1. Introduction

The use of silicon photonics to realize practical integrated optical circuits has become increasingly popular over the last decade buoyed by the performances of the different required optical functions and CMOS compatibility allowing integration directly with electronics as well as high volume, high yield, low cost fabrication. Multiple approaches have been proposed and demonstrated for the modulation of the optical signal. For high speed modulation, approaches involving the plasma dispersion effect or the hybridization of other electro-optic materials are generally used. For lower speed applications (<100 KHz), the use of the thermo-optic effect can be very effective due to the simplicity of the structure and low optical losses. There are a wide range of low speed applications of optical modulators including phase tuning in high speed modulators [1], ring resonator resonance tuning [2,3], optical switching arrays, beam steering elements for LIDAR applications [4] and programmable optical circuits [5]. When designing a photonic integrated circuit (PIC) the thermal crosstalk between adjacent components is an important consideration to ensure the overall circuit operates correctly. Designers usually seek to make the overall footprint of the circuit as small as possible in order to make efficient use the silicon real estate, allowing a lower fabrication cost per chip. Understanding the lateral heat spread from a thermo-optic element of device is therefore a very important consideration, limiting the compactness of closely packed components before thermal crosstalk becomes an issue and dictates how small the overall chip can be. Recently, different heating mechanisms have been studied, and measures to mitigate thermal crosstalk have been proposed [6,7]. Isolation approaches have been demonstrated in the literature to confine the heat in order to increase the device efficiency and/or to reduce thermal crosstalk [8,9], however these add additional fabrication steps, complexity and do not typically feature in standard silicon photonic fabrication services. Therefore, we have focused on a device without isolation.

In sections 2 and 3, we have used simulation and experimental approaches to assess the spread of heat from a typical silicon photonic thermo-optic heating element. The results show that in a 220 nm SOI based fabrication platform without any isolation approaches, if the center of a silicon waveguide is offset from the center of the heating element by more than 6um then thermal crosstalk is minimal. The simulation analysis is extended to other available platforms in section 4, giving generalized guidelines for minimal thermal crosstalk.

2. Design and fabrication

2.1 Design of the thermo-optic test structures

For this study a typical silicon photonic thermo-optic element involving a metallic heating element running along the top of the silicon waveguides has been used (as shown in the cross section of Fig. 1). The thermo-optic phase modulators were based upon the standard components from process design kit (PDK) of the Cornerstone multi-project wafer (MPW) fabrication service [10,11]. As shown in Fig. 1, the fabricated devices are based in a 220nm thick silicon strip waveguide platform with a 2um thick buried oxide layer and 1um of top cladding oxide. Heating elements consisting of 150nm thick, 108um long and 2um wide TiN strips positioned above the waveguide were connected to Ti/Au pads at either end for electrical probing. A separation of 5um was used between the waveguide edge and the surrounding silicon.

 

Fig. 1. Cross-section of the thermo-optic phase modulator.

Download Full Size | PDF

Two different test structures were used to experimentally study the spread of heat away from the heating elements as shown in Fig. 2. The first (Fig. 2(a)) consisted of a Mach-Zehnder interferometer (MZI) with multiple heating elements, placed along one of its arms, each with a different offset ranging from 0um, 2um, 4um, 6um and 8um from the center of the waveguide. Large separation between the 2 arms of MZI was used to avoid any thermal crosstalk between them. In this test structure, we can assess the phase shift of the light against the power required for switching for each offset case on the same optical structure and therefore assess how much the heat spreads from the heating element in each case. The second test structure (Fig. 2(b)) consisted of a set of 3 MZI with different arm separations (2um, 4um and 6um). In this test structure, thermal crosstalk from one arm to the other will result in a reduction in the relative phase shift between the two arms. The thermal crosstalk between the arms of the MZI can be assessed through the additional driving power required to achieve a certain phase shift.

 

Fig. 2. Test structures used to characterize the lateral spread of heat. (a) Consists of multiple heating elements along one arm of MZI, each with a different positional offset from normal position, which is with the center of the waveguide and the center of the heating element aligned. (b) Consists of a set of MZI each with a different waveguide arm separation.

Download Full Size | PDF

2.2 Fabrication of the test phase modulators

Fabrication of the test phase modulators was performed using the Cornerstone MPW service [10] in 220nm overlayer silicon on insulator. First, 248nm deep ultraviolet (DUV) lithography is used to expose the grating couplers in photoresist before inductively coupled plasma (ICP) etching of 70nm into the silicon overlayer. DUV lithography and 220nm ICP silicon etching was then used to define strip waveguides. One micron of silicon dioxide was then deposited as a top cladding of the waveguide using plasma enhanced chemical vapor deposition (PECVD). Two DUV lithography, metal deposition and lift-off steps were then used to define first the titanium nitride heater filaments and then gold pads for electrical connection to the filament. An optical microscope image of one of the fabricated thermo-optic MZI is shown in Fig. 3.

 

Fig. 3. Microscope image of fabricated device.

Download Full Size | PDF

3. Results and discussion

The experimental test structures were measured by coupling light at 1550nm into and out of the input and output waveguides of the MZI respectively using grating couplers etched into the top waveguide surface. An electrical probe head with two tips was used to contact the pads at either end of the heating filament. A computer-based scan program then scanned the voltage applied across the heating element whilst measuring the electrical current and optical power at the output of the MZI. From this the electrical power was calculated at each voltage point. The MZI output power versus electrical power data was then fitted to sinusoidal functions to extract the switching power (or power required to produce a π radian phase shift).

The normalised optical transmission against electrical drive power on the heater for the first test structure is shown in Fig. 4(a), Dataset 1, Ref. [12]. The different colour plots represent the optical power measured for the heater elements with different offset positions. Figure 4(b), Dataset 1, Ref. [12] shows the extracted switching power for each heating element offset. A switching power of 13mW is required for the zero-offset case. It can be seen that the switching power required to achieve a 2π phase shift rises dramatically at 6um and beyond. It should be noted again that the switching power is extracted by fitting the data of Fig. 4 in linear form to a sinusoidal function and that a 2π phase shift was not reached over the range of electrical power applied to the heating element for offsets of 6um or larger.

 

Fig. 4. (a) MZI normalised transmission against electrical drive power for the different heating element offset positions. (b) Electrical power required for switching (2π phase shift) with the heater off-set from the waveguide by different amounts.

Download Full Size | PDF

Since the thermo-optic effect in silicon is well known [13] it is possible to calculate the rise in temperature in the waveguide with the switching power required for a 2π phase shift in the zero offset case applied to the heating element. This data is plotted in the graph shown in Fig. 5, Dataset 2, Ref. [14]. Using Lumerical HEAT, steady state simulations for the spread of heat through the optical structure has also been performed (results plotted as red dots in Fig. 5). Structures identical to test structure 1 were used for the simulation with the temperature of bottom of the simulation region fixed at 300 K, and the region above the heating element bound by air. A uniform heat source object was used as heater wire and temperature monitors were placed around the waveguide to capture the temperature profile of the waveguide. Although, the propagating mode would be affected by temperature profile overlapping the mode area, temperature data at the centre of the waveguide was used for comparison with the experimental results (Fig. 5). A good agreement between simulation and experimental results can be observed. As can be seen the temperature change in the waveguide drops from approximately 38°C with zero offset to around 2°C with an 8um offset. According to the simulations the temperature change in the waveguide drops to around 0.6°C with a 10um offset.

 

Fig. 5. Temperature change in the waveguide for different heater element position offsets with the electrical power required for switching in the zero offset case applied.

Download Full Size | PDF

Next the second test structures consisting of the 3 separate MZI with 2um, 4um and 6um separation between the two waveguide arms. The normalised optical transmission against electrical drive power on the heater for each MZI is shown in Fig. 6(a), Dataset 3, Ref. [15].

 

Fig. 6. (a) Normalised transmission against electrical drive power for the three MZI with waveguide arm separations of 2um, 4um and 6um. (b) Electrical power required for switching (2π phase shift) for the 3 different MZI with different arm separation.

Download Full Size | PDF

As with the first test structure the switching power required for each of the MZI was then extracted using a fitting method and the data plotted in Fig. 6(b), Dataset 3, Ref. [15]. As the separation is reduced, the electrical drive power required for switching is increased since the heat spreads from the heater on one arm to the other meaning that the phase shift in the intentionally heated arm is partially cancelled out and therefore needs to be driven harder. At a separation of 6um the electrical power required for switching is 28.6mW which is just slightly higher than the 26mW required in the first test structure where a very large separation >220um was used. This indicates that thermal crosstalk is minimal when a separation of 6um is used.

4. Extension to other SOI platforms

The simulation techniques utilized above can be extended for analysing other SOI platforms for example those offered by CORNERSTONE and in general use as well. Three different silicon overlayer thickness as the starting wafer have been considered as shown in Table 1.

Table 1. Different SOI platforms considered for simulation.

View Table | View all tables in this article

Figure 7(a) shows the basic configuration of the structure used in the simulations to calculate the change in temperature at the centre of reference waveguide and heated waveguide. The heater filament (2um wide and 150nm thick) and waveguide (width = 0.5um) underneath it moves (to the left) in x the direction as the separation ΔS increases, and the reference waveguide is kept stationary at x = 0um.

 

Fig. 7. (a) Cross-section of the simulated structures showing 2 STRIP waveguides with a separation of ΔS. (b) Cross-section for two different waveguide separations, ΔS₂>ΔS₁.

Download Full Size | PDF

The resultant temperature changes along the cross section passing through the centre of the waveguide are shown in Fig. 8(a), Dataset 4, Ref. [16], for different ΔS, for the 220nm SOI STRIP platform.

 

Fig. 8. (a) Temperature line profile across the centre of waveguide cross section (as shown in Fig. 7(a)) for different separations. Red line (ΔS=0) represents the heated and reference waveguide without separation (i.e. they are the same waveguide), and blue, orange, green line represents heated waveguides at position −3, −6, −9um distance with reference waveguide placed at 0um. (b) Percentage temperature change in the reference waveguide for different waveguide separations in various platforms.

Download Full Size | PDF

It is clearly visible that for different ΔS, the reference waveguide is also heated to an extent which causes the lowering of the effective phase shift between the two arms. It can also be seen that the temperature of reference waveguide reduces as the heated arm and filament moves further away from it.

As a guide in the simulation analysis, the temperature change in the reference arm as a percentage of the heated arm is plotted, as the separation between the waveguides ΔS is varied (Fig. 8(b)). From the experimental analysis, we can conclude that there would be minimal thermal transmission effects from one arm of the MZ to other when the distance is between the waveguides is more than 6um which calculates to a percentage change of 90% compared to the peak temp of the waveguide undergoing phase transition. We extend this number to use for the other platforms and the data is plotted in Fig. 8(b).

From Fig. 8(b), the following guidelines (Table 2) for minimal crosstalk (−90% level) can be inferred.

Table 2. Guidelines for waveguide separation with minimal thermal crosstalk for various platforms.

View Table | View all tables in this article

The data shows that the RIB waveguide-based platforms require a larger separation to achieve minimal crosstalk due to the silicon slab layer which is comparatively more conductive of heat than SiO₂. The high lateral spread of heat with 500nm SOI platform utilizing greater silicon slab height is also clearly noticeable in the results.

5. Conclusion

In conclusion, the spread of heat from typical metallic micro heaters used in silicon photonic components has been studied through experimental and modelling techniques. Two different experimental test structures have been used, the first consisting of multiple heating elements positioned along one arm of a single MZI each with a different positional offset from the center of the waveguide and the second consisting of a set of MZI each with a different waveguide arm separation. Both test structures are in good agreement that for 220nm SOI Strip based platform, at distances of approximately 6um and beyond, thermal crosstalk and therefore the spread of heat from one structure to the next is minimal. Other common platforms have also been studied through simulation. These results can serve as a guide to how compact and closely packed such devices can be placed without the use of more complex isolation techniques.