1. Introduction

With increasing demand for higher transmission rates and larger data capacities owing to the fast growth of big data, cloud computing, and Internet of Things, wavelength division (de)multiplexing (WDM) configuration has assumed a great deal of importance and is being widely utilized in commercial telecommunication networks [1]. One of the most commonly used mechanisms for (de)multiplexing optical signals is arrayed waveguide grating (AWG).

AWGs have attracted considerable public attention in the telecommunication realm since the advent of the devices proposed in the early 90s [2–4]. By leveraging both imaging and dispersive properties based on free-space propagation and phase control of waveguides, respectively, the optical beam can be engineered and focused at the specific waveguide ports in terms of corresponding wavelengths for large data capacity over (de)multiplexing [5]. In the past few decades, many AWGs have been presented and implemented on different material platforms such as silica [6,7], indium phosphide (InP) [8,9], silicon-based polymer [10,11], lithium niobite [12], silicon nitride (SiN_x) [13], and silicon-on-insulator (SOI) [14–16]. Among these platforms, SOI has become a promising and powerful candidate given the high maturity and mass production leveraged by the complementary-metal-oxide-semiconductor (CMOS) technology. This has given rise to silicon photonics (SiPh) in the realm of data communication [17] for high-quality and cost-effective photonic integrated circuits [18,19]. Devices with smaller footprints can be designed in terms of a higher index contrast and a smaller bending radius [20–24] featured by the SOI platform compared with other platforms, greatly decreasing the required optical power for devices and thus systems. Given these advantages, footprints of AWGs implemented on the SOI can be significantly reduced.

For conventional AWGs, a Gaussian shape is obtained in the filtering response, bringing about tight restrictions on the wavelength tolerance of optical sources and a requirement for accurate temperature control for both the AWG and the sources. To relax these restrictions, different approaches [5,14,25–36] have been proposed to flatten the passband shapes of AWGs. Among these, multimode waveguides are utilized at the receiver side of an AWG to obtain a flat response. However, this is not preferred for single-mode systems [36]. Another method involves converting the optical field at the transmitter or receiver into a double image, i.e., double-peaked electric field distribution, which can be realized using a Y-junction [5,25], a short MMI coupler [26–29], or a parabolic horn [25,30,31]. Furthermore, a different approach leveraging Fourier transform has been proposed to obtain a double image at the receiver [14,32–35]. Given the pairing relation of Fourier transform between sinc and rectangular functions, a sinc-like envelope electric field distribution is designed by introducing additional phase retardation of plus or minus π to the portion of arrayed waveguides (AWs), leading to a double image at the interface between the second slab and output waveguides, and consequently a flat spectral response at the output ports. Although a flattened passband shape can be achieved, the insertion loss (IL) is normally increased by 3–5 dB, which is a critical issue for transceiver modules. Therefore, an alternative method based on the relation between heat and the material refractive index of silicon has been proposed to address this issue.

Given the tight restrictions on the wavelength tolerance brought by conventional Gaussian-shaped filtering responses of AWGs and refractive index deviations of the fabricated silicon layer, thermally tunable AWGs (TTAWG) have been implemented on different platforms [37–41]. In the literature, either red- or blue-shifted spectra can be respectively realized using core materials with a positive [38,39,42] or negative [37,40] thermo-optic coefficient without compromising the shapes of spectra. Although bi-directionally tunable functions were achieved in [41], the TTAWGs usually require several tens of volts for electrical voltages [38] or large thermal power of 1–3 W [41] to shift the spectra by 2–5 nm due to thermo-optic coefficients of core regions and waveguide dimensions. This paper proposes a thermally bi-directionally tunable AWG (TBDTAWG) with a wavelength tuning range of 8 nm using a sub-watt thermal power driven by an ultra-low electrical voltage of below 3 V on the SiPh platform. In addition, an automatically stabilizing function for the required filtering responses of WDM systems can be achieved with a scheme using a proper feedback control circuit.

In this paper, a TBDTAWG at a low voltage for tuning wavelengths of ∼8 nm is proposed and demonstrated on an SOI platform. Design principle and simulation results are presented in Section 2, consisting of passive and active patterns for the functions of (de)multiplexing and heating. For the passive pattern, shallow-etched patterns connected with bending waveguides with designed radii are employed at the interface of free propagation regions (FPRs) in an S-shaped AWG for lower IL and crosstalk (XT). For the active pattern, a parallel configuration is utilized for heating. Electrical layers above two triangular thermal-tuning regions with complementary phase distributions in the S-shaped architecture are employed to achieve feasible functions of thermally bi-directional tunings, i.e., both red- and blue-shift tunable functions, under an ultra-low thermal power and electrical voltage. In Section 3, measured and simulated spectra in terms of different electrical voltages on both triangular regions are compared and presented in figures followed by detailed descriptions of differences as well as their attributions. In Section 4, discussions on possible approaches to improve performances of the device and on the automatically stabilizing process are made before the conclusion of this paper.

2. Design principles and simulation results

Figure 1(a) shows a schematic top view of the proposed TBDTAWG for 1×4 coarse WDM (CWDM) system with a channel spacing of 20 nm operating at O-band and transverse electric (TE) polarization state. The device is based on the SOI platform with a 220-nm-thick silicon layer on a 2-µm-thick buried silicon dioxide interlayer. For single-mode and O-band operations, a waveguide width W of 380 nm is used for optical input (skin)/output (light blue) ports with a pitch Λ_i of 3 µm of the device composed of passive and active designs. The passive design is realized with an AWG, while the active design is utilized with heating (black) and electrical (dark gray) patterns, as shown in Fig. 1(a)–(e). For the passive design, the AWG consists of two star couplers and an AWs design for far-field imaging and phase control, respectively. Both star couplers given in Fig. 1(b) and 1(c) are implemented with FPRs as well as input and output ports of the AWG. Image planes formed by a Rowland circle are designed in both star couplers for reduced aberrations. Radii of shapes of the image planes at the slab interfaces with AWs and with input/output ports are denoted by R_o and R_i, respectively. They are designed to have dimensions of 48.64 µm and 24.32 µm in terms of the effective index at slab region, pitch of tapered waveguides at interfaces of slab regions, the channel number, and the central wavelength. Tapered waveguides with a pitch D of 1.9 µm are employed at the interfaces as bridges for straight waveguides (light gray) with a width W of 380 nm so that sufficient spacings can be provided for the evanescent wave and shallow-etched patterns at both sides of the tapered waveguides. This brings about a reduced IL suffering from abrupt change at interfaces. In the AWs design, an array number N_a of 16 is used to cover the mode field diameter (MFD) of the far-field pattern imaged from the input waveguide of the first star coupler. The shallow-etched patterns (cyan) are linearly tapered to make the tapered waveguides (skin) transit between W_t of 1.7 µm and W of 380 nm within a length L_t of 6.67 µm. For single-mode operation in each AW and low channel crosstalk (XT) in the filtering responses of the AWG, a width W of 380 nm is utilized for bending (light green and dark blue) and rotated (light gray) straight waveguides. A wider width W_a of 680 nm is employed for vertical (red) and horizontal (red and dark green) straight waveguides to reduce phase errors that suffer from side-wall roughness fabricated on a high-index-contrast SOI platform. To connect between the waveguides with two different widths, tapered waveguides (purple) with a length L_t of 10 µm are used as transitions. For a lower non-uniformity in the filtering response of the AWG, six instead of four channel spacings, namely 120 nm, are chosen as free spectral response (FSR). Based on the FSR as well as the calculated dispersive property, i.e., effective indices versus wavelengths, of the individual waveguides with a wide width of 680 nm in the incremental regions, a required grating order m of 8 and a length difference ΔL₀ of 3.644 µm between the adjacent AWs are utilized. Bending waveguides (BWGs) with designed radii are employed to meet waveguide arrangements with pitches Λ_o of 2.6 µm at the head/end of AWs and Λ_i of 3 µm at output ports. Given that the required length difference ΔL₀ for dispersive property is smaller than twice of Λ_o, an S-shaped architecture with a larger footprint instead of a rectangular one is employed for the AWs design. To meet the required phase difference of the integer multiple of 2π for the adjacent AWs at the central wavelength, fifteen length differences are replaced with ΔL_i evaluated on the basis of the effective indices of the designed BWGs (dark blue) and horizontal waveguides (orange), where the subscript i represents the index of array number ranging from 1 to 15 for the corresponding AWs. A simulation for AWG is performed using three-dimensional (3-D) beam propagation method (BPM) provided by RSoft. In the simulation, a phase correction followed by a 3-D field generation is included between the simulation steps of two star couplers to ensure that the corrected phases contributed by incremental length differences of arrayed waveguides are obtained at the entrances of the tapered waveguides. Given the incremental regions implemented with the wider mode, i.e., the mode for waveguide width of 680 nm, the default phase correction provided by the software is replaced by a user-defined python function which is called along with other simulation steps under batch command mode. Simulated field patterns of the output start coupler operated at four channel wavelengths are shown in Fig. 2 while the simulated spectrum is given in Fig. 3. The result given in Fig. 3 shows that an IL of ∼3.6 dB with non-uniformity below 1 dB and an XT of less than −20 dB can be achieved without considering side-wall roughness.

 

Fig. 1. Schematic top view of (a) overall TBDTAWG, (b) input star coupler, (c) output star coupler, (d) magnified top view and cross section of one heater unit, (e) magnified length difference (dark green), and (f) magnified straight waveguide (orange).

Download Full Size | PDF

 

Fig. 2. (a–d) Simulated field patterns of output star coupler operated at four channel wavelengths.

Download Full Size | PDF

 

Fig. 3. Simulated filtering response of the proposed AWG for CWDM configuration.

Download Full Size | PDF

For the active design, a tungsten wire (vertical black traces within dashed triangles in Fig. 1) is used as a heater to tune the effective index of the silicon waveguide and consequently shift the filtering response of the AWG. The tungsten wire with a thickness of 300 nm and a width of 600 nm is placed 1 µm above the 220-nm-thick silicon layer. Density of 19300 kg/m³, specific heat of 134 J/kg-K, and thermal conductivity of 173 W/m-K are assumed and assigned to the heat transport properties while a sheet resistance of 650 mΩ/□ is used for the electrical property of the tungsten material. To achieve bi-directionally tuning functions and to simplify the tuning issue, two dashed triangular regions with complementary incremental phase shifting contributions in Fig. 1(a) are chosen as heating regions. Each triangularly heating region is composed of 120, i.e., 1 + 2 + … + (N_a − 1), heater units with each unit length of 6.6 µm for both the heating tungsten wire (black) and the heated silicon core (red). All heater units are expected to have the same values of voltage drop and current flow, respectively, in the tungsten wire, providing an equivalent thermal power and thus an equivalent phase shift for all heated units of silicon core. In each triangular region, 120 heater units are divided into 6 parallels, with each parallel consisting of 20 heater units in series to reduce the required electrical voltage for tuning. For example, 15 heater units inside the 16^th arrayed waveguide are connected to 5 heater units inside the 6^th one to reach 20 units. Under the arrangement, the thermal crosstalk between heater units is ignored and the simulation for thermal tuning could be simplified to an issue of how much phase shifts can be achieved under different thermal powers provided by each heater unit. Two-dimensional (2-D) simulations for the heater unit, shown in the cross section of Fig. 1(d), are performed using both heat transport and finite difference eigen-mode solvers provided by Lumerical Inc. for evaluation of the effective index changes under different unit thermal powers increased from 0 to ∼2.185 mW. To calculate the spectrum shift Δλ of the filtering responses of the tuned AWG, an expression obtained by (ΔL Δn_eff) / m [1] is utilized and derived into an equivalent one as (Δn_eff ΔL λ₀) / (n_eff0 ΔL₀), where Δn_eff and ΔL are parameters for the active design, representing refractive index change and the length difference, respectively, of the heated core, while n_eff0, ΔL₀ and λ₀ are parameters for the passive design, representing the effective index at the central wavelength, the length difference between adjacent waveguides of the AWs design, and the central wavelength, respectively. Note that the length difference of the heated core ΔL in the expression is the unit length in this design owing to each heater unit consisting of both heating tungsten wire and heated silicon core. Further discussions on tuning performances in terms of the heater unit length are given in the last section.

Figure 4(a)–(b) illustrate the simulated curves of Δn_eff and Δλ versus required total thermal power P and electrical voltage V_B, respectively, showing a direct proportionality between P and Δλ. The results show that shifted spectra by approximate 1.51, 6.04, and 9.45 nm are achieved with electrical voltages of 1, 2, and 2.5 V, i.e., total thermal powers of 41.9, 167.8, and 262.2 mW, indicating a linear shift-to-power ratio of ∼36.07 nm/W. Given the parallel configuration for heating patterns, the total thermal power P and electrical voltage V_B are obtained by P = 120 P_unit and V_B = (P_{20 units} R_{20 units})^1/2, respectively, where the subscript “unit” represents one heater unit. The expressions indicate that the required total thermal power P is directly proportional to a unit thermal power P_unit while the voltage is proportional to the root of P_unit and thus the root of Δλ, as shown in Fig. 4(b).

 

Fig. 4. Simulated refractive index difference (left y axis) of each heater unit and the corresponding wavelength shifts (right y axis) of filtering response in terms of different (a) thermal powers and (b) applied electrical voltages.

Download Full Size | PDF

3. Measurement results

The proposed TBDTAWG is implemented with the foundry service provided by Interuniversity Microelectronics Centre (IMEC) and measured with a pair of surface grating couplers operated at O-band and TE-polarized state owing to the convenience of surface coupling for measuring and verifying on-chip optical devices. The mask layout of the device is given in Fig. 5(a), wherein red and blue regions imply full- and shallow-etched patterns while the white background indicates the non-etched silicon layer. Note that there are dummy patterns being placed above the layer of silicon core to meet the processing window of the metal layer and to absorb un-guided or scattered light at the white region of the mask layout in the fabrication. However, no dummy patterns should be implemented inside FPRs of both star couplers given that the diffracted lights might be affected due to absorption. To prevent this, a pattern describing a region without dummy patterns is defined and covered on the entire AWG. In Fig. 5(b), the dark-gray region outside the AWG comes from the dummy patterns while the non-etched silicon regions such as FPRs and waveguides are visible with a lighter color. In the fabricated device, an additional input and four output ports are employed at two corresponding star couplers for autocorrection of the filtering responses by red- or blue-shift tuning in terms of the optical transmissions in the middle two of these extra output ports for Channels 1.29 and 1.31, i.e., two channels adjacent to the central wavelength of 1.3 µm, if a low-power input source operated at 1.3 µm is introduced. Detailed concepts for monitoring process will be given in the next section. In Fig. 5(a), the vertical black strips illustrate the heating tungsten wires connected for 6 parallels with each parallel consisting of equivalent 20 heater units to reduce the required tuning voltages. The electrical connections are realized with metal patterns in dark gray, as shown in Fig. 5(a), and a pad configuration of S₁GS₂ with a pitch of 100 µm. Under these arrangements, red-shifted spectra can be obtained with electrical voltages applied on GS₂ while blue-shifted ones can be measured with S₁G.

 

Fig. 5. (a) Mask layout of the proposed TBDTAWG and (b) photograph of the fabricated device on the SOI chip.

Download Full Size | PDF

Figure 6(a) and 6(b) show the measured results of red- and blue-shifted spectra, respectively, under electrical voltages of 0, 1, 2, and 2.5 V which correspond to the solid, dashed, dotted, and dash-dotted lines, respectively. An IL and XT of approximate 4 dB and −15 dB, respectively, are given in the figures. Compared to simulated spectrum given in Fig. 4, the measured IL for the passive design is increased by less than 1 dB, which could be attributed to side-wall-roughness losses due to (a) multi-mode excitations at the interfaces between tapered waveguides and FPRs and/or (b) inevitable waveguide couplings when modes exit/enter FPRs and enter/exit AWs. Some efforts discussed in the next section can be made to address the issue. In addition, the measured XT of the passive design is increased by ∼5 dB, which could be due to phase errors resulting from inaccurate incremental phase differences or side-wall roughness contributed by the arrayed waveguides. The measured spectra also imply that the thickness of 220-nm silicon layer might be deviated by a small amount, resulting in a blue shift of 7 nm for the passive design. A proper silicon thickness of 215 or 210 nm for simulation might lead to a better agreement with the measured results. For tuning performances, the measured results show that spectrum shifts of ±1.4, ±5.5, and ±8 nm are achieved with voltages of 1, 2, and 2.5 V, respectively, for each channel, where the symbol “±” represents bi-directional tuning functions. The results also show that differences of channel spacings due to tuning are less than 1.2 nm under these voltages, indicating a good shifting feature of the device. In addition, nonsignificant IL and XT differences in terms of tuning shown in Fig. 6 could be attributed to a thermally induced misalignment between the fiber and the chip due to the thermal expansion of the chip during the measurement. Fortunately, this issue could be resolved by permanently bonding the input/output fibers to the chip. Note that XTs are much more sensitive to phase errors when compared to ILs. Inaccurate thermal tunings of incremental phase shifts of arrayed waveguides will significantly increase XTs which are not observed in the measured result. Comparisons between simulated (black curve) and measured (red square) results are given in Fig. 7(a)–(d) and 7(e)–(h) for red- and blue-shifted spectra, respectively, at each channel, showing a good agreement between simulation and measurement, and indicating the feasible functions of thermally bi-directional tuning for the proposed device. Moreover, Fig. 8 illustrates a direct proportionality, as shown in Fig. 4(a), between total thermal powers P, i.e., 120 units of P_unit, and spectrum shifts Δλ. For a total thermal power P of 262 mW, an index change Δn_eff of ±0.0114 and a shift Δλ of ±8 nm are obtained from the measured results, indicating a linear bi-directional shift-to-power ratio of ±30.5 nm/W with a wide tuning range of 8 nm. The difference of absolute values of linear ratios between simulation (36.07 nm/W) and measurement (30.5 nm/W) might come from the power loss in the electrical metal traces or from a deviated thickness and/or sheet resistance of the fabricated heating wire. A performance comparison is given in Table 1, showing a better thermal-tuning efficiency with bi-directional tuning functions for the device proposed in this paper.

 

Fig. 6. (a) Red-shifted and (b) blue-shifted filtering responses for different applied voltages V_B of 0, 1, 2, and 2.5 V are represented by solid, dashed, dotted, and dash-dotted lines, respectively.

Download Full Size | PDF

 

Fig. 7. Effective index differences (left y axis) of adjacent AWs and wavelength shifts (right y axis) of filtering response in terms of electrical voltages at four output Channels 1.27, 1.29, 1.31, and 1.33 µm for (a–d) red-shift tuning and (e–h) blue-shift tuning, where the solid line and squared red marker represent the simulated results from Fig. 4 and the measured data from Fig. 6, respectively.

Download Full Size | PDF

 

Fig. 8. The linear relationship between the spectrum shifts and the required thermal powers at four output Channels 1.27, 1.29, 1.31, and 1.33 µm for (a–d) red-shift tuning and (e–h) blue-shift tuning, where the solid line and squared red marker represent the simulated results from Fig. 4 and the measured data from Fig. 6, respectively.

Download Full Size | PDF

Table 1. Thermal-tuning performance comparison of thermally tunable AWGs in the literature

View Table

4. Discussion and conclusion

From the simulated and measured (V_B = 0 V) spectrum shown in Figs. 4 and 6, respectively, slightly higher ILs and XTs are obtained for the passive design of the fabricated device, as mentioned in the former section. The higher IL of the passive design might come from side-wall-roughness loss due to multi-mode excitations and/or inevitable waveguide couplings near the interfaces between AWs and FPRs, where a tip width of 1.9 µm is used for the tapered waveguides. To reduce the IL, a mono-mode condition as well as the best coupling length for tapered waveguides with shallow-etched patterns in junction with FPRs proposed in [43] could be utilized. On the other hand, the increased XT of the passive design could be attributed to phase errors due to length difference errors between adjacent AWs as well as fabricated side-wall roughness at the bending waveguides. For a better agreement with the simulated passive spectrum, snapping patterns to integer points on mask layout without compromising simulated spectra and using improved fabrication processes might be helpful.

Given that both triangular regions with complementary incremental phase shifting contributions for active tuning are implemented outside two regions, which determine passive filtering responses, with modified incremental length differences of arrayed waveguides, a longer/shorter unit length can be realized with a larger/smaller pitch of arrayed waveguides at the upper one of two incremental passive regions determined by the modified length differences ΔL_i. Although a longer/shorter unit length can be achieved, a smaller/larger thermal power would be obtained under the same electrical voltage, leading to the same linear shift-to-power ratio as the one before using different unit length. The reason comes from the relation (direct proportionality) between the heating length, i.e., the length of each heater unit, and the tungsten resistance of each heater unit. The longer/shorter unit length is used in the device, the larger/smaller unit resistance would be obtained, resulting in a reduced/increased spectrum shift and thermal power by a simultaneous ratio under the same electrical voltage and thus the same linear shift-to-power ratio. In other words, a wider tuning range under the same electrical voltage can be implemented with a shorter unit length while no differences of linear shift-to-power ratio, which comes from material properties, would be made.

To implement a function for automatically stabilizing the filtering responses, i.e., thermally bi-directional tuning the spectral responses automatically until the condition is stabilized, a scheme employing an additional input port for a central wavelength and four extra output ports for the corresponding channels is proposed in Fig. 9. In this scheme, a low-power optical source operating at the central wavelength of 1.3 µm is utilized and introduced into the additional input port to monitor transmissions in the middle two channels, i.e., Channels 1.29 and 1.31, of the additional output ports. If the transmission at Channel 1.29 is higher than the one at Channel 1.31 when operated at 1.3 µm, the filtering responses of the AWG should be blue-shifted for the WDM system. On the contrary, if the comparison result is reversed, then the spectral responses of the device should be red-shifted. Because the speed of thermal conduction might be less than the one of voltage change that is controlled by the feedback loop, a proper proportional integral derivative (PID) control [44] can be applied to the feedback control circuit so that the stable condition of T_1.29 (1.3 µm) T_1.31 (1.3 µm) is satisfied. Here T_λch (λ₀) represents the transmission at the monitoring port for Channel λ_ch when operated at the wavelength of λ₀. Note that the PID control should be carefully optimized to compensate for the environmental thermal loss and to avoid temperature oscillation before the stable condition occurs so that the amplified differential voltage is in the range of 1–2 V.

 

Fig. 9. Scheme for automatically stabilizing the spectral responses of thermally bi-directionally tunable AWG for WDM communication systems. CWLPLD, continuous-wave low-power laser diode; SAWG, S-shaped arrayed waveguide grating; PD, photodiode; DA, electrical differential amplifier; D, diode; RSH, red-shifting heater; BSH, blue-shifting heater; TBDTAWG, thermally bi-directionally tunable arrayed waveguide grating. Green and blue routing wires represent optical and electrical connections, respectively.

Download Full Size | PDF

Considering the system signals and monitoring low-power optical source are both introduced into the device using the corresponding input ports when demultiplexing, the low-power source operated at 1.3 µm at the monitoring ports for Channels 1.31 and 1.29 will result in the spatial diffraction with lower (m − 1) order to the system port for Channel 1.33 given the designed FSR of 6 channel spacings and the monitoring input port being placed 4 channel spacings away from the system input port in the layout. On the other hand, the higher (m + 1) order of the 1.33-µm signal at the system port for Channel 1.33, i.e., the upper-right focused beam shown in Fig. 2(a), could also affect the monitoring one for Channel 1.29. However, the issue can be easily addressed with two solutions including: (a) using a wider FSR of 7 or 8 channel spacings at the expense of larger footprint of the entire device, or (b) placing the monitoring input port 3 instead of 4 channel spacings away from the system input port, so that XTs due to spatial diffractions with either the higher (m + 1) or the lower (m − 1) order can be avoided.

In conclusion, a thermally bi-directionally-tunable (TBDT), namely both red- and blue-shift tunable, arrayed waveguide grating (AWG) was proposed and demonstrated based on 193-nm lithographic complementary-metal-oxide-semiconductor (CMOS) technology. The device was composed of passive and active designs for realization of an AWG and fine tuning of its filtering responses, respectively. By leveraging an S-shaped architecture for the passive design and six parallel components with each one consisting of twenty equivalent heater units in series for the active one, a thermally bi-directionally tunable AWG (TBDTAWG) applied with ultra-low electrical voltages for wide spectrum shift can be achieved. An even wider tuning range can be realized with a shorter length of the heater unit without compromising the shift-to-power ratio or tuned/untuned spectra if sufficient spacings are provided for evanescent waves of arrayed waveguides. Measurement results show that both red- or blue-shifted spectra can be achieved and a linear bi-directional shift-to-power ratio of ±30.5 nm/W as well as a wide tuning range of ∼8 nm can be obtained under an electrical voltage range of 0–2.5 V. This shows a good agreement with simulation data and highlights a great potential for the proposed device to deal with fabrication errors and environmental temperature change.