1. Introduction

With ever-increasing demand of global IP traffic, the large volume of information in the limited capacity of the physical channels is causing bottleneck in data communication [1]. This imposes a challenge on bandwidth density, computation speed, and energy consumption in future zetta-scale data centers [2–4]. To meet the escalating bandwidth-demand and energy-cost, mode division multiplexing (MDM) has emerged as an alternate multiplexing method, along with wavelength division multiplexing (WDM) [5–7]. MDM in silicon-on-insulator (SOI) is promising for high-capacity on-chip optical links and high-throughput optical switches due to the large difference in refractive indices between the silicon waveguide core and the surrounding oxide (SiO₂). MDM silicon photonics systems offer efficient generation, conversion, propagation and phase-tuning of orthogonal guided optical modes enabling the deployment of on-chip multimode reconfigurable optical space switches [8,9].

Multimode switches are a key component in an MDM based silicon photonics system, which can switch payload data between different spatial channels exploiting multiple orthogonal optical modes. Significant research effort is given towards the development of mode-multiplexed silicon photonics switches. On-chip two-mode switch with data exchange capability is reported using micro-ring resonators (MRRs) [10,11]. Reconfigurable switching between two spatial modes are demonstrated using symmetric Y-junction [12], multimode interference (MMI) couplers [13], and three-dimensional directional coupler [14]. On-chip mode-selecting switch is reported using thermo-optically tuned MMI based Mach-Zehnder interferometers [15], and asymmetric directional couplers [16]. MDM compatible matrix switches are demonstrated based on 2×2 four-mode switch element using adiabatic directional coupler (ADC) based mode (de)multiplexer [17], and 1×1 multimode Mach-Zehnder interferometers (MZIs) [18]. For reconfigurable switching, MMIs are an attractive design choice for their compactness, fabrication tolerance, and higher switching extinction ratio for a comparable insertion loss (approximately 0.2 dB). Recently, we reported on a scalable mode switch using MMI based thermally controlled MZI structure [19].

In this work, we demonstrate a three-mode switch (3MS) enabling reconfigurable switching of the first three transverse electric (TE) modes in the C-band using a 3 ×3 unbalanced MZI (UMZI) formed of a 120°optical hybrid MMI coupler. To the best of our knowledge, this is the first experimental demonstration of an on-chip 3MS using 120° optical hybrid. The 3MS exploits the relative phase difference of the optical hybrid to enable interferometric switching among the output ports of the UMZI. An ADC based 3-mode multiplexer and MMI based mode decomposer, reported in [20], are used to excite the TE0, TE1 and TE2 modes, and then demultiplex them into the fundamental mode components for phase-tuning. Titanium/tungsten (Ti/W) based thin film metal heaters are used as phase-shifters. The 3MS exhibits less than 12.0 μs switching time and greater than 12.0 dB switching extinction ratio (ER) while dynamically switching high-speed non-return-to-zero (NRZ) 2³¹−1 pseudo random bit sequence (PRBS31) optical payloads for an aggregated bandwidth of 30 Gb/s (3×10 Gb/s). An average power consumption of the heater of 94.8 mW is measured enabling energy efficient 3-channel switching.

2. Design and working principle of the three-mode switch (3MS)

The 3MS is designed for TE polarization using 220 nm thick silicon channel waveguides, surrounded by 2.0 μm buried oxide (BOX) and 2.2 μm cladding layers of SiO₂. Ti/W thin film of 200 nm thickness and 6 μm width is deposited over the silicon waveguide as metal heater phase-shifter. The waveguide cross-section is shown in Fig. 1(a). The waveguide widths for TE0, TE1 and TE2 mode propagations are optimized to be 0.5 μm, 1.0 μm and 1.45 μm, respectively according to the eigenmode simulation using a commercial CAD tool (Lumerical Mode solution), as shown in Fig. 1(b).

 

Fig. 1 (a) Cross-section of the single-mode (TE0), and multimode (TE1, and TE2) waveguides with Ti/W metal heater phase-shifter. The heater is placed 2.0 μm over the single-mode waveguide; (b) simulated effective refractive indices for the first four TE modes showing the simulated electric field for each mode.

Download Full Size | PDF

The proposed three-mode switch, as shown in Fig. 2, consists of a tapered mode decomposer MMI (MMI-A) followed by the 3×3 unbalanced MZI (UMZI), comprised of two tapered cascaded MMIs (MMI-B and MMI-C). Four 200 μm long metal heaters are used as phase-shifters. The PS-1a and the PS-1b phase-shifters are used for tuning the relative phases of the decomposed mode components to optimize the insertion loss (IL) and the crosstalk. The PS-2a and the PS-2b phase-shifters are used for applying required phase-shift in the UMZI to implement reconfigurable switching. The arms of the MMIs containing the phase-shifters are bent to reduce the device footprint using 90° bends of 15 μm radii. The widths of all MMIs are 6 μm. The detail design and dimensions of the MMIs are reported in [19].

 

Fig. 2 2D schematic of the three-mode switch (3MS) with three multimode inputs denoted as TE0-in, TE1-in0, and TE2-in; and three single mode outputs denoted as Out1, Out2, and Out3. the input ports of MMI-B (120° hybrid) are denoted as I1, I2 and I3, and its output ports are denoted as O1, O2 and O3.

Download Full Size | PDF

The three TE modes, i.e., TE0, TE1 and TE2, are first multiplexed using a broadband ADC based mode multiplexer. The dimension and normalized optical transmission in this mode-multiplexer are shown in the experimental result section (Figs. 9(a)–9(d)). After different optical modes are multiplexed in the mode-multiplexer, the mixed-mode signal is demultiplexed, and decomposed to its fundamental components in MMI-A, which is a reconfigurable multimode demultiplexer/switch (RMDS), as reported in [20]. The decomposed mode components for the TE0-in, TE1-in and TE2-in input modes are mapped to the MMI-A output ports as follows (Fig. 3):

 

Fig. 3 Simulated electric fields of the MMI-A (left) and MMI-B (right) at 1550 nm. The top (a, b), middle (c, d) and bottom (e, f) images represent the propagation of TE0, TE1 and TE2 modes, respectively.

Download Full Size | PDF

A longer MMI can potentially improve the conversion efficiency and reduce the scattering loss for the TE2-in input mode [21], but significantly increases the device length hence this method is avoided. Another approach is to use cascaded MRRs, but limiting the operating range of the 3MS due to the wavelength sensitivity of the micro-rings [10]. Using asymmetric directional couplers [22] and cascaded asymmetric Y-junctions [23] can improve the TE2→TE0 conversion efficiency. These devices, however, are more susceptible to fabrication process variation than MMIs. The outputs of the MMI-A are coupled to MMI-B’s middle input port (I2) for the TE0-in and TE2-in input modes, and to the upper (I1) and the lower (I3) input ports for the TE1-in input mode, as shown in Fig. 3. The incoming light is equally distributed among the output ports of MMI-B for TE0 and TE2 modes. The imbalance in power splitting observed in the TE1 mode (Fig. 3(d)) is compensated by the PS-2a and PS-2b phase-shifters.

2.1. Design of the 120° optical hybrid

MMI-B and MMI-C are identically designed to operate as a 120° optical hybrid [24], where each of the output ports (O1, O2, and O3) receives one-third of the input optical power from any of the input ports (I1, I2, and I3). According to the principle of general interference, the optical phases of a N×N MMI are determined by Fourier analysis in [25] and can be expressed as

For a 6.0 μm width and 89.1 μm estimated beat length, the first (M = 1) three-fold (P = 3) image appears at approximately 89.1 μm away from the input location. A length sweep of MMI-B using Lumerical’s eigenmode expansion (EME) simulation estimates the optimal length to be 90.0 μm to achieve 33% power splitting and 120° relative phase-shift in each output port regardless of the input ports, as shown in Fig. 4. The length and width of the taper, and the locations of the input/output ports are also optimized using EME simulation with a 20 nm uniform mesh in all cases.

 

Fig. 4 Simulated optical transmission (left) and relative phase difference in the output ports of the 120°optical hybrid (MMI-B and MMI-C) as a function of MMI length. The 2D schematic of the MMI is shown above the transmission plot.

Download Full Size | PDF

2.2. Design of the 3×3 Unbalanced Mach-Zehnder interferometer (UMZI)

The 3×3 UMZI is designed analytically using the transfer matrix method [26]. Although the more rigorous EME method can be used to generate the scattering matrix (S-matrix) utilizing bidirectional wave propagation, a large number of modes need to be considered to ensure the EME accuracy. This increases calculation time and complexity, especially in a phase sensitive device like the UMZI. Alternately, the TMM is attractive for its ability to decompose a complex structure into smaller components and calculate the transfer matrix of each component to directly generate the transmission spectrum of the complete device.

First, the analytical model is applied on a balanced 3×3 MZI to understand the switching in the ideal case. Then the transfer matrix of the UMZI is developed by compensating the optical phase delay associated with the bends and unbalanced arms. The characteristic equation for the output transmission of a balanced 3×3 MZI is defined as:

Where U_I and U_out are the input and output matrices of the switch, U is the transfer matrix of the 3×3 general interference MMI (MMI-B and MMI-C), and Φ is the matrix of the phase-shifters. Using the amplitudes and phase-shifts of the 120° hybrid from Fig. 4, the transmission matrix of each input to each output of the 3×3 balanced MZI switch is generated in Matlab with respect to PS-2a and PS-2b phase-shifters. This phase matrix is shown in Fig. 5. The applied phases for each individual transmission comply with the 2π/3 phase requirement of the 120° hybrid. The optimal transmission maintains three groups of symmetry in phases I, II, III, which is summarized in Table 1:

 

Fig. 5 Simulated phase-matrix of the 3×3 balanced MZI showing optical transmission for the inputs (I1, I2 and I3) to outputs (Out1, Out2 and Out3) as a function of the combined phase-shifts of the PS-2a and the PS-2b phase-shifters. The schematic of the 3×3 MZI is shown above.

Download Full Size | PDF

Table 1. Phase-symmetry in the 3×3 balanced MZI

View Table | View all tables in this article

In the UMZI, the optical path difference, ΔL, between the bent waveguides on the upper and the lower arms along the PS-2a and the PS-2b phase-shifters, and the smaller straight waveguide in the middle arm results in a phase difference of Δφ defined by,

This phase difference is taken into account in the transfer matrix of the UMZI to determine the required phase shift for reconfigurable switching. The phase-compensated transfer function of the UMZI becomes:

The three groups of symmetry I, II and III still applies in UMZI but the 2π/3 phase requirement no longer exists. The transmission matrix of the UMZI with the corrected phase shift in PS-2a and the PS-2b is shown in Fig. 6. In the 3MS, the UMZI needs to be tuned according to these phases to achieve reconfigurable switching.

 

Fig. 6 Simulated phase-matrix of the 3×3 unbalanced MZI (UMZI) showing optical transmission for the inputs (I1, I2 and I3) to outputs (Out1, Out2 and Out3) as a function of the combined phase-shifts of the PS-2a and the PS-2b phase-shifters. The schematic of the 3×3 UMZI is shown above.

Download Full Size | PDF

The UMZI transfer-matrix in the Eq. (6) is solved in Matlab by taking the phase shifts of Fig. 6 into account. The solutions are plotted in Figs. 7(a) and 7(b) for I2 and (I1+I3) inputs, respectively. I2 input corresponds to the TE0-in/TE2-in input and (I1+I3) corresponds to the TE1-in input, as shown in Fig. 3. The switching states are determined for the maximum transmission in the expected output port and, at the same time, the minimum transmission in the crosstalk ports. The optimal transmissions are encircled by black dotted lines in each plot. For the I2 input condition (TE0-in/TE2-in input mode) in Fig. 7(a), switching occurs between Out1 and Out3 output ports with the optimal IL and crosstalk. From Fig. 6, the I2→Out2 switching should occur when PS-2a = PS-2b = 45°. As both phase-shifters simultaneously need a non-zero phase shift, the manual tuning of the test bed makes it difficult to achieve of the precise bias voltage needed. Moreover, the relative phase difference between the UMZI arms varies from the theoretical value due to the fabrication non-uniformity making phase tuning more challenging. An automated biasing technique is required to precisely tune both the phase-shifters to enable switching at all output ports. This two-port switching will potentially limit the reconfigurability of the device but it is necessary for error-free data transmission. The phase condition required for three-port switching causes large crosstalk (more than 7% power leakage in each of the output ports), which will degrade the signal integrity and eye quality during high-speed data transmission. For the (I1+I3) input condition (TE1-in input mode) the optimal transmissions enable reconfigurable switching among all three output ports, as shown by the delimited dotted black line in Fig. 7(b).

 

Fig. 7 Solutions to the UMZI transfer matrix as a function of the combined phase shifts of PS-2a and PS-2b phase-shifters showing optical transmission and switching in all three output ports for (a) the I2 input port (TE0-in and TE2-in modes), and (b) the (I1+I3) input ports (TE1-in mode). The phase condition for the optimal transmission in each case is delimited by a black dotted line.

Download Full Size | PDF

3. Fabrication and characterization of the three-mode switch (3MS)

The 3MS is fabricated through Applied Nanotools Inc. in Alberta, Canada. The silicon device layer is patterned using a 100 keV electron-beam lithography (EBL) followed by an inductively coupled plasma-induced reactive ion etching (ICP-RIE) process. The Ti/W thin film as metal heater and aluminum thin film for metal routing are deposited using electron-beam evaporation. A thin (300 nm) SiO₂ passivation layer is deposited by chemical vapor deposition (CVD) to protect the metal layers. The optical micrograph of the fabricated chip is shown in Fig. 8(a). Surface grating couplers (GCs) are used to vertically couple the optical signal to and from the chip. The device footprint is estimated to be 0.157 mm² excluding the optical and electrical I/O ports.

 

Fig. 8 (a) Optical micrograph of the fabricated 3MS showing the complete device footprint excluding electrical pads; (b) experimental setup for the high-speed data transmission measurement for the 3MS. The optical and the electrical connections are shown as black and blue lines, respectively. PC: polarization controller; MZI: Mach-Zehnder interferometer modulator, EDFA: Erbium doped fiber amplifier, DUT: device under test, BPF: band-pass filter, PD: photodetector, RTO: real-time oscilloscope, DCA: digital communication analyzer, CLK: clock synthesizer, PPG: pulse pattern generator, NRZ: non-return-to-zero data signal.

Download Full Size | PDF

The 3MS is first characterized with continuous wave (CW) optical input. The polarization controlled optical input from a tunable C-band laser is swept from 1520 nm to 1600 nm wavelength, and the output optical power is measured by an optical power meter. The IL and the crosstalk of the switch are minimized by tuning the DC bias voltages applied to the phase shifters. Fig. 8(b) is the experimental set up for the high-speed data transmission measurement, where the 10 Gb/s NRZ PRBS31 payload, generated by a pulse patter generator (PPG), modulates the CW optical signal. The modulated and amplified signal is transmitted through the device under test (DUT), and then detected by a 46 GHz off-chip photoreceiver of 0.7 A/W sensitivity. The eye diagrams are recorded by a digital communication analyzer (DCA). The dynamic switching response is recorded in real-time using a 350 MHz oscilloscope (RTO) by applying a 18.6 kHz electrical square wave gating signal from the PPG to the PS-2a phase-shifter. The other phase-shifters are biased at fixed voltages to achieve minimum IL and crosstalk.

3.1. Optical transmission characterization

A test structure of the ADC based 3-mode (de)multiplexer is first characterized with a CW optical input. Fig. 9(a) is the schematic of this ADC based mode (de)multiplexer and Figs. 9(b)–9(d) are the normalized transmissions for the optical inputs for (de)multiplexing of TE0, TE1 and TE2 modes, respectively. As the optical I/Os, i.e., the GCs, are designed for single-mode coupling, the multiplexed modes need to be demultiplexed and converted to the fundamental modes to be detected at the output GCs. The IL varies from −0.3 dB to −2.9 dB for the TE0 mode, from −0.9 dB to −3.7 dB for the TE1 mode, and from −0.3 dB to −2.2 dB for the TE2 modes from 1520 nm to 1600 nm operating range. The worst crosstalks are −18.3 dB, −17.0 dB, and −19.6 dB for TE0, TE1 and TE2 modes, respectively.

 

Fig. 9 (a) 2D schematic of the ADC based 3-mode (de)multiplexer; the normalized optical transmissions of this (de)multiplexer as a function of wavelength are shown in (b) for the TE0 mode, in (c) for the TE1 mode and in (d) for the TE2 mode.

Download Full Size | PDF

The optical transmission in the 3MS, as a function of wavelength from 1520 nm to 1600 nm, normalized to a small waveguide test structure connecting two GCs, are shown in Fig. 10. TE0-in, TE1-in and TE2-in input modes are shown in the upper, middle and lower rows, respectively. The imbalance in effective optical path lengths between the spiral bent waveguide in the upper and the lower arms, and the straight waveguide in the middle arm of the UMZI results in an interferometric response with a small (∼1.5 nm) free spectral range (FSR). This is different than the 2.5 nm FSR we reported in [15] due to the different lengths of the MZI arms. As this interference adds negligible variation in optical transmission (<0.5 dB), a larger sampling wavelength (1 nm) is taken for the experimental measurements. An extended FSR up to approximately 20 nm is observed, as shown in Figs. 10(a)–10(f). This is attributed to the Vernier-effect induced FSR broadening caused by the multiple waveguide cavities formed of the upper arm (O1→ PS-2a→ O2) and the lower arm (O3→ PS-2b→ O2) [27,28]. This extended FSR is not observed in our 2×2 based mode (de)multiplexer/switch reported in [20], nor in [15] due to the absence of multiple cavities. For the TE0-in input mode (Figs. 10(a)–10(b)), the on-off switching occurs between the Out1 and the Out3 output ports, as explained in Fig. 7(a). The phase-shifters are biased for the lowest crosstalk at 1560 nm. As the electro-optic modulator, used in the data transmission (Fig. 7(b)), is most stable at 1563.5 nm, the IL and the crosstalk are optimized around 1560 nm, which will facilitate data transmission measurement. For the Out1 output port, the IL varies from −0.56 dB to −3.8 dB and the switching ER varies from 14.0 dB to 20.1 dB. For the Out3 output port, the range of IL is −1.5 dB to −6.8 dB, and the range of switching ER is 12.7 dB 18.2 dB, respectively. The crosstalk varies from −24.1 dB to −6.3 dB over the full spectrum. As a multi-port interferometric device, the optical transmission in the UMZI is sensitive to the phase error and power imbalance in its arms, which imposes strict requirement of phase-control by precise tuning of bias voltages. A small imbalance in power splitting ratio and phase deviation due to fabrication non-uniformity can cause non-optimal switching resulting in higher IL and crosstalk, and lower ER [29,30]. A look-up table based control algorithm can be used for the pre-compensation of the phase-error, which will potentially improve the switching performance [31]. The total heater power is estimated from the bias voltages to be 17.8 mW in Fig. 10(a) and 30.2 mW in Fig. 10(b).

 

Fig. 10 Normalized optical transmission as a function of wavelength showing reconfigurable switching between output ports for (a–b) TE0-in, (c–d) TE1-in and (e–f) TE2-in input modes. The applied bias voltages in each phase-shifter are shown next to each transmission spectra.

Download Full Size | PDF

The normalized transmissions for the TE1-in input mode are shown in Figs. 10(c) and 10(d) for the Out1/Out2 switching and the Out2/Out3 switching, respectively with the estimated heater power to be 70.0 mW and 75.9 mW, respectively. The IL and switching ER at 1560 nm are less than −3.0 dB and greater than 12.3 dB. The worst case crosstalk at 1560 nm is −12.0 dB between Out1 and Out2 outputs in Fig. 10(c). For the TE2-in input mode, the IL and switching ER are less than −5.1 dB and greater than 11.6 dB at 1560 nm, respectively. The worst case crosstalk at 1560 nm is −10.8 dB in Fig. 10(f). The average IL and crosstalk increase in the TE2-in input mode. This is due to the imperfect TE2→TE0 conversion in MMI-A (Fig. 3(e)). Indeed, a design improvement is required to reduce this loss by optimizing the length, width and I/O position of MMI-A. A design trade-off using an ADC based mode decomposer instead of the MMI-A can potentially reduce the loss at the cost of a large footprint. The average heater powers for the TE2-in input mode are 175.1 mW and 181.7 mW in Figs. 10(e) and 10(f), respectively. The overall average power consumption of the switch is 91.8 mW. The total power consumption can be significantly reduced by using a resistive phase-shifter of 21 mW/π-phase tuning efficiency [20].

3.2. Data transmission characterization

The switching performances in response to the gating signal (see Fig. 8(b)) applied at PS-2a phase-shifter are shown in Figs. 11(a) and 11(b) for the TE0-in input mode, in Figs. 11(c) and 11(d) for TE1-in input mode, and in Figs. 11(e) and 11(f) for the TE2-in input mode. The rise time (10% to 90% increase in switching response) and the fall time (90% to 10% decrease in switching response) are estimated from the static switching without payload transmission. The slowest rise times in TE0-in, TE1-in and TE2-in inputs are 12.0 μs, 11.9 μs, and 11.1 μs, respectively. The slowest fall times are 10.0 μs, 10.9 μs, and 10.3 μs for TE0-in, TE1-in and TE2-in inputs, respectively. The switching time can be reduced to 2.0 μs by replacing the Ti/W heater with a doped silicon resistive heater [32], and further reduced to 2.5 ns using a p-i-n phase-shifter [13]. The dynamic switching responses with 10 Gb/s NRZ PRBS31 optical payload are shown next to each channel. The higher crosstalk and lower switching ER in TE2-in input is attributed to the higher IL and crosstalk caused by the lower mode conversion efficiency, as shown in Figs. 10(e) and 10(f). The corresponding eye diagrams in each transmission are shown next to the switching responses. The electrical signal-to-noise ratios (SNRs) and peak-to-peak amplitudes are shown for each eye. Open eyes are observed in all cases confirming the signal integrity with distortion free high-speed data transmission.

 

Fig. 11 Measured static (left) and dynamic (middle) switching of the 3MS for (a–b) TE0-in, (c–d) TE1-in and (e–f) TE2-in input modes. The corresponding eye diagrams (right) with recorded electrical SNR and peak-to-peak voltage are shown next to each switching response.

Download Full Size | PDF

3.3. Optical power budget

MDM silicon photonics switches offer energy advantage over the single-mode switches due to the use of a single laser to transmit optical signal over multiple guided modes. As most on-chip power is consumed by turning on the laser [35], using a single laser significantly improves the energy efficiency. The on-chip power budget of the 3MS is estimated considering the measured average IL of −7.0 dB from 1520 nm to 1600 nm operating range and the average heater power consumption of 94.8 mW. The measured fiber-to-chip optical coupling loss through the surface GCs is −13.2 dB from the input to the output. The total estimated on-chip power consumption is 237.8 mW including the transmitter and the receiver power consumptions. As low-power VCSELs with multimode fiber (MMF) and/or polymar waveguide are the current state-of-the art in silicon photonics short-reach interconnects [3], a 10 Gb/s directly modulated on-chip VCSEL of 66 mW driving power is considered as the transmitter [33]. A 77 mW power consumption of a 40 Gb/s Ge-on-Si photodetector, wire-bonded to a transimpedance amplifier (TIA), is considered as receiver power [34]. Although simultaneous transmission and switching of three 10 Gb/s data signals over three modes is not demonstrated, we estimate the energy efficiency at 3×10Gb/s, considering the maximum bandwidth capacity achievable by the device [36]. For the 3×10 Gb/s aggregated bandwidth, the energy efficiency of the 3MS is estimated to be 7.9 pJ/bit, including the driver power of the laser and the receiver. This can be significantly improved by using a more efficient phase shifter of resistive doped heater, and optimizing MMI-A for a higher TE0→TE2 conversion efficiency, which will lead the way towards the targeted attojoule optical interconnects [37]. The on-chip power budget is summarized in Table 2.

Table 2. Estimated power budget for the 3MS at 3×10 Gb/s aggregated bandwidth

View Table | View all tables in this article

The 3MS can be scaled up to switch higher order TE modes (e.g. TE3 and TE4) by optimizing the width and length of the mode decomposer MMI (MMI-A) with additional output ports followed by the necessary phase-shifters. However, the higher order modes will increase both the IL and the power consumption leading to some scalability challenges. For example, the estimated IL and power consumption for the TE3 input mode in a similar 4-mode switch, calculated using numerical approximation, is −8.5 dB and 125 mW, respectively. The scalability can be further improved by introducing polarization division multiplexing (PDM) alongside MDM, where the MMIs are designed to be polarization insensitive such that the beat lengths for the TE and the TM polarizations are equal [38]. A hybrid WDM-MDM switch can be implemented by adding DWDM filters, before the ADC based mode (de)multiplexer, using either arrayed-waveguide gratings (AWGs) [6] or MRRs [7].

4. Conclusion

To summarize, we demonstrate, for the first time, a silicon photonics three-mode switch using 120° optical hybrid based thermo-optically tuned 3×3 unbalanced Mach-Zehnder interferometer (UMZI). The design methodology of the UMZI is discussed. The proposed device enables reconfigurable switching of high-speed modulated signal over the first three TE modes (TE0, TE1, and TE2). Less than 12.0 μs switching time is demonstrated while switching 10 Gb/s NRZ PRBS31 optical payload, and open eye diagrams are recorded in each switching channel. The highest IL, lowest switching ER, and worst case crosstalk at 1560 nm are measured to be −5.1 dB, 12.3 dB, and −11 dB, respectively, for the TE2-in input mode. An average 91.8 mW overall power consumption is estimated. The same structure can be used for switching first two (TE0 and TE1) and first three (TE0, TE1 and TE2) TE modes offering footprint efficiency and higher bandwidth density. By carefully designing the width and length of the mode decomposer MMI (MMI-A), and adding necessary phase-shifters, the 3MS can be scaled up to accommodate higher order modes (e.g. TE3, TE4, etc.) without significant change in the device footprint. The proposed switch can be used in reconfigurable mode-division-multiplexed silicon photonics interconnects for the deployment of high throughput energy efficient switch matrix.