Recently, reconfigurable silicon photonics have attracted significant research interest to address the challenges in artificial intelligence (AI) [1], the Internet of Things (IoT) [2], and telecommunications [3]. Among the reconfigurable devices, microring resonators (MRRs) are becoming crucial components in various systems due to their flexible spectral processing capabilities. For instance, reconfigurable MRRs can function as wavelength-selective switches (WSSs) in reconfigurable optical add-drop multiplexers (ROADMs) for dense wavelength division multiplexing (DWDM) [4], external cavities in tunable lasers [5], weight banks in photonic neural networks [6], and tunable delay lines and infinite impulse response filters in microwave photonic signal processing systems [7,8]. Specifically, DWDM technology is ubiquitous for optical communication and computation; hence, it is one of the most important applications for reconfigurable MRRs.

Phase shifters and tunable couplers with low power consumption and high tuning efficiency are key components in reconfigurable MRRs. However, current thermo-optic (TO) and electro-optic (EO) devices on silicon photonics have several limitations. Besides their static power consumption, their long tuning length due to low tuning efficiency will decrease the free spectral range (FSR) of MRRs when they are incorporated as parts of the reconfigurable MRRs. In detail, TO phase shifters face inherent trade-offs between tuning efficiency and response time [9]. Usually, their tuning efficiency is about $20\text{mW/}\pi$ for μs-scale response time, resulting in $\sim 100\mathrm{\mu m}$ length for $\pi$-phase shifting. As for EO phase shifters that utilize free carrier dispersion, their phase-shifting efficiency is constrained to reduce free carrier absorption [10], resulting in sub-mm-scale to mm-scale $\pi$-phase-tuning lengths. As for tunable couplers, TO and EO effects cannot directly tune the coupling between two waveguides. Instead, they directly tune the effective indices of the waveguides within the couplers. The implementations of tunable couplers include Mach–Zehnder interferometers (MZIs) [11], multimode interference (MMI) couplers, and directional couplers with TO or EO phase shifters [12,13]. Thus, these tunable couplers share the same limitations as the phase shifters above. To accommodate more tunable components with larger FSR, cascaded MRRs have been proposed to extend the FSR through the Vernier effect [14].

As an emerging technology, micro-electromechanical systems (MEMS)-based low-power silicon photonic devices are promising for future practical large-scale integrated photonic chips. Compared to their TO and EO counterparts, electrostatically actuated MEMS phase shifters and tunable couplers have demonstrated high tuning efficiency, near-zero static power consumption, and μs-scale response time [15–19]. Moreover, these devices can operate at low temperatures, making them suitable for photonic circuits interfaced with superconducting single-photon detectors [20].

In 400G optical transport networks (OTNs), the baud rate per wavelength has already reached 50 GBd. In next-generation optical networks, this rate is anticipated to increase to 100 GBd, with the spectral range expanding to encompass both the C and L bands. Future high-speed elastic optical networks will require fully tunable wavelength-selective filters capable of adapting to any channel with variable symbol rates. For instance, to support optical networks covering the C band (1530–1565 nm), a single-channel optical filter with 35 nm FSR is required. Furthermore, to accommodate the variable channel spacings between 50 and 125 GHz for diverse high-speed signal rates, coupling tunability supporting a linewidth range of 0.4–1 nm is necessary. Moreover, if each port of the optical network requires flexible adaptation to different wavelength and bandwidth allocation schemes, the resonance of the filter at each port should be continuously tunable. However, as summarized in Table 1, current research on silicon photonic MEMS reconfigurable MRRs is still insufficient to meet the high-speed elastic DWDM network demands. In previous studies, numerous designs [21–26] based on the silicon photonic MEMS platform have only realized either phase or coupling tunability. The limited number of controllable degrees of freedom constrains their applicability in elastic optical network implementations. In fully reconfigurable ring resonator designs [27,28], the incorporation of multiple tunable elements inevitably requires an extended perimeter, consequently limiting the FSR to approximately 4 nm. This restricted FSR proves incompatible with the broad spectral bandwidth requirements of high-speed DWDM networks. Recent advancements in fully reconfigurable MEMS double-ring resonators have achieved remarkably narrow FSRs (0.1 nm and 0.125 nm for each single ring, respectively) and linewidths (0.013 nm) [29,30], yet these ultra-precise spectral properties simultaneously hinder their applicability in DWDM networks.Table 1.

Performance Comparison of Silicon Photonic MEMS Ring Resonators

In this work, new designs for fully reconfigurable silicon photonic MEMS MRRs have been proposed and implemented for high-speed elastic DWDM networks with channel spacing beyond $\sim 50\mathrm{GHz}$, including all-pass MRR (AP-MRR), add-drop MRR (AD-MRR), and add-drop double-MRR (AD-DMRR), with resonance tuning and linewidth tuning, as summarized in Table 1. For single ring devices, the FSRs are broadened as much as possible while ensuring full-FSR resonance tuning range, and high extinction ratio (ER) is simultaneously maintained. The AP-MRR has a 7 nm FSR and full-FSR resonance tuning range. The AD-MRR has a 3.5 nm FSR, full-FSR tuning range, and 0.5–1 nm FWHM tuning range. After pushing the performance of the single ring filters to near their limits, the AD-DMRR with the Vernier effect is further demonstrated, which features a 34 nm FSR with broad-range discrete tuning capabilities. The proposed devices are promising for practical applications, including WSS for ROADM, external cavities in tunable lasers, weight banks in photonic neural networks, and tunable delay lines and infinite impulse response filters in microwave photonic signal processing systems.

As illustrated in Figs. 1(a)–1(c), several suspended silicon waveguides and bars supported by MEMS structures are arranged around the suspended MRRs. The suspended silicon waveguides together with the ring waveguides form the symmetric directional couplers. The silicon bars that serve as phase shifters change the effective index of the ring waveguides by perturbing the evanescent field outside the ring waveguides. All the suspended silicon structures utilize two mechanical supporting schemes in our design. Tether-support structures [Figs. 1(h) and 1(i)] consist of mechanical supports at several discrete points with tethers while their other parts are fully suspended. Slab-support structures [Fig. 1(g)] are single-sided rib structures, in which only one side of them is fully connected with thin slabs. Using these two types of mechanical supports, MRRs are connected to central anchors, and the silicon waveguides and bars are connected to MEMS comb actuators, by which the in-plane distance between the silicon waveguides, bars, and MRRs can be tuned.

As shown in Fig. 2(a), moving combs and folded-flexure springs are connected to the shuttle beam, and the upper spring is connected to the ground plane. Meanwhile, the fixed combs are connected to the electrode isolated from the ground plane. When a constant voltage difference is applied between the fixed comb and the moving comb, the generated electrostatic force will move the shuttle beam in plane until it is balanced to the elastic force generated by the deformed spring. The displacement in mechanical equilibrium is governed by the virtual work principle of electrostatic force and Hooke’s law, which is given by $$F_x=\frac{1}{2}\frac{\partial C}{\partial x}{V}^2=k_{x}x,$$(1)where $x$ is the displacement component, $F_x$ is the Coulomb force in the $x$ direction, $C$ is the comb capacitance, $V$ is the voltage, and $k_x$ is the spring constant in the $x$ direction. The suspended structures are perforated to increase tuning speed by reducing total mass and to facilitate structure undercut during the hydrofluoric (HF) vapor releasing process.

Pushing-type electrostatic combs are employed in our designs, which allow for extremely small distances between the moving waveguides, bars, and MRRs while the initial gap size still meets the design rules of the fabrication processes. When silicon structures come into contact, if the restoring force of springs is insufficient to overcome the surface forces, such as capillary and van der Waals forces, stiction will occur. Thus, stoppers [Fig. 2(a)] with small contact areas have been designed to prevent this stiction.

As illustrated in Figs. 2(a)–2(d), silicon bars connected to the MEMS comb actuator can be pushed toward the waveguide in plane. The effective index of an optical waveguide will continuously increase when its evanescent field is perturbed by an approaching silicon bar. This will also lead to an increase in propagation loss due to the enhanced electric field at the dry-etched sidewalls with roughness [31]. Therefore, the distance between the silicon bars and the waveguide should be no less than 30 nm to prevent excessive propagation loss. To improve the tuning efficiency, the widths of the waveguides are designed as 350 nm because narrower waveguides with enhanced evanescent field are more susceptible to perturbation. To avoid coupling with the ring waveguides, these phase-shifting silicon bars are only 150 nm wide. The simulated change of $n_{\mathrm{eff}}$ with respect to the horizontal coupling gap is plotted in Figs. 2(e) and 2(f) (COMSOL Wave Optics). The $\mathrm{\Delta }{n}_{\mathrm{eff}}$ of the slab-support phase shifter is 0.1, which is less than 0.16 of the tether-support counterpart because the thin slabs for mechanical supports decrease the evanescent field of the waveguide.

As illustrated in Figs. 3(a)–3(d), for the tether-support tunable couplers, the moving waveguides are connected to the mechanical shuttle beam through waveguide crossings and are connected to the rest of the photonic circuit through rib-strip waveguide converters. The meandering waveguide sections connecting the rib-strip converters to the moving waveguides, which also serve as a flexure spring, are sufficiently compliant to minimize the deformation of the moving waveguides. Unlike EO and TO tunable couplers, comb actuators can change the distance between the two coupled waveguides so that the coupling strength is directly changed. The transfer matrix of the directional coupler according to coupled mode theory ($e^{j\beta z}$ for sign convention) is $$\left[\begin{array}{cc}\cos ({\beta }_{0}z)-j\frac{\mathrm{\Delta }\beta }{2{\beta }_0}\sin ({\beta }_{0}z)& j\frac{\kappa }{{\beta }_0}\sin ({\beta }_{0}z)\\ j\frac{\kappa }{{\beta }_0}\sin ({\beta }_{0}z)& \cos ({\beta }_{0}z)+j\frac{\mathrm{\Delta }\beta }{2{\beta }_0}\sin ({\beta }_{0}z)\end{array}\right]·e^{j(\frac{{\beta }_1+{\beta }_2}{2})z},$$(2)where ${\beta }_1$ and ${\beta }_2$ are propagation constants of the two waveguides, $\mathrm{\Delta }\beta ={\beta }_2-{\beta }_1$, ${\beta }_0=\sqrt{\mathrm{\Delta }{\beta }^2/4+{\kappa }^2}$, and $\kappa$ is the amplitude coupling coefficient [32]. The power coupling $K$ and additional phase $\phi$ are $$K={\left(\frac{\kappa }{{\beta }_0}\sin ({\beta }_{0}z)\right)}^2,$$(3)$$\phi =\text{angle}(\cos ({\beta }_{0}z)-j\frac{\mathrm{\Delta }\beta }{2{\beta }_0}\sin ({\beta }_{0}z)).$$(4)The function angle (z) will calculate the angle of complex number $z$. Then Eq. (2) can be reduced to $$\left[\begin{array}{cc}\sqrt{1-K}{e}^{j\phi }& j\sqrt{K}\\ j\sqrt{K}& \sqrt{1-K}{e}^{-j\phi }\end{array}\right]·e^{j(\frac{{\beta }_1+{\beta }_2}{2})z}.$$(5)The influence of $\phi$ on resonance will be discussed in Section 2.D. For the symmetric directional coupler (${\beta }_2={\beta }_1$), the power coupling coefficient is $$K={\sin }^2(\kappa z).$$(6)To improve the coupling strength, the widths of the waveguides are designed as 350 nm because narrower waveguides with an enhanced evanescent field are more strongly coupled. The simulated change of the $\kappa$ with respect to the horizontal coupling gap is plotted in Figs. 3(e) and 3(f). The $\mathrm{\Delta }\kappa$ of the slab-support tunable coupler is $0{\text{.}07\mathrm{\mu m}}^{-1}$, which is less than $0.2{\mathrm{\mu m}}^{-1}$ of the tether-support counterpart because the thin slabs for mechanical supports decrease the evanescent field of the waveguide.

As illustrated in Fig. 1, several phase shifters and tunable couplers are arranged around the suspended MRRs for linewidth tuning and resonance tuning. As for the linewidth tuning, the FWHM of the AP-MRR and AD-MRR is given by [33] $${\mathrm{FWHM}}_{\mathrm{AP}}=\frac{(1-t\alpha )·{\mathrm{FSR}}_{\mathrm{AP}}}{\pi \sqrt{t\alpha }},$$(7)$${\mathrm{FWHM}}_{\mathrm{AD}}=\frac{(1-t_1{t}_2\alpha )·{\mathrm{FSR}}_{\mathrm{AD}}}{\pi \sqrt{t_1{t}_2\alpha }},$$(8)where $t$ is the amplitude transmission in the coupler of the AP-MRR, $\alpha$ is the round-trip amplitude transmission, taking into account both propagation loss in the ring and excess loss in the coupler(s) (zero loss: $\alpha =1$), and $t_1$ and $t_2$ are the amplitude transmission of the couplers for the through port and drop port, respectively, in the AD-MRR. Thus, the linewidths of MRRs can be tuned through (1) the tunable couplers by changing the coupling ($t$) between the moving waveguides and MRRs and (2) the phase shifters due to the additional loss ($\alpha$) induced by the silicon bars that approach the MRRs. Assuming the $t_{\min }$ ($t_{1\min }$ or $t_{2\min }$) is already estimated from the desired linewidth tuning range according to Eq. (7) or (8), there is no excess loss in couplers, and the coupling coefficient $\kappa$ in Eq. (2) increases monotonically with respect to wavelength, then the length of coupling region is given by $$L_{\mathrm{tc}}=\frac{\arcsin \sqrt{1-t_{\min }^2}}{\kappa ({\lambda }_{\min })}.$$(9)

Note that $K+t_{\min }^2=1$ according to Eq. (5). To ensure sufficient tolerance for the uncertainties in experiments, such as the propagation loss of MRRs and fabrication errors, tunable couplers have been designed to achieve tunability from complete decoupling to full coupling. The tunable couplers in the AP-MRR and AD-MRR have been designed to achieve full coupling at the 215 nm gap with a length of 12 μm. Meanwhile, due to lower coupling efficiency of the slab-support tunable couplers in the AD-DMRR, they have been designed to achieve full coupling at the 100 nm gap with a length of 15 μm. In practice, due to the residual stress relaxation, the moving waveguides may buckle and offset vertically, and their coupling efficiency will be reduced. Nevertheless, in this case, the coupling gap can be tuned to below the designed full-coupling gap to compensate the reduced coupling due to the potential vertical offset.

As for the resonance tuning, both the phase-shifter gap and dispersion should be taken into account for the shifted resonant wavelength. In a simplified model as illustrated in Fig. 4(a), the resonance tuning can be expressed as $$(n({\lambda }_0,g_0)+\mathrm{\Delta }n)L_{\mathrm{ps}}+(n({\lambda }_0,\infty )+\frac{\partial n}{\partial \lambda }\mathrm{\Delta }\lambda )(L-L_{\mathrm{ps}})=m({\lambda }_0+\mathrm{\Delta }\lambda ),$$(10)where $L$ is the round-trip length, $L_{\mathrm{ps}}$ is the phase-shifting length, $m$ is the azimuthal order of the MRR resonance, $n(\lambda ,g)$ is the effective index as a function of wavelength and the gap within the phase shifter, and $\mathrm{\Delta }\lambda ={\lambda }_1-{\lambda }_0$ is the change from initial resonant wavelength ${\lambda }_0$ to shifted resonant wavelength ${\lambda }_1$, as the phase shifter gap is changed from $g_0$ to $g_1$, and $\mathrm{\Delta }n=n({\lambda }_1,g_1)-n({\lambda }_0,g_0)$. To simplify the following derivations, we assumed that the initial phase shifter gap is sufficiently large, such that $g_0=\infty$, and then the required phase-shifting length is given by $$L_{\mathrm{ps}}=\frac{n_g({\lambda }_0,\infty )\mathrm{\Delta }\lambda L}{\mathrm{\Delta }n{\lambda }_0+(n_g({\lambda }_0,\infty )-n({\lambda }_0,\infty ))\mathrm{\Delta }\lambda },$$(11)where $n_g({\lambda }_0,\infty )$ is the group index of MRR waveguides. Then the phase-shifting length for full-FSR resonance tuning is obtained when $\mathrm{\Delta }\lambda =\mathrm{FSR}$. Substituting $\mathrm{\Delta }\lambda =\mathrm{FSR}={\lambda }_0^2/n_{g}L$ into Eq. (11), the phase-shifting length for full-FSR resonance tuning is given by $$L_{\mathrm{ps}}^{\mathrm{FSR}}=\frac{{\lambda }_0^2}{\mathrm{\Delta }n{\lambda }_0+(n_g({\lambda }_0,\infty )-n({\lambda }_0,\infty ))·\mathrm{FSR}}.$$(12)

According to Eq. (12), assuming that the full-FSR resonance tuning is achieved at the 60 nm gap, the AP-MRR with the 77 μm perimeter should have a 16.4 μm $L_{\mathrm{ps}}^{\mathrm{FSR}}$, and the AD-MRR with the 134 μm perimeter should have a 17.4 μm $L_{\mathrm{ps}}^{\mathrm{FSR}}$. The perimeters of the triangular AP-MRR and octagonal AD-MRR are designed as short as possible while providing a sufficient length to accommodate the phase shifters, tunable couplers, and tether supports.

According to Eq. (5), the tunable coupler will introduce additional $\kappa$-dependent phase tuning, when the directional couplers are asymmetrical. In practice, the asymmetry of directional couplers is primarily attributed to fabrication errors that result in different widths of the two waveguides. Thus, the change of additional phase $\mathrm{\Delta }\phi$ with respect to $\kappa$ under different width differences $\mathrm{\Delta }w$ is plotted in Fig. 4(b). Fully suspended waveguides with 350 nm width and 12 μm length are used in calculations. Note that the sign of $\mathrm{\Delta }\phi$ is determined by the sign of $\mathrm{\Delta }w$, which means the resonance may randomly blue shift or red shift as the $\kappa$ increases due to the random variation of $\mathrm{\Delta }w$ after fabrication.

As illustrated in Fig. 5(b), the AD-DMRR enlarges its effective FSR and resonance tuning range through the Vernier effect. Specifically, the resonance tuning of the AD-DMRR is achieved by tuning the resonance of its two rings simultaneously and then aligning them at a targeted wavelength. The effective FSR of the AD-DMRR is $${\mathrm{FSR}}_{\mathrm{eff}}=n{\mathrm{FSR}}_1=m{\mathrm{FSR}}_{2}m,n\in \mathbb{N},$$(13)where ${\mathrm{FSR}}_1$ and ${\mathrm{FSR}}_2$ are the FSR of its two MRRs. Figure 5(a) presents a simplified model for AD-DMRR resonance-tuning simulation. The MRRs in the AD-DMRR configuration have perimeters of 91 μm and 109 μm, corresponding to FSRs of 5 nm and 6 nm, respectively. According to Eq. (13), the designed effective FSR of the AD-DMRR is 30 nm. Following Eq. (11) with minor algebraic manipulation, the mapping between the target resonance tuning range $\mathrm{\Delta }\lambda$ relative to ${\lambda }_0$ and the required $\mathrm{\Delta }n$ can be established once the $L_{\mathrm{ps}}$ and initial resonant wavelength ${\lambda }_0$ are determined: $$\mathrm{\Delta }n=\left(\right(\frac{L}{L_{\mathrm{ps}}}-1)n_g({\lambda }_0,\infty )+n({\lambda }_0,\infty ))\frac{\mathrm{\Delta }\lambda }{{\lambda }_0}.$$(14)

The phase-shifting lengths $L_{\mathrm{ps}}$ for Ring1 and Ring2, as illustrated in Fig. 5(a), are designed to achieve full-FSR resonance tuning for each single ring. According to Eq. (12), assuming that the full-FSR resonance tuning is achieved at the 30 nm gap, Ring1 with 91 μm perimeters requires a 19.3 μm $L_{\mathrm{ps}}^{\mathrm{FSR}}$, and Ring2 with 109 μm perimeters requires a 19.6 μm $L_{\mathrm{ps}}^{\mathrm{FSR}}$. To compensate for potential vertical offset-induced tuning efficiency reduction, the phase-shifting lengths in both rings were extended to 24 μm. Based on the above parameters, Fig. 5(c) illustrates the target wavelength ${\lambda }_t={\lambda }_m+\mathrm{\Delta }\lambda ({\lambda }_m)$ versus the required $\mathrm{\Delta }n$ for different resonant wavelengths ${\lambda }_m$ of Ring1 and Ring2 across from 1525 nm to 1565 nm. Additionally, Fig. 5(c) indicates the specific modes requiring tuning for target wavelength synthesis. Obviously, the sufficient condition for the AD-DMRR to be continuously tunable over the entire effective-FSR is the full-FSR-tunability of its two single rings. Figure 5(d) illustrates the simulation results for the effect of dispersion and gap on the $n_{\mathrm{eff}}$ of the slab-support phase shifter. By using $n({\lambda }_t,g_r)=n({\lambda }_m,\infty )+\mathrm{\Delta }n$, the required gap $g_r$ can be mapped in Fig. 5(d). Figure 5(e) illustrates the simulated transmission spectrum at the drop port of the AD-DMRR with 30 nm FSR and 30 nm continuous resonance tuning.

As for the AD-DMRR, the coupling strength between its two rings will determine its resonance-splitting for the tunable flat-top band. However, it is infeasible to accommodate another MEMS comb actuator between the two rings due to the limited layout area. Instead of tuning the coupling between the rings, we slightly detune the resonance of the two rings to realize the flat-top band. Thus, the resonance-splitting introduced by ring-ring coupling should be negligible for the entire flat-top tuning range. To determine the design parameters of the AD-DMRR, we have resorted to numerical simulations because analytical derivations are cumbersome and unnecessary. To choose the suitable ring-ring coupling strength, assuming the two couplers between the MRRs and the waveguides share the same power coupling coefficient $K_{\mathrm{wr}}$, the simulated results shown in Fig. 6(a) illustrate that the ring-ring power coupling coefficient $K_{\mathrm{rr}}$ should be less than 1/10 of $K_{\mathrm{wr}}$ (assuming 7500 intrinsic Q). Higher $K_{\mathrm{wr}}$ can realize lower insertion loss but also increase its FWHM, so we choose the $K_{\mathrm{rr}}$ ranging from 0.003 to 0.03, corresponding to the gap ranging from 300 to 150 nm, as illustrated in Fig. 6(b). The simulation results above provide rough estimations for the sweeping range of design parameters for the AD-DMRR.

The devices were fabricated on SOI wafers with a 220-nm-thick top silicon layer and a 2-μm-thick BOX layer. Initially, ridge and strip waveguides were patterned using electron beam lithography (EBL), followed by 70 nm shallow etching and 220 nm full etching processes, respectively. Subsequently, metal electrodes consisting of 50-nm-thick chrome and 300-nm-thick gold films were patterned with photolithography, followed by electron beam evaporation and lift-off processes. Finally, the suspended structures of the devices were released from the BOX layer with HF-vapor etching. Areas other than the suspended structures and waveguides were covered with gold to serve as ground planes. The electrodes of the fixed combs were isolated from the ground plane by a 3-μm-wide deep trench. The MEMS structure itself has a footprint of $87\mathrm{\mu m}\times (74-125)\mathrm{\mu m}$.

To demonstrate the resonance and linewidth tuning, the transmission spectra at different actuation voltage combinations were measured using a tunable laser and an optical power meter. The optical signal couples into/out of the chip through grating couplers. In order to preclude the occurrence of thermo-optic bistability or optomechanical oscillations [34,35], a variable optical attenuator was employed during the test. The metal electrodes of the devices were connected to a multi-channel voltage source through direct current (DC) probes.

Figure 7 illustrates the linewidth and resonance tuning at the through port of the AP-MRR shown in Fig. 7(a), and the results are normalized. As illustrated in Fig. 7(c), the measured maximum ER is about 26 dB when critical coupling is achieved. As the voltage of the tunable coupler ($V_{\mathrm{TC}}$) increases from 23 to 27 V, the AP-MRR changes from under-coupling to over-coupling, while FWHM changes from 0.44 to 0.85 nm. Meanwhile, the resonance blue shifts 0.15 nm because the tunable coupler also introduces additional phase change, which is a side effect as explained in Section 2.D. As illustrated in Fig. 7(d), the FSR of the AP-MRR is 6.7 nm, and full-FSR resonance tunability has been achieved. As the voltage of phase shifter ($V_{\mathrm{PS}}$) increases from 0 to 31.6 V, the resonance red shifts from 1542.9 to 1549.4 nm while the linewidth changes from 0.41 to 0.51 nm.

Figure 8 illustrates voltage sweeping plots of the linewidth and resonance tuning at the through and drop ports of AD-MRRs shown in Fig. 8(a). The measured AD-MRRs exhibit lower driving voltage than AP-MRRs due to their smaller initial gaps of phase shifters and tunable couplers. In the measurement, two independent phase shifters are equipotential, corresponding to the $V_{\mathrm{PS}}$ in Fig. 8(a). The resonance shift induced by the tunable coupler is random as interpreted in Section 2.D; thus, Fig. 8(a) only shows the resonance shift caused by $V_{\mathrm{PS}}$ with certain voltages of tunable couplers ($V_{\mathrm{TC}1}=5\mathrm{V}$ and $V_{\mathrm{TCr}}=6\mathrm{V}$). By increasing $V_{\mathrm{PS}}$ from 15 to 20.8 V, the resonance red shifts from 1548.36 to 1552.39 nm. According to the measured data, the FSR of the AD-MRR is 3.5 nm, and full-FSR resonance tunability has been achieved. After the resonant wavelength is determined by $V_{\mathrm{PS}}$, the linewidth of the resonance can be tuned by adjusting $V_{\mathrm{TCl}}$ and $V_{\mathrm{TCr}}$ according to Figs. 8(d)–8(k) at certain $V_{\mathrm{PS}}$. Figures 8(d)–8(g) illustrate the FWHM tuning at the through port, and Figs. 8(h)–8(k) illustrate the FWHM tuning at drop port. The FWHM of the through port changes from 0.5 to 0.88 nm, while the FWHM at the drop port changes from 0.52 to 0.98 nm; as the $V_{\mathrm{PS}}$ changes from 15 to 20.8 V, $V_{\mathrm{TCl}}$ changes from 5 to 11 V, and $V_{\mathrm{TCr}}$ changes from 6 to 10 V.

Figure 9(c) illustrates the resonance tuning at the drop port of the AD-DMRR with a 275 nm ring-ring gap as shown in Fig. 9(a). The gray background in Fig. 9(c) indicates that the resonance in this wavelength region is continuously tunable. The resonance tuning illustrated in Fig. 9(c) is achieved by aligning the resonance of its rings at a new wavelength at certain voltages of tunable couplers, where the ER reaches 19 dB. Figure 9(c) demonstrates the tuning range based on the Vernier effect. While simulations predict that the AD-DMRR should achieve full-FSR tuning, structural deformation induced by stress release during fabrication hinders coplanar alignment between the phase shifter and ring waveguide. This misalignment results in reduced phase-shifting efficiency, ultimately restricting the individual ring’s tuning range. In Fig. 9(c), the resonance tuning ranges of the phase shifters were 2.1 nm and 0.5 nm, respectively. As a consequence, experimental measurements reveal that the AD-DMRR exhibits continuous tunability only within several discrete spectral regions. The flatness factor is defined as the ratio of the 1 dB bandwidth to the 3 dB bandwidth. The closer this ratio is to one, the flatter the spectral shape. Figure 9(b) presents the filter’s flat-top operational characteristics. When $V_{\mathrm{PSl}}$ decreased from 10.2 to 8 V, insertion loss increased from 4 to 12 dB, and the 3 dB bandwidth expanded from 0.23 to 0.44 nm. The flatness factor increased from 0.61 to 0.78, as the $V_{\mathrm{PSl}}$ decreased from 10.2 to 9.5 V. However, when $V_{\mathrm{PSl}}$ is 8 V, the passband ripple reaches 3 dB, which may be unacceptable for DWDM applications. In practical multi-filter cascading configurations, the significant insertion loss penalty may potentially outweigh the benefits of flat-top operation. Nevertheless, when a sufficient optical power budget is available, flat-top operation can substantially improve the system’s tolerance to thermal drift in both filter and laser center wavelengths.

Figure 10(a) illustrates the schematic of the setup for transient measurement. A 1 kHz square wave signal with a 50% duty cycle, generated by a signal generator, is amplified by a voltage amplifier with a gain of 20 before being applied to a tunable coupler through a DC probe. As shown in Fig. 10(b), the measured tunable coupler, featuring an initial gap of 1 μm, requires an applied bias voltage of 22 V to achieve the appropriate DC operating point. Figure 10(c) illustrates the measured normalized optical power and the output voltage of the signal generator. The output voltage of the signal generator ranges from 0.85 to 1.35 V, corresponding to the output voltage of the voltage amplifier from 17 to 27 V. The optical power is normalized using the output optical power when the gap is 1 μm. The measured 10%–90% rise time is 5.4 μs, while the 90%–10% fall time is 94 μs. The prolonged fall time is primarily caused by the limited discharge current resulting from the Schottky contact formation between the metal electrode and the silicon substrate under reverse bias conditions. If ohmic contacts were used instead, the fall time would be expected to be similar to the rise time.

We have demonstrated novel and versatile reconfigurable MRRs with independent tunable MEMS phase shifters and tunable couplers on a silicon photonics platform, including various configurations such as AP-MRR, AD-MRR, and AD-DMRR. The AP-MRR achieves full-FSR tunability with 7 nm FSR and 26 dB ER, demonstrating the largest FSR and the widest FSR-limited effective resonance tuning range among all devices listed in Table 1 featuring both resonance and coupling tunability. The AD-MRR demonstrates full-FSR with 3.5 nm FSR, 0.5–1 nm linewidth tuning range, and 5 dB insertion loss at the drop port. These performance parameters are comparable to previously reported devices [27,28], suggesting limited potential for further significant improvements through conventional approaches. Due to structural deformation induced by silicon layer stress release during fabrication, the efficiency of the AD-DMRR’s phase shifter was compromised. This constraint prevented the AD-DMRR from achieving the simulated 34 nm continuous full-FSR resonance tunability, limiting the device to broad-range discrete tuning capabilities. Despite this limitation, the implemented devices demonstrate superior performance compared to previous MEMS double-ring structures [29,30] for high-speed DWDM networks, which typically exhibit FSRs below 1 nm. This advancement positions our device as a promising solution to high-speed elastic DWDM network applications. Furthermore, the continuous full-FSR tunability for the AD-DMRR can be attained through stress management technologies.

To enhance tuning efficiency, the external evanescent field is strengthened by the designed 350 nm waveguide. However, combined with the substantial refractive index contrast at the air-silicon interface, the waveguide scattering loss is significantly enhanced. This optical loss can be substantially reduced through optimization of nanofabrication processes. Specifically, low-loss waveguide technologies compatible with MEMS fabrication processes have been reported [31]. The device exhibits a typical response time of approximately 5.4 μs. The observed anomaly in fall time is attributed to the limited reverse bias discharge current at the unstable Schottky contact, resulting from the silicon-metal contact during fabrication. Implementing an ohmic contact process during manufacturing can effectively mitigate this issue. In addition to the overall improvement, performances of the above devices can also be optimized for specific applications. For example, only a small power coupling coefficient tuning range is needed for filters in WDM. Thus, the tunable coupler can be optimized for lower driving voltage by designing an appropriate initial distance between MRRs and moving waveguides. Moreover, the full tuning range of phase and coupling is valuable to photonic systems for universal purposes, such as programmable photonics [19,36].

Acknowledgment. The authors thank the ZJU Micro-Nano Fabrication Center and the Westlake Center for Micro/Nano Fabrication and Instrumentation for the facility support.