Low thermal crosstalk silicon MZI optical switch with high speed and low power consumption

Figure 1 shows the structure of the directly heated compact MZI optical switch developed in this study. As shown in Fig. 1(a), two heavily doped Si electrodes matching the MMI period were connected to a wide multimode waveguide that was heated by applying direct current to the Si core waveguide. In addition, each phase shifter in the upper and lower arms was connected to non-doped Si with a high thermal conductivity. 7.15 μm long phase shifter section composed of lightly-doped Si is loaded on both arms. In this study, the number of electrodes is reduced to two, and they are composed of heavily doped Si of 400 nm width and 2 μm length, and connected to the light convergence section by MMI. By applying a voltage to the left and right aluminum (Al) electrode pads for the wire bonding, current flows in the longitudinal direction of each phase shifter and heats them independently, modulating the phase of the propagating light by the TO effect. Thus, in this switch, the waveguide is heated by direct current injection without heating the SiO₂ cladding layer, which is expected to reduce the thermal crosstalk with the surroundings compared with the microheater type. The phase shifters of each arm are connected to each other by non-doped Si, leading to convergence of the MMI. This non-doped Si lead maintains the structural symmetry of the core waveguide, and because non-doped Si has high thermal conductivity, it contributes to faster switching by reducing the temperature difference between the arms at a high speed when cooling the phase shifter. In this study, the performance is evaluated by changing the distance L between these phase shifters. The top of the waveguide was loaded with a heatsink structure consisting of TiN/Al layers embedded in a plate shape, as shown in Fig. 1(b). The heat dissipation of the heatsink suppresses the temperature change in the SiO₂ layer of the cladding, thus enabling faster switching. ³¹⁾ Simultaneously, heat dissipation by the heatsink suppresses heat propagation to the surrounding area, contributing to a reduction in thermal crosstalk during current injection.

Zoom In Zoom Out Reset image size

In this study, we first simulated the time variation in the MZI transmittance when power is injected into the Si core phase shifter of the TO switch using a heat transfer simulator. The transmittance T of the TO-MZI switch is given by

$$\begin{eqnarray}&&T=\frac{\left(1+\cos \unicode{x02206}\theta \right)}{2},\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&\unicode{x02206}\theta =\frac{2\pi {L}_{{\rm{heater}}}}{\lambda }\cdot \frac{\partial n}{\partial T}\cdot \unicode{x02206}T,\end{eqnarray} \tag{ 2 }$$

where ${L}_{{\rm{heater}}}$ is the length of the phase shifter section (7.15 μm), λ is the wavelength of the propagating light (1.55 μm), and ∆T is the temperature difference between each arm. $\frac{\partial n}{\partial T}$ is the TO constant of Si (1.86 × 10⁻⁴ 1/K). As shown in Eqs. (1) and (2), The phase difference caused by the temperature difference ΔT between the arms changes the output state of the TO-MZI optical switch. The variation of temperature of the Si waveguide over time during power injection is simulated by the heat transfer module of Comsol Multiphysics. Through this process, the switching behavior of the TO-MZI optical switch can be predicted, and we estimated the π-phase shift power consumption P_π and thermal time constant τ. The time constant is the time required for the light intensity to change to 1/e during the heating and cooling of the phase shifter.

In this study, we simulated the structural dependence of power consumption and switching time constant by varying the distance L between phase shifters to 1, 2, 4, 12, and 20 μm, with and without heatsink for each. Table I shows that the L = 20 μm without heatsink has the lowest power consumption and the L = 1 μm with heatsink, has the fastest switching operation. Here, τ_heat is the thermal time constant during heating, and τ_cool is the thermal time constant during cooling. When FOM = P_π τ mWμs, $\tau$ is the average value of τ_heat and τ_cool. The simulation results of FOM are shown in Fig. 2. The longer the distance L between phase shifters and without heatsink, the smaller the power consumption P_π is. On the other hand, the shorter the distance L between phase shifters and with heatsink, the smaller the time constant τ is and the faster the switching operation. Generally, when heating a material, the power consumption is proportional to the heat capacity and thermal conductivity, whereas the time constant of the temperature change is proportional to the heat capacity and inversely proportional to the thermal conductivity. Because a heatsink has the effect of reducing heat capacity, and a short L has the effect of increasing thermal conductivity, the results obtained from the simulation agree with the general trend. However, the thermal conduction properties of this device are not so simple. The length of L and the presence or absence of a heatsink will be discussed in the Sect. 3.2 with experimental results.

Table I. Simulation results of power consumption and switching speed for different distances between phase shifters and with and without heatsink.

Phase shifter distance L (μm)		1		2		4		12		20
Heatsink		w/	w/o	w/	w/o	w/	w/o	w/	w/o	w/	w/o
Switching time constant (μs)	${\tau }_{{\rm{heat}}}$ (μs)	0.15	0.23	0.19	0.27	0.29	0.44	0.33	0.63	0.34	0.97
	${\tau }_{{\rm{cool}}}$ (μs)	0.22	0.42	0.28	0.40	0.37	0.73	0.51	1.40	0.54	1.57
Pπ (mW)		24	21	19	17	16	14	15	12	15	11
FOM (mWμs)		4.49	6.76	4.47	5.70	5.28	8.19	6.30	12.2	6.60	14.0

Zoom In Zoom Out Reset image size

From a wavelength tunable laser diode (Santec TSL-550), TE polarized light with a wavelength of 1.55 μm was input to the TO-MZI optical switch, and a photodetector (Terahertz Technology TIA-525) detected transmitted light. The output light intensity was measured by a digital oscilloscope (TBS2000B, Tektronix). We applied the signals from an arbitrary function generator (Tektronix AFG3102C) amplified by a high-speed voltage amplifier (NF Corporation, HSA4011) to a phase shifter. For the measurement of power consumption P_π , a 10 kHz triangular wave voltage was applied to the heater from the AFG, and the current flowing in the circuit was calculated by measuring the voltage value when the output light level became zero and the voltage value of a 10 kΩ resistor connected in parallel to calculate the power applied to the phase shifter. For the measurement of switching operation, a rectangular pulse voltage of 10 μs width and voltage V_π was applied and the switching of output light from the steady state of ON state at a voltage of 0 V was measured. The time constant τ_heat , at which the output light intensity becomes 1–1/e from the falling edge of the output light during heating, and the time constant τ_cool , at which the output light intensity becomes 1/e during cooling when the applied voltage is reduced from V_π to 0 V, were measured. The rise and fall time constants of the pulse voltage applied to the measurement system were 16 and 10 ns, respectively.

To measure the thermal crosstalk, the following measurements were performed for the conventional TO-MZI switch with a microheater and the TO-MZI switch with a direct current injection heater with a heatsink, as shown in Fig. 1. First, as shown in Fig. 3, Si ring resonators were designed to be located at 30, 50, 70, and 90 μm from one arm of the MZI for the microheater type and the direct heater type. The microheater used in this measurement is placed 1.2 μm above the Si waveguide and heats the Si core through the SiO₂ cladding layer. The power equivalent to P_π is applied to each of the TO-MZI switches. The transmission spectra of the four-ring resonators were measured. Ring resonators filter peak wavelengths at regular interval FSR which is free spectral range. In this measurement, the spectrum of the ring resonator is measured before power injection and during P_π injection, respectively. When the thermal crosstalk affects the spectra, the ring resonator is heated, and the resonator length changes, resulting in a shift in the resonance peak wavelength. From this shift of the peak wavelength ∆λ, the phase change ∆θ of the ring resonator is expressed as

$$\begin{eqnarray}&&\unicode{x02206}\theta =\frac{\unicode{x02206}\lambda }{{FSR}}\,* \,2\pi \,\left({rad}\right).\end{eqnarray} \tag{ 3 }$$

To compare the phase change due to thermal crosstalk for each switch of the microheater type and the direct heater type, the phase change ratio α was defined as the ratio of the phase change of the surrounding waveguide to the phase change π of the waveguide in the phase shifter section during P_π injection from ∆θ using the following equation:

$$\begin{eqnarray}&&\alpha =\frac{\unicode{x02206}\theta }{\pi }\times 100\left( \% \right).\end{eqnarray} \tag{ 4 }$$

Zoom In Zoom Out Reset image size

The insertion loss of phase shifter using MMI in both the previous study ^27,28) and this study were measured. The results are shown in Fig. 4. The propagation modes converge periodically due to MMI in the MMI waveguide. By attaching a heavily doped silicon electrode to the position where the propagation mode converges, the scattering by the electrode is reduced. However, in an MMI phase shifter with a large number of electrodes, scattering loss increases due to the difference between the electrode period and the MMI period which have wavelength dependence. Compared to the 10-electrodes MMI phase shifter of the previous study, the 2-electrode MMI phase shifter has low loss over a wide wavelength bandwidth. This is because our improved phase shifter has fewer electrodes and the phase shifter length is as short as 7 μm. The extremely compact MMI phase shifter with an insertion loss of less than 0.1 dB over the entire C-band has been realized.

Zoom In Zoom Out Reset image size

In the experiment, phase shifters with distances L of 2, 4, and 20 μm between phase shifters were fabricated, and power consumption and switching behavior were measured. Table II presents the measurement results of power consumption P_π and thermal time constant τ for each switch structure. The power consumption P_π becomes small when the distance between phase shifters L is long and the heatsink is removed. On the other hand, the time constant τ is small and the switching operation becomes faster when the distance between phase shifters L is short and the heatsink is loaded. Figure 5 shows the measurement results for L = 2 μm, which showed the fastest switching operation in sub-μs can be realized for both heating and cooling, especially in the case of with heatsink. As shown in Table II, there is a trade-off relationship between ${P}_{\pi }$ and $\tau .$ Therefore, we will use FOM to compare the length of L and the effects of the heatsink. When the distance L between the phase shifters is short, the phase shifters are closely connected by undoped-Si with high thermal conductivity, so the heat from the heated phase shifter is quickly transferred to the other phase shifter during cooling. Therefore, ${\tau }_{{\rm{cool}}}.$ decreases as L decreases. However, since the thermal conductivity between the phase shifters is large even during heating, ${P}_{\pi }$increases as L becomes smaller, and if the phase shifters are too close together, a very large ${P}_{\pi }$ is required. As can be seen from Fig. 2, L = 2 μm is optimal to obtain the minimum FOM in simulation, and as shown in Table II, the smallest FOM was experimentally obtained with L = 2 μm. The heatsink has the effect of suppressing the heating of the over-cladding SiO₂, which has a large heat capacity, and reducing the heat capacity of the entire phase shifter. Therefore, $\tau$ is significantly reduced by heatsink loading. However, the power consumption increases slightly due to the increased thermal conductivity with the entire chip through the heatsink. The FOM is reduced by the heatsink, as the speed-up effect due to the heatsink is dominant. By improving the device structure above, we have achieved the smallest FOM of 8.89 mWμs in the MMI phase shifter with L = 2 μm and having the heatsink.

Table II. Results of power consumption and switching speed measurements for different distances between phase shifters and with and without heatsink.

Phase shifter distance L (μm)		2		4		20
Heatsink		w/	w/o	w/	w/o	w/	w/o
${P}_{\pi }$ (mW)	sim	19	17	16	14	15	11
	ex	17.4	15.1	16.5	14.7	15.8	12.4
${\tau }_{{\rm{heat}}}$ (μs)	sim	0.19	0.27	0.29	0.44	0.34	0.97
	ex	0.736	0.864	0.832	0.832	0.768	0.928
${\tau }_{{\rm{cool}}}$ (μs)	sim	0.28	0.40	0.37	0.73	0.54	1.57
	ex	0.286	1.180	0.420	1.860	0.670	2.750
FOM (mWμs)	sim	4.47	5.70	5.28	8.19	6.60	13.97
	ex	8.89	15.43	10.33	19.79	11.36	22.80

Zoom In Zoom Out Reset image size

Simulations were first performed to evaluate the thermal crosstalk performance of the direct-heater-type TO-MZI switch with a heatsink and the conventional microheater-type TO-MZI switch. We placed similar Si waveguides (test_wg) at different distances near the Si core waveguide to be heated (main_wg) and simulated the temperature change of each test_wg during P_π power injection into the main_wg. The dotted line in Fig. 6 shows the simulation results. The black dotted line in Fig. 6 shows the criteria for an acceptable phase change ratio based on the phase error per MZI required to maintain a high classification accuracy in a large-scale 64 × 64 ONN circuit. ¹⁷⁾ This value is approximately 1.6 %, and a very small phase error is required in a large-scale ONN such as a 64 × 64 ONN with multiple MZIs in a row. First, for the conventional microheater type, Fig. 6 shows that the phase change ratio cannot fall below the ONN criteria even at a distance of 90 μm. This is owing to the large thermal crosstalk caused by the large heating area. The distance between the devices must be large to maintain a high classification accuracy when microheaters are densely integrated. On the other hand, for the direct heating phase shifter with heatsink in this study, the phase change ratio is shown to be below the ONN criteria even when the distance between main_wg and test_wg is 30 μm. This can be attributed to the narrow heating region due to direct heating and the fact that the heatsink allows sufficient heat dissipation to keep the thermal crosstalk sufficiently small.

Zoom In Zoom Out Reset image size

Subsequently, we discussed the results of the thermal crosstalk measurements for the microheater- and direct-heating-type switches with a heatsink. Figure 7 shows the results of the microheater-type switch measurements for the transmission spectra of the ring resonator at distances of 30 and 90 μm from the microheater-type switch for both P = 0 and P = P_π power injection into the switch. The resonance peak wavelength of the ring resonator is shifted by thermal crosstalk when P_π power is applied at both 30 and 90 μm. Particularly, at 30 μm, which is closer to the heater, the resonance peak wavelength of the ring resonator is significantly shifted due to thermal crosstalk. As shown above, the microheater type has a large heating area; therefore, the effect of thermal crosstalk on neighboring devices is large. Therefore, to suppress thermal crosstalk, it is necessary to provide a certain long distance between the devices, which limits high-density integration. On the other hand, Fig. 8 shows the measurement results of the direct heating structure with a heatsink. Figure 8 shows that the resonance peak wavelength shift is very small even at the closest distance of 30 μm, and such a spectral shift can hardly be observed at a distance of 90 μm. The peak wavelength shift of the resonator at a distance of 30 μm is also very small compared to the microheater type, indicating that this structure contributes significantly to the reduction of thermal crosstalk. The phase change ratio α was calculated for each of these measurements. The measured and simulation results are compared in Fig. 6. As shown in Fig. 6, the experimental and simulation results were in good agreement. That is, the direct heating structure with heatsink in this study is well below the strict thermal crosstalk criteria required by ONN, even at an adjacent distance of 30 μm. Compared with the conventional microheater type, the thermal crosstalk is sufficiently low, allowing for high-density integration and close proximity to other devices. Furthermore, because the footprint of this switch is very small compared with conventional switches, it is expected to make a significant contribution in reducing the footprint of the entire circuit when configuring large-scale integrated optical circuits.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The performances of the TO-MZI switch in this study and switches in other studies are compared in Table III. In this study, the direct heating-type heaters achieved lower power consumption and faster switching speeds than conventional microheaters. Regarding the direct heating type, this and previous studies ³¹⁾ had particularly small FOM. Particularly, the FOM is significantly reduced by installing a heatsink, slightly increasing the power consumption but speeding up the switching operation by an order of magnitude. In addition, compared with previous studies, ³¹⁾ the footprint of this study was very small and reduced by one order of magnitude. Because TO-MZI switches are densely integrated as matrix switches consisting of multiple stages of MZI structures and are applied to ONN configurations, the reduction of the footprint as a single device enables downsizing of the overall footprint of the integrated circuit. Furthermore, as shown in the previous section, the thermal crosstalk of the switches in this study is much smaller than that of conventional microheaters. For example, compared with a report of thermal crosstalk measurement of a directly heating TO phase shifter using an N-type doped Si heater. ³²⁾ From these thermal crosstalk measurements, the phase change ratios calculated were 5.25% and 3.89% at a distance of 15 and 45 μm, respectively. The phase change ratio at 30 μm was estimated by the approximate curve and remains 3.67%, more than twice the ONN criteria. In addition, a distance of approximately 90 μm was required to obtain a distance below the ONN criteria. In comparison, this structure has less than 1.6% thermal crosstalk even at 30 μm, indicating that the thermal crosstalk is very small. This performance is one of the features that can significantly reduce the footprint of the entire circuit during high-density integration.

Table III. Comparison of performance between this study and TO-MZI switches in other studies.

		Switching time (μs)
Thermo-Optic switch	Pπ (mW)	Heat	Cool	FOM (mWμs)	Footprint (μm2)
Conventional heater 24)	40	8.1	8.8	338	30 × 300
MMI heater with 10 electrodes 27,28)	28	3.46	3.37	95.6	15 × 218
Adiabatic bend 33)	12.7	1.2	2.4	22.9	500 × 400
doped-Si-based TOPS with deep trench 32)	22.8	2.0	2.2	47.88	50 × 320
Asym-doped MMI w/o heatsink 31)	15	1.3	3.28	34.4	30 × 60
Asym-doped MMI w/heatsink 31)	22.6	0.360	0.500	9.72	30 × 60
Compact Si TO-MZI w/o heatsink (this study)	14.8	0.864	1.180	15.13	6 × 24
Compact Si TO-MZI w/heatsink (this study)	17.4	0.736	0.286	8.89	6 × 24

Furthermore, the thermal time constant inherent in the TO switch is extremely small because the thermal capacitance of the switch, which is determined by the device structure, is small. Simultaneously, the MZI structure allows independent control of the phase shifter of both arms. Therefore, it is possible to further speed up the switching operation using overdrive control, ²⁸⁾ assist pulse control, or differential control methods ^34,35) for voltage applications. However, the issue with this device is that the drive voltage V_π is large, approximately 10 V. This is because the current was injected in the longitudinal direction of the phase shifter, resulting in a large resistance because of the elongated structure. However, V_π can be reduced by optimizing the doping concentration in the phase shifter and studying the power injection structure in the short direction of the phase shifter width.