Demonstration of a hitless bypass switch using nanomechanical perturbation for high-bitrate transparent networks

: We demonstrate an optical hitless bypass switch based on nanomechanical proximity perturbation for high-bitrate transparent networks. Embedded in a single-level π -imbalanced Mach-Zehnder interferometer, the two nanomechanical-based ∆β -directional couplers permit broadband signal rerouting on-chip, while the selected wavelength remains unaffected at all times for optical filter reconfiguration. The optical hitless switch is implemented in the silicon nanophotonics platform, with experimental measurements matching well with numerical and theoretical modeling.


Introduction
In current wavelength division multiplexing (WDM) systems and photonic on-chip interconnect networks [1][2][3][4][5][6], rerouting of wavelength signals is a critical element to optical information processing. This can be achieved with reconfigurable optical add/drop multiplexers (OADMs) [7][8][9]. The present challenge involving such OADMs is that their reconfiguration to reroute a different wavelength must take a finite time [10][11][12][13], which is typically in the range of microseconds to even milliseconds. During this reconfiguration time, undesired wavelengths get rerouted if they are in the same channel as the OADMs. This is further illustrated in Fig. 1. For example, for a 40 Gbit/s channel, a 1 microsecond reconfiguration time of the OADMs results in 4 × 10 4 bits lost when the targeted wavelength is tuned from one position to the next or when the filter is switched on or off in an optical network. Our objective is therefore to demonstrate a design that permits wavelength filter reconfigurability and on/off switching without any loss of information bits. Also, the entire operation needs to be performed without affecting other wavelength channels -in other words, the operation needs to be transparent or hitless [14]. Two main requirements emerge from this scenario: (1) enabling a binary on/off switching of the filters without any loss of data bits, and (2) enabling filter reconfigurability without any loss of data bits by temporary transfer of data from the signal bus to a ``bypass" bus without disturbing other wavelength channels. Secondary objectives are to achieve small form factors and low power requirements for scalability to high-density integrated optics, materials compatibility with silicon photonics for potential processing integration, and minimal design and fabrication complexity in order to for high-density scalability. Fig. 1. A challenging problem in current WDM networks. The flow of signals through a reconfigurable optical add-drop multiplexer (OADM) before, during and after reconfiguration is shown schematically. The different wavelength signals are depicted by different colored arrows. Stage 1 depicts the situation before any tuning of the OADM has taken place. The desired wavelength signal λ1 is dropped out and re-routed. Now, if it is desired to drop λ3 instead of λ1, reconfiguration of the OADM takes place. During this finite time, the OADM is tuned through all wavelength channels in between λ1 and λ3 and undesired wavelength signals get re-routed. As an example, in Stage 2 at some time step during reconfiguration, λ2 (where λ1 < λ2 < λ3) although not desired, gets re-routed to a different channel.
Recently a few solutions to this approach of hitless tuning has been suggested [14-19], along with spatial non-blocking switches [20] and earlier studies [21]. The approach towards solving this important problem described in Ref. 14 to 18 employs a "slow" hitless switching scheme for lossless reconfiguration. Here we demonstrate experimentally, for the first time, this hitless switch scheme using integrated nanomechanical proximity perturbation in a singlelevel π-imbalanced Mach-Zehnder interferometer.
The operation of the hitless bypass switch is described schematically in Fig. 2, and a practical implementation of this scheme is illustrated in Fig. 3. It consists of two inversesigned symmetrically actuated 2 × 2 switches arranged in a balanced Mach Zehnder configuration. In this implementation, the 2 × 2 switch is a ∆β switched directional coupler (where ∆β is the propagation constant mismatch between two waveguides) with nanomechanical proximity perturbation [22-25] to control the power transfer ratios [26][27][28] with tuned propagation constant mismatch ∆β. A φ-phase shift, relative to the phase response of the reconfigurable filter, is introduced to one arm of the Mach Zehnder, with a π-phase shift desired if no reconfigurable optical device is present. The switch does not need to be fast as it serves as a bypass router for the reconfigurable filter. Fig. 2. The hitless bypass switch -a possible solution to WDM reconfiguration. The hitless bypass switch schematically described here attempts to provide a possible solution to the problem described in Fig. 1. Similar to Fig. 1, the flow of signals through a reconfigurable optical add-drop multiplexer (OADM) before, during and after reconfiguration is shown schematically. The different wavelength signals are once again depicted by different colored arrows. Stage 1 depicts the situation before any tuning of the OADM has taken place. The desired wavelength signal λ1 is dropped out and re-routed. Now as described in Fig. 1, if it is desired to drop λ3 instead of λ1, reconfiguration of the OADM takes place. During this finite time, the OADM is tuned through all wavelength channels in between λ1 and λ3. To prevent undesired wavelength signal loss, during the reconfiguration of the OADM, the hitless bypass switch provides a path where all the signals are temporarily transferred, thus completely bypassing it. This is shown in Stage 2. In Stage 3, after reconfiguration of the OADM is complete, the hitless bypass switch allows all wavelength signals to remain in the original channel and the desired wavelength λ3 gets re-routed.

Hitless bypass switch operation
The operation of the hitless bypass switch is described in Ref. 14, and illustrated here in Fig. 4 for convenience. The directional couplers are designed such that in the unperturbed case (∆β = 0) complete power transfer takes place from one waveguide to the other.  (3) inclusion of a φ-phase shift in one of the interferometer arms.
Case I: (∆β ≠ 0) In the off-state, the ∆β switches are completely detuned (Fig. 4a) and all input wavelengths from the top port (a 1 in Fig. 4a) go through the top port a 3 of the directional couplers to the wavelength selective device, which in this case is a ring resonator filter. The desired wavelength then gets rerouted while the other wavelength channels remain unaffected and emerge out of a 5 .
Case II: (∆β = 0) During reconfiguration of the ring resonator filter, the switches are synchronously turned to the on-state (unperturbed case; Fig. 4b) and all input wavelengths travel from the top input port (a 1 in Fig. 4b) to port a 4 and are then transferred back to a 5 . Thus, the switch creates a bypass port for the wavelengths during the tuning of the OADM.
It is important to note that during all intermediate states between the unperturbed case and completely detuned case or vice versa, the undesired wavelength channels remain unaffected and always exits out of the port a 5 . This is achieved because of the reversal of the input of the second directional coupler with respect to the first, and also by the π-phase shifter which causes destructive interference of the signals at the bottom port port a 6 of the second directional coupler. As an example, we illustrate the 3-dB scenario in Fig. 4c. In the 3-dB coupler, the signal input a 1 power is evenly split between two arms, with a π/2 phase lag in (for example) the bottom arm. The signal in the bottom arm is then additionally phase-lagged by π. Upon exit of the second cascaded perturbed directional coupler, the wave in a 6 is destructively interfered, while the wave in a 5 maintains unchanged in this transition state. In all other transition states, we emphasize that this analysis continues to holds true, as with the 3-dB transition state. This will be further verified with coupled-mode calculations in the next section. The resultant signal intensities for various states are summarized in Table 1, with α the signal intensity in port a 3 . For any phase shifts introduced by optical devices in the bypass region, a relative phase shift of π is sufficient to ensure destructive interference at port a 6 . Fig. 4. Schematic of signal propagation through the hitless bypass switch. a, for limiting condition of completed detuned state. b, for limiting condition of completely unperturbed state. c, for the 3-dB transition state. Notice in the 3-dB transition state, the power in the top arm is π-phase shifted, leading to destructive interference at the exit of the perturbed directional coupler. This maintains power in a5 throughout the transition. As described in Ref. 14, the coupled mode theory formalism [29,30] has been employed to illustrate the operation and performance sensitivity of the hitless switch in the weakly coupled regime. The net transfer matrix for the two cascaded directional couplers and the phase shifter, phase-shifted by ± ∆θ/2, is described in Ref. 14 as: where r (r'), t (t') and φ (φ') are the reflection, transmission and phase of the first (second) directional coupler. As defined in Ref. 14, the transmitted wave amplitudes ' b is expressed by T T = − . If we assume that we have identical ∆β switches under synchronous operation, then r = r' and t = t'. To achieve complete transmission, Eqs. (1) and (2) verifies that two conditions need to be true: (1) switch 2 input has to be reversed with respect to switch 1 so that φ = -φ' and, (2) ∆θ = π or a π differential phase shift must be introduced to one of the arms of the directional coupler [14]. Figure 5a illustrates the results of coupled-mode analysis on the two cascaded perturbed directional couplers in the Mach-Zehnder configuration. The extinction ratio, defined as With zero detuning, all power is found in a 4 ; as the detuning δ goes up to 3κ , the signal in a 4 drops to zero and all the power goes to a 3 . We note that this transfer does not affect the power levels in a 5  can be observed in experiment. Alternatively, this suggests that even if there are device nanofabrication variations from the optimal design parameters, a sharp dip will still be observable in experiments, since the range of δ is sufficiently large -using ∆β perturbation -to compensate for deviations of κ from design.

Performance and sensitivity analysis of the design
Variations on the symmetry of the ∆β perturbation and coupling, and π-imbalanced Mach-Zehnder are investigated through closed-form solutions from coupled-mode theory. Figure 5b shows the power in the hitless port 2 5 a (normalized by input power 2 1 a ) as the device goes through the transition states, when the π-phase shift departs from ideal. In particular, we observe an excess loss of ~0.1 dB (0.4 dB) for a 10% (20%) variation in the desired π-phase shift. During the bypass switch transition, the maximum excess loss occurs when δ/κ = 3 /2. Furthermore, as shown in Fig. 5b, we emphasize that the switch remains hitless with zero excess loss for all phase variations when δ/κ = 3 . This is due to zero cross-over when the perturbed directional couplers are completely detuned. Figure 5c shows hitless port 2 5 a as the device goes through the transition states, when the two perturbed directional couplers departs from ideal. In particular, to keep the excess losses below 0.5 dB, variations in the perturbations of the two directional couplers (δ a /δ b ) should be kept between 0.76 and 1.31 of each other. In addition, unlike the variations on the π-phase shift analyzed in Fig. 6a, there are residual transmission losses even at δ/κ = 3 due to incomplete field interference when the perturbations are asymmetric. Figure 5d shows the sensitivity of the two-cascade perturbed directional coupler design to asymmetry in two couplers, which may arise from nanofabrication variations in the directional coupler gap separation. To keep the excess losses below 0.5 dB, variations in the coupling coefficients of the two directional couplers (κ a /κ b ) should be kept between 0.78 and 1.20 of each other. Residual transmission losses are also due to incomplete field interference when the couplers are asymmetric.

Nanomechanical proximity perturbation based ∆β directional coupler switch
A key component of the hitless bypass switch is the ∆β directional coupler, where the propagation mismatch can be controlled through thermal heating [31,32], electro-optic tuning [33,34], or suspended movable waveguides [35][36][37][38]. The first two approaches require relatively high powers, while with an electrostatic approach for the movable waveguides, significantly lower power switching is required. Here we employ electrostatic actuation of a fabricated movable dielectric nanobeam (100 nm in width and 200 nm in height, as shown in the inset of Fig. 6b) for proximity perturbation of the directional coupler. Fig. 6. Numerical calculation of detuning parameter (δ/κ) for a high index contrast system. a, 3D FDTD method was used to find the coupling length for various separations between the two arms of the directional coupler. Here coupling length refers to the length required for complete coupling of light from one port of the directional coupler to the other. For a particular separation, the length is found and then κ is calculated. The inset gives the geometry and dimensions of the directional coupler used in the simulation. Inset also gives the 3D FDTD calculation showing the transfer of power from one port to the other looking from the top of the directional coupler. b, The detuning parameter (δ/κ) is plotted against the proximity distance.
The figure shows that the required detuning parameter value (δ/κ = 3 ) can easily be achieved using the proximity perturbation method. Inset shows the geometry and dimension of the device that was studied.
As shown in Fig. 6a, our particular implementation is for a silicon-based directional coupler, with each (500 nm × 200 nm) waveguide designed to support a single transverse electric (TE) mode. 3D finite-difference time-domain simulations [39] were performed to obtain the coupling lengths versus the separation distances (for example, a κ of ~1.85 × 10 4 m −1 for a gap separation of 280 nm). We then examined numerically the perturbed waveguide propagation constant β' in the first waveguide, for varying proximity distance d (labeled in the inset of Fig. 6b) of the nanobeam from the directional coupler. δ was obtained from (β' -β 2 )/2 where β 2 is unperturbed propagation constant of the second waveguide. Figure 6b shows the resulting designed δ/κ values for varying nanobeam proximity distance d. A large set of δ/κ values can be obtained within the controllable range of d. Particularly, we note that the desired detuning δ/κ of 3 can now be achieved for complete switching, as required in the hitless switch. We further confirm the switching with finite-difference time-domain results as shown in Fig. 7, for perturbed directional coupler with a 85 um coupling length. The top panel of Fig.  7b shows the directional coupler without any perturbation [see also Fig. 7a(3)]; the bottom panel shows the completely detuned state, with power remaining in the source waveguide at the output [see also Fig. 7a(1)]. In the completely detuned state (Fig. 7b bottom panel), we also note the weak spatially-oscillating fields between the two waveguides, which correlates with earlier suggestions from coupled mode theory, with an estimated maximum power transfer of 0.25 at half the original coupling length.

Nanofabrication of the hitless bypass switch
Encouraged by the numerical and theoretical simulations, we fabricated the complete hitless bypass switch on silicon-on-insulator wafers with 200 nm silicon device thickness and 3000 nm oxide substrate. The devices are patterned using deep-UV or electron-beam lithography, with reactive ion etching as shown in Fig. 8a to 8e. The two ∆β directional couplers are cascaded in the Mach Zehnder arrangement, with the π-phase shift is implemented with extension of one of the Mach-Zehnder arms by half its guided wavelength. The directional coupler gap is 250 nm. To demonstrate the hitless switch operation, the coupling lengths are chosen to be 40 um. This ensures a complete cross state when unperturbed (with 0.4-dB excess loss) and a mixed cross / bar state (74.6% power in cross state, 16.5% in bar state, with remaining in excess loss) when perturbed with 90 nm proximity distance. d Fig. 8. Fabrication results of the hitless bypass switch. a, Optical microscope image of the hitless bypass switch with the cascaded nanomechanical proximity perturbation based directional coupler ∆β switches and a ring resonator coupled to one of the arms of the Mach Zehnder interferometer. Scale bar: 100 µm. Inset shows the cross-section profile of a waveguide (not same device) with vertical sidewalls and small sidewall roughness. b, Close-up SEM image of the perturbing dielectric and the waveguides. Perturbing distances of ~90 to 100 nm was achieved for some of the devices. c, SEM image of the comb-drive actuators. The perturbing dielectric is connected to the comb-drives through thin beams. d, Optical microscope image of the mask created from photoresist. The clear area is directly exposed to the vapor HF and the oxide underneath that gets taken away releasing the comb-drives and the directional coupler. e, SEM image of a ∆β switch after vapor HF and oxygen plasma to remove the photoresist layer.
Almost vertical sidewalls with very little sidewall roughness (less than 5 nm) were achieved. Overall, less than 6% deviation from the designed dimensions were obtained in the hitless switch nanofabrication, within the tolerance levels for less than 0.5 dB excess loss targets. Gold electrodes were next patterned onto the silicon-on-insulator samples through electron-beam lithography and thermal evaporation. A photoresist mask is then patterned and the oxide underneath the device region undercut with vapor hydrofluoric acid [40-43] to suspend the perturbed directional couplers and the comb-drive actuators for measurements.

Experimental results and discussions
The fabricated devices were tested using a tapered lensed fiber-chip-fiber transmission setup (Fig. 9), consisting of a tunable laser (1480 nm to 1580 nm) as the input source with lock-in amplification, polarization controllers and precision xyz-stages for input/output fiber positioning to the waveguides. To actuate the nanomechanical beams and the comb-drives, four external probes were connected to the electrodes (two for each ∆β directional coupler switch) with static DC differential bias applied during the optical measurements. Fig. 9. Experimental setup for testing the performance of the hitless bypass switch. a, Schematic of the experimental setup. Inset shows a close up image of the chip under test together with the input and output tapered lensed fibers and two electrical probes. To measure the hitless bypass switch four probes was used. b, Zoomed in top view photograph of the chip without the probes. c, Modified setup to collect the output light using an objective lens instead of the tapered lensed fiber. Inset shows the output IR image seen from the cross-section of the waveguide ends. It shows that the output light from one of the arms of the bypass switch is much stronger than the output from the other arm.
We first measure directly the switching contrast of the nanomechanical proximity perturbation based ∆β directional coupler switches. The drop port transmission of the ring resonator (coupled to one of the arms of the Mach Zehnder interferometer) was measured, in both the perturbed (0V DC bias) and unperturbed states. To achieve the unperturbed state, a 40V DC bias was applied to the electrodes of the comb-drive actuators to pull the nanomechanical dielectric beam away from the directional couplers. The result of the experiment is presented in Fig. 10c. Almost 31 dB of switching contrast was achieved at the resonance peak wavelength of 1540 nm. The variation in the peak transmission is suspected to be due to wavelength-dependences of the coupler. The high extinction ratio is achieved because of the small proximity perturbation distance (~90 nm) obtained for the nanofabricated devices. Figure 10b shows the topview infrared images of the chip collected in the far-field, corresponding directly with the input and output port positions of Fig. 10a. To verify the hitless performance of the bypass switch, the transmission through port B and port C (illustrate in Fig. 10a) was measured for the two example unperturbed (40V) and perturbed (0V) cases. The result is shown in Fig. 10d. From the results, it is clear that the measured transmission through port C always has suppressed transmission for our perturbed and unperturbed states, for the full range of wavelengths examined with our tunable laser source. Note that the perturbed case has light intensity in both arms of the π-imbalanced Mach-Zehnder (74.6% power in cross state and 16.5% in bar state). With the π-phase shift, the fields destructively interference in the second directional coupler, with output transmission in port C suppressed. Moreover, the transmission through port B is always high for the non-resonant wavelengths irrespective of the state of the individual ∆β switches, demonstrating the functional performance of the bypass switch. We also note that in port B there are local resonances (dips in the dash-dot green plot) during the perturbed (0V) state, which arises from the finite reflections at the interfaces of the ∆β switches and can be reduced through adiabatic tapering at the interfaces. We emphasize that the average extinction ratio (between port B and C) preserves an approximately 17.3-dB and 11.8-dB intensity contrast for the unperturbed and perturbed states, across the wavelength range demonstrated for the hitless switch.

Conclusion
We present the first implementation of a nanoelectromechanical single-level hitless bypass switch, in a silicon photonics platform. The proximity perturbation method was introduced to demonstrate the cross-to-bar state switching in each of the two cascaded ∆β switches in the Mach-Zehnder interferometer. With an included microring filter, an almost 31 dB of extinction ratio with negligible undesired wavelength rerouting was demonstrated. This allows an optical hitless switch for future high density integrated circuits. Future studies include high bitrate measurements to demonstrate the implementation of this hitless switch in on-chip nanophotonic networks for optical signal processing.