An on-chip tunable photonic delay line is a key building block for applications including sensing, imaging, and optical communication. However, achieving long and tunable delay lines within a small footprint remains challenging. Here, we demonstrate an on-chip tunable photonic delay line using ultralow loss high confinement Si3N4 waveguides with integrated microheaters. As an example of potential application, we embed a 0.4 m delay line within an optical coherence tomography (OCT) system. We show that the delay line can extend the OCT imaging range by 0.6 mm while maintaining a high signal to noise ratio. Our tunable photonic delay line is achieved without any moving parts which could provide high stability, critical for interference based applications.
On-chip tunable photonic delay lines are one of the key building blocks for realizing optical systems-on-chip.1–3 However, achieving tunable long delay lines that are broad bandwidth within a small footprint remains challenging. Photonic delay lines can be used for many applications such as optical communication,4 microwave signal processing,5,6 optical gyroscopes,7 and optical coherence tomography (OCT).8 Different devices such as photonic crystals,9,10 Bragg gratings,11 stimulated Brillouin scattering photonic filters,12,13 and microresonators14–17 have been used to provide optical delay on a chip. However, these devices are based on resonances with limited bandwidth.1,18 On the other hand, devices incorporating physical delay lines have been used to provide longer delays with broader bandwidth.2,19 These devices are not tunable and therefore can only provide a fixed optical delay. Realizing long delay lines while maintaining a low loss and small footprint is extremely challenging since it requires one to overcome not only losses due to fabrication misalignments between different lithography fields but also radiation and scattering losses due to tight bends which cause the optical mode to strongly overlap with sidewalls.
Here, we show 0.4 m of length in an 8 mm2 area using 720 bends, each with a bending radius of only 80 µm using a high confinement silicon nitride (Si3N4) waveguide (WG) (mode simulation shown in Fig. 1). We tailor the waveguide to be single mode and close to zero dispersion around 1.3 µm, which is critical for applications such as optical coherence tomography. The dispersion simulation is shown in Fig. 5 of the Appendix. In order to achieve such a tight bending radius, we choose the width to be 780 nm (in contrast to the 2500 nm used in ring resonators20) and show that despite the much stronger overlap with the waveguide surfaces than the ones typically experienced in individual micrometer-size devices,20 we achieve tens of millimeters long delay lines with low propagation losses (0.17 dB/cm ± 0.01 dB/cm). In order to enable tunability, we use the thermo-optic effect of Si3N4 and design the waveguides to ensure high thermal overlap between the optical mode and microheaters. At the same time, the design minimizes the loss from the metallic heaters. As shown in Fig. 1(a), the optical mode is not affected by the heater, which ensures minimum losses. The waveguide has a height of 730 nm and a width of 780 nm. In Fig. 1(b), we show a numerical simulation of the heat distribution profile for the integrated microheaters with a width of 5 µm and a height of 100 nm. The heat dissipation is calculated using the finite element method (COMSOL Multiphysics). The heater temperature used in the simulation is 400 K. This temperature corresponds to waveguide temperature of 368 K, which is equivalent to the waveguide temperature that we have achieved in our experiment. The waveguide temperature variation vs heater power is shown in Fig. 6 of the Appendix. One can see that since the microheater is designed in close proximity to the waveguide, it can efficiently change the temperature of the waveguide. The relative change in the optical path length of a Si3N4 waveguide, ΔlOPL, after a temperature tuning of ∆T, can be calculated as
We measure the propagation loss to be 0.17 ± 0.01 dB/cm using a design that ensures low loss despite fabrication imperfections. In order to achieve low optical losses for delay lines of tens of centimeters over several millimeters square area, we design the waveguides to be fundamentally robust to misalignments between different lithography fields. This robustness to misalignment is crucial since misalignments at the field boundary would dramatically increase propagation loss in the waveguide due to field shifts and stage instability. These misalignments are typically on the order of tens of nanometers (depending on the field size and field numbers22,23). With large and multiple fields, these misalignments can be even larger. Note that the errors due to misalignment are not only related to maskfree lithography, such as electron beam (e-beam) lithography, but also relevant to lithography which requires masks such as deep ultraviolet lithography since most of high-end masks are written using e-beam lithography. In Fig. 1(c), we show the adiabatic design of the waveguide taper that has a wider width at the lithographic field boundaries and, therefore, increases robustness to misalignment. We also show, in Fig. 1(c), simulations of the light transmission efficiency with different taper widths for a misalignment of 100 nm. We fabricated such a taper with 5 µm width. In order to confirm that the adiabatic taper design helps reduce the stitching loss, we measured the propagation loss of waveguides across different numbers of fields [see Fig. 1(d), the inset showing the schematic of the on-chip photonic delay line]. We fabricated devices that cross different numbers of fields on the same wafer, and we used an etched facet to ensure constant coupling losses for different devices.24 Each field is 5 cm long. Therefore, for a total distance of 0.4 m, we cross 8 field boundaries. We show that the propagation loss has a linear dependence with the waveguide length, which indicates that the additional loss due to the misalignment is negligible. The waveguide loss vs wavelength is shown in Fig. 7(b) of the Appendix.
As an example of an application of a tunable long delay line, we show that the photonic delay line can enhance the capabilities of an optical coherence tomography (OCT) system. OCT is an interferometric imaging technique capable of providing high-resolution, cross-sectional, and three-dimensional images with micrometer-scale axial resolution at high speed.25 Recently, great efforts have been made toward integrating and miniaturizing OCT components, including beam splitters,26 reference arm,8 and sample arm27 for miniature and inexpensive OCT systems printed on a wafer scale. However, the demonstrated components to date have been passive. It is necessary to tune the path length difference to achieve high contrast images. The signal-to-noise ratio (SNR) of a spectral domain (SD) OCT decreases very fast along the depth due to the finite sampling area of the spectrometer pixel.28 OCT is an example of a technique that can significantly benefit from such an on-chip tunable delay line. This is due to the fact that these high contrast OCT signals rely on zero-order interference, where the interferogram features lowest possible spatial frequencies and its visibility can be maximized under quantized detection.29 To demonstrate the capability of the on-chip tunable delay line, we couple it into a commercial SD OCT system (Thorlabs Telesto I) around 1.3 µm to compensate the path length difference with a small footprint by replacing the reference arm in the system. The on-chip tunable delay line is shown in Fig. 2(a). It has a total length of 0.4 m with platinum microheaters integrated on the top. The total chip size is only 8 mm2. The comparison images taken with and without our tunable photonic delay line using the commercial SD OCT system are shown in Fig. 8 of the Appendix. The images show that our tunable delay line chip does not distort the image. Figure 2(b) shows the schematic of the experimental setup to test the tunable photonic delay line. A broadband fiber coupler with a 75:25 splitting ratio is used to divide the input power into the sample and reference arms. We tune and monitor the temperature of the platinum heater by applying a current to the heater and measuring the resistance. In the sample arm, a low-NA objective (Thorlabs LSM05) is used to ensure a long depth of focus (0.63 mm) that is necessary for high-topology imaging. In addition, a fixed optical delay line (ODL) is inserted into the sample arm to match the fiber length differences in the two arms. Polarization controllers (PCs) are used to maintain TE modes. We acquire the OCT raw data with the commercial software, and the dispersion mismatch between the two arms is compensated numerically during post processing. The OCT images are reconstructed using MATLAB, following standard OCT data processing steps, including background subtraction, linear-k interpolation, dispersion compensation, and apodization.
We show that the delay line can extend the imaging range of OCT by 0.6 mm while maintaining a high SNR. The measured axial resolution of the OCT system is 6.5 µm in air. In Fig. 3, we show OCT B-scans taken from the endocardium side of the tissue before and after delay line tuning. The surface at the lower part of the OCT B-scan suffers from a reduced SNR [shown in Fig. 3(a) with zoomed-in views shown in the red box] due to the falloff of the SD-OCT system despite the use of a low-NA objective that ensures the surface remaining within the depth of focus. In Fig. 3(b), one can see that after delay line tuning, the SNR of the surface area with the lower part of the OCT scan is increased.
Using the on-chip tunable delay, we demonstrate 3D high-SNR OCT imaging (see Fig. 4) of different samples with high-topology for applications ranging from imaging to blade detection. In Fig. 4(a), we show images of a metal razor blade covered by a piece of lens tissue. The experimental arrangement is illustrated by the camera image shown in the inset of Fig. 4, where the blade is placed on the stage and covered by the lens tissue. One can see that only the lens tissue is resolved by the SD-OCT system before tuning the delay line. By tuning the delay line, the razor blade underneath the lens tissue is clearly resolved after tuning the delay line. In Fig. 4(b), we show images of a human skin tissue sample covered by a piece of gauze. The gauze and top parts of the tissue are resolved by the SD-OCT system before tuning the delay line, while more parts of the tissue that have a lower surface profile are resolved only after tuning the delay line. Similarly, in Fig. 4(c), we show images of a human aorta tissue sample with high surface elevation. One can see that only part of the aorta can be captured with a high SNR before tuning the delay line. By tuning the delay line, the two volumes of the aorta sample taking at different depths can be stitched together to reconstruct the full surface topology with a high SNR. All the volumetric images in Fig. 4 were processed using standard techniques such as multiband blending and gain compensation.28,30 The imaging depth of the OCT system is increased from 2.52 mm to 3.12 mm by adding the delay line.
In this work, we demonstrated a tunable ultralow loss Si3N4 on-chip tunable photonic delay line. With further optimization of the heater design, we could potentially provide a delay of up to 2 mm in the current waveguide. The upper limit bandwidth of our 0.4 m long tunable photonic delay line is 229 GHz at 1.3 µm. OCT is an example of a technique that can significantly benefit from such an on-chip tunable delay line. Our tunable photonic delay line is achieved without any moving parts which could provide high stability. This high stability is crucial for any interferometric measurements and, therefore, could benefit applications such as light detection and ranging (LIDAR), sensing, and coherent communication systems.
The authors would like to thank Dr. Steven Miller and Dr. Samantha P. Roberts for helpful discussions. This work was performed in part at the Cornell NanoScale Science and Technology Facility (CNF), a member of the National Nanotechnology Coordinated Infrastructure (NNCI), which was supported by the National Science Foundation (Grant No. NNCI-1542081). The devices were partially fabricated at the City University of New York Advanced Science Research Center and Cornell Nanoscale Facility. X.J. is also grateful to the China Scholarship Council for financial support.
This work was supported by the Defense Advanced Research Projects Agency (DARPA) (Grant No. N66001-16-1-4052), the Air Force Office of Scientific Research (AFOSR) (Grant No. FA9550-15-1-0303), and the National Institutes of Health (NIH) (Grant No. 1DP2HL127776-01).
APPENDIX: DEVICE DESIGN AND EXPERIMENTAL METHODS
1. Device fabrication
Starting from a silicon wafer, a 4 µm thick oxide layer is grown for the bottom cladding. Silicon nitride (Si3N4) is deposited using low-pressure chemical vapor deposition (LPCVD) in steps. After Si3N4 deposition, we deposit a silicon dioxide (SiO2) hard mask using plasma enhanced chemical vapor deposition (PECVD). We pattern our devices with JEOL 9500 electron beam lithography. Ma-N 2403 electron-beam resist is used to write the pattern, and the nitride film is etched in an inductively coupled plasma reactive ion etcher (ICP RIE) using a combination of CHF3, N2, and O2 gases. After stripping the resist and oxide mask, we anneal the devices at 1200 °C in an argon atmosphere for 3 h to remove residual N–H bonds in the Si3N4 film. We clad the devices with 500 nm of high temperature silicon dioxide (HTO), deposited at 800 °C, and followed by 2.5 µm of SiO2 using PECVD. CMP and multipass lithography technique can be applied to further reduce sidewall scattering losses. Above the waveguide cladding, we fabricate integrated microheaters by sputtering platinum and using a lift-off approach. We integrated microheaters on our device to control the core index change by temperature tuning.
2. Dispersion simulation
We design the waveguide to be single mode and close to zero dispersion at 1300 nm which is critical for applications such as optical coherence tomography (OCT). The OCT signal is generated based on broadband interference. Therefore, to allow all wavelengths to have the same time of flight, dispersion introduced by the delay line in the reference arm needs to be minimized. The simulated dispersion for the 730 nm height waveguide is shown in Fig. 5. We choose the waveguide width to be 780 nm, and the waveguide dispersion is −8.4 ps/(nm km).
3. Waveguide tuning
The waveguide temperature variation vs dissipated power is shown in Fig. 6. Since the waveguide is cladded with a layer of SiO2, the temperature of the waveguide cannot be directly measured. Instead, we have measured the optical path change in the waveguide. The optical path length change is measured using the OCT system. In the OCT B-scan, the signal appears at z = zs − zr, where zs is the optical path length of the samples in the sample arm and zr is the optical path length of the reference arm. In our experiment, the sample is placed at a fixed position in the sample arm. While tuning the delay line, the signal shifts along the axial (z) direction due to a change in both the refractive index and effective optical path length of the photonic delay line in the reference arm. Therefore, by measuring how much the signal shifts in the axial direction, we can extract the optical path length changes. Using the equation shown in the main text, we can calculate the waveguide temperature variation with respect to the input voltage. Note that the waveguide temperature variation has a linear dependence with the electrical power applied to the heaters which is also matched with theory.
In our study, the OCT line rate is set at 28 kHz. The delay line is tuned on the same time scale, and the speed of the heaters can be further improved above 100 kHz by optimizing the design.31 The maximum tuning range is measured to be 0.6 mm, and it is currently limited by the power supply. The tuning resolution is determined by the minimal incremental voltage applied to the heater. With an output voltage resolution of 1 mV, the tuning resolution of the delay line is 3 nm.
The measured waveguide loss vs tuning range is shown in Fig. 7(a). One can see that the waveguide tuning does not affect the loss. The measured waveguide loss vs wavelength is shown in Fig. 7(b). One can see that the waveguide loss does not change with wavelengths, which indicates that our delay line is broadband. With our 0.4 m long delay line chip, we have measured that the power conversion from TE to TM is only around 1%. So, the polarization state is well preserved in these long delay lines.
4. OCT performance
The commercial OCT system has an axial resolution of 6.5 µm (in air), a lateral resolution of 15 µm (in air), a sensitivity of 104 dB (at an A-line rate of 28 kHz), and an imaging range of 2.52 mm. The 3 dB bandwidth of the light source is 108 nm. The speed of the camera is set at 28 kHz for all the imaging experiments. In our study, we have modified the commercial system by adding a fiber splitter and a circulator [shown in Fig. 2(b)] which allows us to utilize parts of the commercial system in our experiment. For the modified system, the axial and lateral resolutions are the same. The sensitivity drops to 101 dB as the power transmitted to the sample arm is reduced after inserting the fiber splitter and circulator. The modified system still operates under the shot noise limit, since the delay line does not introduce additional noise to the system.
The on-chip delay line extends the high-SNR imaging range by 0.6 mm; therefore, the entire imaging range is extended from 2.52 mm to 3.12 mm. All SD-OCT systems suffer from the sensitivity roll-off effect due to the finite sampling of the spectrometer. The sensitivity drops when the signal appears further away over the axial direction from the DC term. In an OCT B-scan, the region starting from the DC term to the 6-dB roll-off range is defined as the “high-SNR region” and the remainder of the imaging range (Nyquist limit) is the “low-SNR region.” For tissues with a large surface curvature, the image quality is degraded when approaching the low-SNR region. The main advantage of the delay line is that it brings the signal up to the high-SNR region by changing the optical path length of the reference arm.
Figure 8 shows the comparison onion images taken with and without our tunable photonic delay line using the commercial SD OCT system. Figure 8(a) shows the image taken with our chip tunable photonic delay line, and Fig. 8(b) shows an image taken with the commercial system. The images show that our tunable delay line chip does not distort the image.