N × N Polymer Matrix Switches Using Thermo- Optic Total-internal-reflection Switch References and Links

We have developed fully non-blocking optical matrix switches using a thermo-optic polymer 1 × 2 total-internal-reflection (TIR) switch as a unit switching element. The TIR switch consists of crossed multimode polymer waveguides and an offset heater electrode at the switching node. The fabricated 4 × 4 and 8 × 8 optical matrix switch chips show excellent switching performances. The insertion losses are less than 2.5 and 4.5 dB for the 4 × 4 and 8 × 8 matrix switches, respectively, and their switching isolations during a turned-off state are higher than 38 dB. The switching time is about 3 ms, and the power consumption for each switching element is below 30 mW. Compact integration of the 4 × 4 and 8 × 8 switch chips is achieved at sizes of 25 mm × 4.25 mm, and 42.4 mm × 5 mm, respectively, through an optimization of the waveguide and heater geometries. Photonic inter-module connector using 8×8 optical switches for near-future electronic switching systems, " IEICE Trans. Compact optical cross-connect switch based on total internal reflection in a fluid-containing planar lightwave circuit, " in High-extinction ration and low-loss silica-based 8×8 thermooptic matrix switch, " IEEE Photon. Low loss and high extinction ratio strictly nonblocking 16×16 thermooptic matrix switch on 6-in wafer using silica-based planar lightwave circuit technology, " J. " Fabrication of 10-channel polymer thermo-optic digital optical switch, " IEEE Photon. Reconfigurable optical add-drop multiplexer using a polymer integrated photonic lightwave circuit, Ultrahigh index contrast planar polymeric strictly non-blocking 1024×1024 cross-connect switch matrix, " in Polymer waveguide thermo-optic switches with −70 dB optical crosstalk, " Opt. " 4×4 non-blocking polymeric thermo-optic switch matrix using the total internal reflection effect, " IEEE J. Polymer 1×2 thermo-optic digital optical switch based on the total-internal-reflection effect, " ETRI


Introduction
An optical matrix switch is a key device for routing signals in complex optical networks [1,2].To handle various types of data demands, the amount of data traffic in optical communication networks has been increasing drastically.As Internet data traffic has driven optical networks to spread into access networks, such as a real-scale FTTH, optical communication networks have come to have a complex structure with a large number of nodes, requiring routing flexibility for efficient network management.The flexible routing of optical signal paths can be easily provided by reconfiguring the network nodes using an optical matrix switch.However, such switches are usually expensive components when applied at the large scale of optical communication networks.
There have been several approaches used to develop an optical matrix switch [2][3][4][5][6].Among them, devices based on a 2D or 3D micro-electro-mechanical system (MEMS) technology have shown the best performances [3,4].The main type of MEMS matrix switch uses a micro-sized mirror as a switching element [3].Another type of MEMS matrix switch uses an oil bubble moving in a micro-tube machined at a cross region of an optical waveguide [4].These types of matrix switches are technically successful.However, their structures are too sophisticated for cost-effective mass production.
On the other hand, an integrated planar lightwave circuit (iPLC) technology based on a silica or polymer material has been considered as a promising candidate for an optical matrix switch owing to such advantages as easy packaging, mass reproducibility, no moving parts, long-term reliability, and high repeatability [5][6][7][8][9][10][11][12][13][14].A silica optical matrix switch has a good low-loss characteristic; however, it mainly operates using phase control, and is thus sensitive to the wavelength and polarization [5][6][7][8][9].Furthermore, these types of silica optical matrix switches have too complex a structure to overcome the poor switching properties of silica owing to their high power consumption resulting from the low thermo-optic coefficient of the silica material [5,6].In addition, they are fairly large in size.Therefore, polymer-based optical switches have been preferable for overcoming these problems [7][8][9][10][11][12][13][14].
In the initial development stage for multi-channel polymer matrix switches, the creation of cascaded 1 × 2 symmetric Y-branch digital optical switches (DOSs) was attempted [10][11][12]; however, they still have a highly complex structure [10,11] and a relatively bulky size.As an alternative, it was reported that a polymer thermo-optic 4 × 4 matrix switch was fabricated using the total-internal-reflection (TIR) effect [13].The structure of this matrix switch was rather simple compared to the silica matrix switches or Y-branch type polymer matrix switches.However, it did not show sufficient performances for stable use in optical network systems owing to a high insertion loss range of 4.5 -8.7 dB and a low switching isolation of ~23.3 dB.In addition, the 4 × 4 switch had a relative large chip size of 39.3 mm considering the number of channels.
In this paper, we report fully non-blocking optical matrix switches using polymer iPLC technology.Our optical matrix switch employs a polymer thermo-optic 1 × 2 TIR switch as a unit switching element, which has an asymmetric Y-branch of multimode polymer waveguides and a heater electrode to induce a TIR through a thermo-optic effect on the branched node [14].The heater electrode on the crossed multimode waveguides can act as a TIR mirror since, when heated up, the refractive index of the polymer near the heater is decreased, causing an internal-reflection phenomenon.The fundamental mode in a multimode waveguide propagating with equiphase front vertical to the propagation axis can be approximated as the light propagating in a 2D free space, and it is therefore much easier to undergo the TIR phenomenon than in a well-confined guided mode of a single-mode waveguide.If a cross angle between the multimode waveguides is small, only a slight change of the refractive index is enough to cause the TIR phenomenon, which enables the light to switch with low loss.
Using this TIR switching scheme, we realized very compact 4 × 4 and 8 × 8 optical matrix switch chips.Our optical matrix switches consist simply and distinctively of a plurality of multimode waveguides and offset heater electrodes in the cross regions, single-mode waveguides in only the input/output region, and tapered waveguides between the single-mode waveguides and multimode waveguides.In particular, a two-layer wiring technique forming electrodes at both the top and bottom layers of the waveguides is applied to the 8 × 8 optical matrix switch for a simple wiring configuration.We demonstrate experimentally that the matrix switches show excellent characteristics such as a low loss, high switching isolation, and low power consumption.

Structure of polymer optical matrix switch
A conceptual schematic configuration of our N × N optical matrix switch is illustrated in Fig. 1(a).The individual switching element is made up of single-mode input/output ports, straight multimode waveguides are crossed with a shallow angle corresponding to the branched angle of the switching element.The heater electrodes used to induce a TIR are formed on each crossed node or branched node on the edge side of the switch.This electrode is tilted toward the waveguide axis with half of the cross angle of the multimode waveguides, as shown in Fig. 1(b).Note here that the location of the heater is offset toward the upper side of the cross point to improve the internal reflection efficiency, considering a shift of the reflective interface from the thermal diffusion.To couple the optical matrix chip with external single-mode fibers, the input/output ports of the single-mode waveguides are configured on the left and right sides of the matrix switch, as shown in Fig. 1(a).The principle of the switching operation is as follows.The signal light is coupled into the input ports of the singlemode waveguide.The single-mode light is adiabatically expanded in the tapered region to be coupled into a 0th-order mode (fundamental mode) of a multimode waveguide, and then propagates through the multimode waveguide.At a crossed (or branched) node, if the electrode is turned off, the light is transmitted in a straight manner.If the electrode is turned on, the light is reflected into the crossed waveguide by a decrease of the refractive index.Through a simultaneous setting of the on/off states of the whole electrodes, each incident light suffers a cascaded path change, and an N × N switching function can therefore be achieved.Through this mechanism, our matrix switch can be structurally operated in a nonblocking manner.
To change the direction of the light with low power consumption, i.e., with a small temperature change, the cross angle between the two multimode waveguides must be small enough to induce a TIR.However, if the cross angle becomes small, the coupling between the waveguides may increase.This coupling induces a crosstalk and increases the propagation loss along the signal waveguide, which affects the switching isolation, particularly in a turned-off state.Thus, an optimized design is required for the waveguide and electrode geometry.
To first optimize the cross angle and width of the waveguide, we calculated the propagation loss of the light using a 3D beam propagation method (BPM) simulator, BeamPROP.A cross-sectional view of the polymer waveguide used in this study is shown in Fig. 2. We used commercially available LFR-series polymer materials [15], synthesized by ChemOptics, Inc.The polymer materials are photoactive UV-curable resins based on perfluorinated acrylate.The thermo-optic coefficient of the polymer materials is about −2.5 × 10 −4 /°C, which is about 30-times higher than the typical value of silica materials.The refractive index of the core is 1.378 at a wavelength of 1550 nm.The index contrast between the core and clad is 0.35∆-%.The single-mode waveguide has a core size of 7 µm × 7 µm, while the multimode waveguide width varies from 20 µm to 60 µm.The tapered waveguide length is long enough for the incident light to adiabatically expand from the tightly guided single-mode to the fundamental mode of the multimode waveguide.Figure 3 shows the calculated transmittance of light passing straight through the crossed multimode waveguide for variations in the heater angle (or half cross angle of waveguides) and the multimode waveguide width.The results indicate that the transmission loss decreases as the heater angle and the multimode waveguide width increase, in which the decrease of the transmission loss also means the decrease of the crosstalk and leads to the increase of the switching isolation in a turned-off state.However, if the heater angle is too large, the temperature of the heater should be raised to achieve a TIR effect.In addition, if the multimode waveguide width is too wide, some problems may occur during the waveguide fabrication process when filling and planarizing the gaps near the cross region for the following step of creating flat electrodes in the structure shown in Fig. 1(b).With this in mind, we chose a 4° heater angle and a 45-µm width for the multimode waveguide.The transmission loss at each crossed node is calculated to be about 0.06 dB (i.e., ~0.985 in transmittance).For the N × N matrix switch shown in Fig. 1, the possible longest path has 2 × (N-1) transmitting nodes and one reflecting node.Thus, the maximum loss caused by the transmitting nodes is estimated to be only about 0.36 and 0.84 dB for 4 × 4 and 8 × 8 matrix switches, respectively.Since every possible channel has only one reflecting node, there is an additional loss caused by the TIR at the reflecting point.The light reflection was also calculated using the 3D-BPM simulator, as shown in Fig. 4. As shown in Fig. 4(a), a switching node has one input waveguide and two output waveguides.The heater is formed with a half of the cross angle between the two waveguides and with an offset distance from the center of the crossed node.To simplify the BPM simulation, we fixed the cross-sectional structure of the multimode waveguide.The lower-clad, core, and upper-clad were set to a thickness of 15.5, 7, and 5 µm, respectively.Thus, only four important variables remain for the BPM simulation: the heater angle, heater offset, heater width, and heater temperature.Figure 4(b) shows the thermal distribution near the heater electrode with a width of 8 µm.We used a silicon wafer as a substrate, which has a thermal conductivity of 163.3 W/(mK), which is about 800-times higher than that of acrylate polymers.Thus, the silicon substrate acts as a heat sink, and the heater acts as a heat source.From these conditions, the thermal distribution can be calculated based on a steady-state heat flow equation.Figure 4(c) shows the propagation of the reflected light at the crossed multimode waveguides.Optical power variations of higher-order modes are also monitored during the reflection.The blue line shows the light power toward the transmission (T) waveguide, while the green line shows the light power toward the reflection (R) waveguide.All other lines show the higher-order modes excited by a refractive index perturbation in the crossed node.As shown in Fig. 4(c), almost all incident light power is converted into fundamental mode in the reflection waveguide.The powers of the higher-order modes were much lower than that of the fundamental mode.For example, the power of the 1st-order mode was less than one-hundredth that of the fundamental mode.In the matrix switch, all higher-order modes will be filtered away at the output tapered waveguides and single-mode waveguides, subsequently.The loss occurred by the reflection at the crossed node is calculated to be less than 0.135 dB (i.e., more than ~0.97 in reflectance), while the loss from the mode conversion at the output tapered waveguide is below 0.05 dB.
Figure 5 shows the switching properties for variations in the angle, offset, and temperature rise of the heater.At an optimal heater offset of around 9 µm, the incident light is efficiently reflected into the reflection waveguide for heater angles of both 4° and 4.5°.Under this optimal heater offset, the switching is completed at a temperature rise of 45°C and 55°C for a heater angle of 4° and 4.5°, respectively.Therefore, to reduce the power consumption of the unit switching element, we need a heater angle of 4°, where the switching isolation in the turned-off state is even higher than 50 dB.Figure 5 also shows that, at the optimal offset, the switching curves are not sensitive to a temperature rise of the heater after the switching is completed.This condition enables a digital-like control of our matrix switch.

Fabrication of 4 × 4 optical matrix switch module
From the simulation results shown in Figs.3-5, the optimized parameters for a basic structure needed to fabricate 4 × 4 and 8 × 8 matrix switch chips were determined as follows.The designed structure of the 4 × 4 matrix switch has a heater angle of 4°, a multimode waveguide width of 45 µm, a heater width of 8 µm, a heater offset of 9 µm, and a heater length of 1150 µm.Gold electro-plating was applied to the terminal pads for easy wire-bonding on the surface of the polymer and for reducing the resistance of the electrode wiring.Figure 6 shows a photograph of the 4 × 4 matrix switch chip and an illustration of the photomask used.The chip has a compact size of 25 mm × 4.25 mm, and a total of 32 chips can be placed on a 4inch silicon wafer.Figure 7 shows the typical optical characteristics of a fabricated 4 × 4 matrix chip.As shown in Figs.7(a) and 7(b), the power consumption needed for full switching was about 25 mW.For an input channel, to send a light signal to one of the output channels, only one of the heaters is turned on, and the total power consumption required for a 4-channel switch will therefore be about 100 mW. Figure 7(a) shows the polarization-dependent loss (PDL) curves.For the two linear polarization states, the switching curves were measured using a polarization scrambler.The PDL in a turned-on state was less than 0.13 dB.The switching isolation in a turned-off state was higher than 38 dB.However, as can be seen in Fig. 7(b), a fluctuation is observed within a power consumption range of between 0 and 5 mW.This fluctuation may be attributed to stress induced by the heaters during the fabrication process and may deteriorate the switching isolation even in a turned-off state.Thus, the switching isolation can be further improved if the stress is released.This suggestion will be discussed again later for the results of the 8 × 8 matrix switch.The insertion losses in the turned-on states were less than 2.2 dB for all 16 switching nodes, and the loss uniformity was within 0.7 dB.The switching time (falling to 10% and rising to 90%) was measured to be faster than 2 ms when turning the switch off and 3 ms when turning the switch on, as shown in Fig. 7(c).To create a matrix switch module, the 4 × 4 matrix switch chips were pig-tailed with ribbon fibers and packaged with a thermo-electric cooler into a metal package.After packaging, all possible cases of channel crosstalk were measured to be far lower than −40 dB.We also developed a circuit board to control the 4 × 4 matrix switch module using RS232 or I 2 C commands with a personal computer or a motherboard in the optical network system.Figures 8 and 9 show the packaged 4 × 4 matrix switch modules and the controller circuit-board developed in this study, respectively.The properties of the 4 × 4 matrix switch modules are summarized in Table 1.

Fabrication of 8 × 8 optical matrix switch chip
We also fabricated an 8 × 8 matrix switch chip.In this scaled-up matrix switch, the electrode wiring configuration becomes a new issue since 64 switching elements are integrated into the chip.We developed a two-layer wiring technique, as illustrated in Fig. 10, for the wiring of 8 × 8 or higher-port matrix switches.A thermal oxidized silicon substrate was used for the electrical insulation of the bottom-layer wiring.For the formation of the electrodes, Cr/Au metal for the bottom-layer wiring was first patterned on the substrate.Next, the polymer waveguides were processed using a standard spin coating and dry etching, and the heater electrodes were then patterned on the surface of the polymer waveguides.After that, to connect the heater electrodes with the bottom-layer wiring, we additionally etched the polymer waveguides with a shadow photomask to form a slanted trench.The shadow photomask was made using a gray scale tone which consists of multiple lines with gradually varying widths and gaps.Finally, Cr/Au metal was deposited on and around the slant-etched trench area.All contact points between the top and bottom layers were tested, and it was proven that the heater electrodes are well connected electrically with the bottom-layer wiring through the metal on the surface of the slant-etched trench.In the 8 × 8 matrix switch, the heater length was 1500 µm, which is longer than in the 4 × 4 matrix switch.Employing our waveguide and electrode configuration, the 8 × 8 matrix switch chip could be fabricated into a very compact size of 42.  Figure 11 shows a drawing and photographs of the 8 × 8 matrix switch chip and its typical switching properties.Unlike in the switching curves of the 4 × 4 matrix switch, no fluctuation in the output powers is observed within a power-consumption range of 0 -10 mW, as shown in Fig. 11(c).Through the optimization of a fabrication process relevant to the formation of the heater electrodes, the stress induced by the heaters could be almost completely released, and thus the switching properties were no longer deteriorated.As a result, we can achieve a switching isolation of over 50 dB in a turned-off state, and a channel crosstalk of much lower than −50 dB for all switching paths.The 8 × 8 matrix switch chip showed a power consumption of between 40 and 47 mW for each switching element, which is higher than that required for a 4 × 4 matrix switch.This high power consumption can be attributed to the increase in resistance proportional to the length of the heaters and electrode wiring, which is longer than that of the 4 × 4 matrix switch.It is expected that the power consumption of each switching element can be further reduced to below 30 mW through an optimization of the waveguide structure (i.e., thickness of the lower-clad and upper-clad) and the electrode wiring (i.e., resistance).The power consumption can also be further decreased through an improvement in the gap-filling in the crossed multimode waveguides near the heater electrodes, particularly for the highly dense structure of a 8 × 8 matrix switch chip, as illustrated in Fig. 11(a).
Figure 12 shows the insertion losses of each switching element in the fabricated 8 × 8 matrix switch chip, the values of which ranged between 3.5 and 4.75 dB, with a loss uniformity of less than 1.3 dB.The loss uniformity can be improved by adding dummy waveguides around the switch patterns.The periodicity of the insertion losses resulted from the periodicity of the number of crossed nodes in the switching paths.The linear fitted line for these insertion losses, as shown in Fig. 12, is inclined with a gradient of −0.0092 dB per switch, which means that the average loss of a crossed node is about 0.074 dB (8 × 0.0092 dB).The average loss was slightly larger than the value of 0.06 dB obtained from the BPM simulation.The referenced straight waveguide showed an insertion loss of 2.2 dB, and thus the loss occurred by the reflection is estimated to be about 1.3 dB by subtracting the straight waveguide loss from the insertion loss (~3.5 dB) of the shortest switching path.
From the results of the 4 × 4 and 8 × 8 matrix switches, those using a TIR show excellent properties in terms of insertion losses, switching isolation, and power consumption.Furthermore, since they have a structural simplicity and compactness, our polymer thermooptic matrix switches are believed to be well applicable for high-performance and low-cost optical switching networks.

Conclusion
Fully non-blocking 4 × 4 and 8 × 8 optical matrix switches were realized using a polymer thermo-optic 1 × 2 TIR switch as the unit switching element.Our optical matrix switches are distinctively composed of a plurality of crossed multimode waveguides and offset heater electrodes in the switching node, single-mode waveguides in only the input/output region, and tapered waveguides between the single-and multi-mode waveguides.Through an optimization of the waveguide dimensions, cross angle, heater angle, and heater offset, we could obtain excellent optical properties in terms of the insertion loss, switching isolation, PDL, and power consumption.Owing to the simplicity of the switching structure using the TIR, we could fabricate very compact 4 × 4 and 8 × 8 matrix switch chips.A compact electrode configuration was also developed for a large-scale matrix switch, employing a twolayer wiring structure.Our polymer thermo-optic matrix switches can be useful for costeffective optical switching systems.

Fig. 1 .
Fig. 1.(a) Conceptual schematic configuration of the N × N optical matrix switch and (b) detailed structure of a switching node at crossed waveguides.atapered waveguide region for mode conversion, and a switching node with branched multimode waveguides, as described in a previous work[14].To form the matrix switch,

Fig. 3 .
Fig. 3. Transmittance of light passing through the crossed multimode waveguide, where half cross angle of waveguides indicates the heater angle.

Fig. 4 .
Fig. 4. 3D-BPM simulation results for the internal reflection at the crossed node of the multimode waveguides: (a) crossed node, (b) thermal distribution near the heater electrode,and (c) beam propagation at the crossed node, where the heater angle is 4°, the heater offset is 9 µm, and the heater temperature rise is 45°C.

Fig. 5 .
Fig. 5. Reflectance of light reflected at the crossed multimode waveguide for variations in the heater angle, heater offset, and temperature rise of the heater electrode: heater angles of (a) 4° and (b) 4.5°.

Fig. 10 .
Fig. 10.Two-layer wiring scheme through the formation of a slant-etched trench of the polymer waveguides in an 8 × 8 optical matrix switch.

Fig. 11 .
Fig. 11.(a) Drawing of a photomask for an 8 × 8 optical matrix switch chip, (b) photographs of the electrodes in the fabricated chip, and (c) optical power switching curves for all cases of 8 × 8 switching.

Fig. 12 .
Fig. 12. Insertion losses of each switching element in an 8 × 8 optical matrix switch chip.