many of the additional complications in the fabrication process that are specific to the use of our EBL tool become unnecessary.

Our fabrication process involves three optical mask steps and one EBL step. Figure 4 shows the layout for the second and third mask patterns and the EBL pattern for one set of equal-length waveguides that have different numbers of air trench bends. As discussed in Section 4, sets of such waveguides are used to measure the optical efficiency of the bends. The second mask defines the blue colored features, which consist of the waveguides, Au EBL alignment marks, and Au identification numbers. The third mask (red) defines regions that are etched down through the thickness of the PFCB uppercladding to reveal the Au EBL alignment marks which are at the same level as the top of the waveguide core layer. Exposure fields are indicated by dashed black lines, which are 342 µm×342 µm. Note that there is an EBL alignment mark in each corner of each exposure field. These are used to align the NPGS system to define the air trench patterns (shown in green) using EBL. Also note that the presence of these alignment marks is specific to our NPGS-based fabrication process. In a high volume fabrication line with typical optical steppers, these would not be necessary and the process flow would be dramatically simplified.

 

Fig. 3. Microscope images of verniers in (a) 120 µm×120 µm and (b) 1,000 µm×1,000 µm exposure fields. The light colored vernier scales in each pair are 57.5 µm long optically patterned Cr/Au (5 nm/50 nm thick) on a silicon substrate. EBL alignment marks (not shown) are patterned in each corner of each exposure field in the same lithography step. The blue colored vernier scales are exposed and developed regions of 400 nm thick e-beam resist, ZEP 520. They are patterned with e-beam lithography using the NPGS two-level autoalignment feature.

Download Full Size | PDF

 

Fig. 4. Mask layout for a set of equal length waveguides, each with a different number of air trench bends. See text for details.

Download Full Size | PDF

The main steps in the fabrication process flow are shown in Fig. 5. First, an 11 µm thick PFCB underclad layer is spin coated on a 75 mm Si wafer and cured at 190C for 22 hours, followed by a 3.6 µm core layer which undergoes an identical cure process as the underclad [Fig. 5(a)]. Next, a lift-off process with AZ701 photoresist is used to define patches of 50 nm thick Au on top of a 5 nm Cr adhesion layer [Fig. 5(b)]. These are etched in a later step to form the EBL alignment marks as well as identification markings. An optical microscope image of the lifted-off Au patches is shown in Fig. 6(a) for the same region as illustrated in Fig. 4.

The next step is sputter deposition of a 150 nm thick SiO₂ layer for the waveguide etch mask. Figure 5(c) shows photoresist on top of the SiO₂ layer that has been patterned with the second optical mask, which defines the waveguides and Au alignment marks. Figure 5(d) shows a schematic illustration of the cross section after the SiO₂ etch in an inductively coupled plasma reactive ion etcher (ICP RIE) (Surface Technology Systems, PLC), while Fig. 5(e) shows the result after ICP RIE etching of the core layer. Note that during the PFCB etch the Au that is not under patterned photoresist/SiO₂ is sputtered away, thereby forming the Au EBL alignment marks. Also note that the core layer underneath the Au alignment marks remains after the PFCB etch although these regions have nothing to do with the actual waveguides.

The SiO₂ etch is based on an O₂ and C₄F₈ etch chemistry, while the PFCB etch is the same as we reported in Reference 2 and is based on a He and O₂ chemistry. We overetch the PFCB core layer to ensure that all of the core material is removed except what is protected by the SiO₂ etch mask. Following the PFCB etch the SiO2 is stripped in buffered hydrofluoric acid. Figure 6(b) shows an optical microscope image of a sample at this stage in the fabrication process.

An 11 µm PFCB upper cladding layer is then spin coated and cured at 190C for 44 hours [Fig. 5(f)]. The additional curing time ensures that its refractive index stabilizes to a similar value as the underclad, which sees multiple curing steps since it is the bottom layer. Prior experimental work in our laboratory has shown that the refractive index of the PFCB formulations we use stabilize somewhere between 22 and 44 hours when cured at 190C.

Before e-beam lithography can be performed to define the air trenches, the Au alignment marks buried under the overclad must be revealed so that they are visible to the electron beam. As shown in Fig. 5(g), a 100 nm thick Al layer is sputtered on top of the upperclad, followed by spin-coated photoresist which is patterned with the third optical mask. The pattern in the photoresist is transferred to the Al layer by wet chemical etching. The PFCB above the Au alignment marks is then ICP RIE etched with He/O₂ to reveal the Au alignment marks [Fig. 5(h)]. The photoresist is also removed during this process. An optical microscope image after the PFCB etch is shown in Fig. 6(c). The 75 mm wafer is then diced into die, and further processing occurs on individual die.

At this point the Au alignment marks can be seen in the SEM since they are no longer covered by the PFCB uppercladding layer. However, since they are surrounded by nonconductive PFCB [Fig. 7(a)], they experience charging which spoils the EBL alignment process. As illustrated in Fig. 7(b), to avoid this problem we conformally sputter deposit 30 nm Al to electrically connect the Au alignment marks to the Al etch mask, which in turn is grounded during EBL. The thin Al layer over the Au alignment marks does not impede imaging of the alignment marks during the autoalignment process. In fact, we find that the Al can be up to ~120 nm thick before imaging of the Au alignment marks is impaired.

After Al deposition, a 400 nm thick positive electron beam resist, ZEP520A (ZEON Chemicals), is spin coated and baked at 90C for 1 minute on a hotplate and then at 170C for one hour in an oven. As illustrated in Fig. 5(i), the air trenches patterns are then written using EBL and the electron beam resist is developed.

 

Fig. 5. Schematic illustration of fabrication process. See text for details.

Download Full Size | PDF

Once the air trench patterns are defined in the e-beam resist, they are transferred to the Al layer by ICP RIE etching with an Argon plasma. The deep PFCB air trench etch is then performed by ICP RIE etching with the same He and O₂ etch chemistry as used to define the waveguide cores [Fig. 5(j)]. While the dimensions of the air trenches can be substantially reduced from what is shown in Fig. 1(b) to decrease the area required for a bend and thereby facilitate even higher levels of integration, we chose to keep them the same as in our previous work [2] for purposes of this study. Figure 6(d) shows a microscope image in which the deeply etched air trenches are clearly visible (as well as the waveguides since the microscope is operated in differential interference contrast mode). Finally, the remaining Al etch mask is removed by wet chemical etching and the two die endfaces containing waveguides are rediced to provide fresh interfaces for butt-coupling to input and output fibers for optical measurement.

 

Fig. 6. Optical microscope image of the sample at different steps in the fabrication process: (a) after Cr/Au lift off, (b) after ICP-RIE waveguide etch, (c) after the removal of upper cladding material on top of the Au alignment mark regions, (d) after the deep air trench ICP-RIE etch and before strip of the Al etch mask.

Download Full Size | PDF

 

Fig. 7. (a). Au alignment marks after the opening etch is not electrically connected to the Al etch mask which results in charging during alignment. (b) 30 nm sputtered Al eliminates this problem and still permits imaging of the Au alignment mark.

Download Full Size | PDF

Figure 8 shows an SEM image of an etched air trench. Note that the waveguide structure is buried under the PFCB overclad and is therefore not visible. Similar to our previous results [2], the verticality of the etched sidewall is excellent (90°+/-0.5°). The etch depth is 18.5 µm. Figure 9 shows a close-up of the sidewall in which it is clear that the smoothness of the sidewall is improved compared to what we reported in Ref. [2]. We attribute this primarily to the EBL process resulting in smoother resist edges for the air trenches compared to UV contact lithography, as well as slightly different PFCB etch characteristics for the Al etch mask compared to the SiO₂ etch mask used in our previous work.

 

Fig. 8. SEM image of single air trench bend.

Download Full Size | PDF

 

Fig. 9. SEM image of a typical sidewall after deep anisotropic ICP-RIE air trench etch.

Download Full Size | PDF

 

Fig. 10. Measured loss of the PFCB waveguide air trench bend as a function of number of bends at λ=1.55 µm for TE and TM polarization. Error bars are not included since they are smaller than the markers used to indicate the experimental measurement results.

Download Full Size | PDF

4. Experimental measurement

Waveguide measurements are made in a Newport PM500 Autoalign System with an Agilent 8164A tunable laser (1480nm–1580nm tuning range) as the optical source. Air trench bend loss measurements are made at a wavelength of 1550 nm. Light from the laser is coupled into a polarization maintaining (PM) fiber that is butt coupled to individual input waveguides. The end of the fiber nearest to the sample is rotated to control the polarization state of the light coupled into a waveguide. Light from the end of a waveguide is butt coupled to a single mode fiber which is connected to an optical power meter. Two computer-controlled precision 3-axis motion stage stacks with 50 nm accuracy are used to optimize the input and output fiber positions to maximize the power through each waveguide.

The measured optical power referenced to the optical power coming out of the PM fiber is shown in dB in Fig. 10 as a function of the number of air trench bends in a waveguide. Note that in this sample the 16 bend waveguide has a fabrication defect and is therefore not measurable. The measured bend loss is 0.124dB/bend (97.2% bend efficiency) for TE polarization, and 0.166dB/bend (96.2% bend efficiency) for TM polarization. We repeated the measurement many times and found that the measurement variation in this system is ±0.02dB. The fabricated bends shows low polarization dependent loss, which is consistent with 2D and 3D FDTD simulation results.

Figure 11 shows the measured optical power through the waveguide with 12 bends divided by the laser output power as a function of wavelength. Note that the wavelength variation is small since the waveguide bend mechanism is based on total internal reflection.

 

Fig. 11. Normalized transmission spectrum for the waveguide with 12 bends.

Download Full Size | PDF

Summary

High efficiency and small area PFCB channel waveguide air trench 45° bends have been designed and demonstrated. A fabrication process with an SEM-based EBL system has been successfully developed to realize high alignment accuracy and smooth etched air trench sidewalls. The measured bend loss at λ=1550 nm is improved to 0.124dB/bend (97.2% bend efficiency) for TE polarization, and 0.166dB/bend (96.2% bend efficiency) for TM polarization.