Wideband E₀₀-E₁₀ Silicon Mode Converter Based on 180 nm CMOS Technology

Abstract

Mode division multiplexing (MDM) is a promising technology for the capacity enlargement of the optical transmission network. As a key element in the MDM system, the mode converter plays an important role in signal processing. In this work, a wideband E₀₀-E₁₀ silicon mode converter constructed by Y-branch and cascaded multimode interference coupler is demonstrated. The theoretical mode crosstalk is less than –29.2 dB within the wavelength range from 1540 nm to 1600 nm. By 180 nm standard CMOS fabrication, the tested mode conversion efficiency of 91.5% and the crosstalk of −10.3 dB can be obtained at 1575.9 nm. The 3 dB bandwidth is over 60 nm. The proposed E₀₀-E₁₀ silicon mode converter is applicable in mode multiplexing.

1. Introduction

Advanced by next-generation communicating and large-quantity signal processing, high-speed data transporting and large-capacity data handling are highly demanded. Due to the unique feature of photonics, various multiplexing methods, such as wavelength division multiplexing (WDM), polarization division multiplexing (PDM), as well as mode division multiplexing (MDM), are explored to enlarge the network capacity [1,2,3,4,5,6,7]. Except for the fiber optic network [8,9], the on-chip MDM transmission implemented by optical waveguide is investigated, too [10,11,12]. The mode signal can be processed by multimode interferometer (MMI) [13,14], directional coupler (DC) [15], Y junctions [16,17], micro-rings [12,18,19,20], and grating-assisted [21,22,23] and inversely designed structures [24]. The mode converter, as one key device in the MDM system, could realize the conversion to desired high-order modes within a certain bandwidth with features of high efficiency and low loss. As reported, the mode conversion is commonly implemented by DC or grating structures [25,26], however, the bandwidth is restrained by the intrinsic quality of DC, which limits the combination with WDM technologies to enlarge the capacity of data transmission.

In this paper, a wideband silicon E₀₀-E₁₀ mode converter is proposed. Finite difference beam propagation method (FD-BPM) calculations are used for the design optimization. The mode crosstalk is less than −29.2 dB within 1540–1600 nm. The theoretical design is experimentally implemented by 180 nm standard CMOS fabrication. The measured highest mode conversion efficiency is 91.5% at 1575.9 nm. The 3 dB bandwidth over 60 nm can be obtained. This E₀₀-E₁₀ silicon mode converter has possible utilization in fields of mode signal multiplexing.

2. Device Design

In MMI waveguides, the self-imaging effect could reproduce the input field profile in single or multiple images at periodic intervals along the propagation direction. Two scenarios exist in the MMI effect. One is the general interference, another case is the restricted interference. The beat length L_π can be described by

$$L_{\mathsf{\pi }}=\frac{\mathsf{\pi }}{{\beta }_0{-\beta }_1}\approx \frac{{4n}_r{w}^2}{{3\lambda }_0}$$

(1)

where λ₀ is the operation wavelength, n_r is the refractive index of the core waveguide, and w is the width of the MMI coupler [27].

When one input waveguide is placed at the center of the MMI coupler, the distance L of Mth N-fold images can be defined as

$$L=\frac{M}{4N}{(3L}_{\mathsf{\pi }}).$$

(2)

For the MMI coupler with a certain length, and when only the fundamental mode or first-order mode is input, we have M = N = 1. The length of the MMI coupler L_MMI is

$$L_{\mathrm{MMI}}=\frac{3}{4}{L}_{\mathsf{\pi }}.$$

(3)

For instance, for the MMI coupler shown in Figure 1 below, two E₀₀ modes with π phase difference are launched into the MMI section through two single modes. The interference between two E₀₀ modes with different phases in the MMI waveguide leads to the E₀₀-E₁₀ mode conversion. The first-order E₁₀ mode will be excited at the output port on the right side of the MMI waveguide.

The proposed silicon E₀₀-E₁₀ mode converter that is constructed by three sections of symmetric Y-branch 3 dB splitter, phase shifter (PS) waveguide, and 2 × 1 MMI coupler is shown in Figure 2a. Firstly, two E₀₀ modes with the same mode power and phase are input into the PS waveguides via the 3 dB splitter. The π-phase difference would emerge between two E₀₀ modes after passing through the PS waveguide constructed by cascaded tapered waveguides with different lengths. The interference between two E₀₀ modes with different phases in the MMI coupler leads to the E₀₀-E₁₀ mode conversion. Figure 2b shows the cross-sectional view of the silicon waveguide.

When the silicon core thickness is fixed to 220 nm, the variation in effective indices (n_eff) of two modes with the silicon core width is illustrated in Figure 3a. To support the E₀₀ mode, the width of the input waveguide (w₁) is chosen as 0.4 μm. We set the width of the output waveguide (w₃) as 0.8 μm to support both the E₀₀ mode and E₁₀ mode. For getting π-phase difference, it is necessary to strictly manipulate the dimensional parameters of PS waveguides. With FD-BPM calculating, the width of the Y-branch 3 dB coupler (w₁) is set to be w₁ = 0.4 μm, and PS waveguides (w₂) are set to be w₂ = 0.28 μm to support E₀₀ mode. Additionally, the lengths of different tapered waveguides 𝑙₁, 𝑙₂, and 𝑙₃ are selected to be 3.6 μm, 12.4 μm, and 8 μm, respectively.

Here, the width of MMI coupler w_MMI is chosen to be 3.3 μm, thereby, the beat length L_π is calculated to be 32.5 μm at 1550 nm according to Equation (1). The length of MMI coupler that can be calculated by Equation (3) is optimized to be 25.3 μm for a compact size. With above parameters, the mode field distribution of the proposed mode converter at 1550 nm is shown in Figure 3b.

To be mentioned, the length of the Y-branch is chosen to be 25 μm, the length of the taper that connects the Y-branch and PS is chosen to be 8 μm, and the length of the straight waveguide that connects PS and MMI is chosen to be 5 μm. These parameters are compromise choices after comprehensive consideration of optical loss and device size.

The normalized transmission spectra and conversion efficiency of E₀₀ mode and E₁₀ mode investigated by FD-BPM calculations under 25 °C room temperature is presented in Figure 4 and Figure 5, respectively. The minimum mode crosstalk of −41.7 dB occurs at 1566 nm. Moreover, the crosstalk is less than −29.2 dB within the wavelength range from 1540 nm to 1600 nm, which implies the conversion efficiency of the E₀₀ to E₁₀ mode is larger than 87.9% within the interested wavelength range. The maximum mode conversion is around 99.6% happened at 1575 nm. The normalized transmission (T) as dB can be represented by Equation (4), where a is the transmittance of the corresponding mode.

T = 10lg(a)

(4)

To confirm the temperature characteristic of the proposed mode converter, the E₀₀ to E₁₀ mode conversion at different temperatures was investigated. When the thermo-optic coefficient of silicon is 1.88 × 10⁻⁴/K [28], the simulated conversion efficiency as a function of wavelength is as shown in the Figure 6. It can be seen that no obvious bandwidth change happens. However, when the operation temperature varies from 25 °C to 40 °C, the maximum conversion efficiency happens at 1575 nm, 1573 nm, 1572 nm, and 1571 nm, respectively. The blueshift of maximum conversion efficiency with the increment of temperature can be observed. Fortunately, the 3 dB bandwidths are all larger than 60 nm at temperatures of 25 °C, 30 °C, 35 °C, and 40 °C, which proves the favorable temperature stability of the proposed device. This phenomenon can be explained by the unremarkable size change of the silicon waveguide and the limited variation in the effective refractive index, which is favorable to stable mode conversion.

3. Results and Discussion

The proposed mode converter was fabricated on silicon on insulator (SOI) wafer with 180 nm CMOS technology. The thicknesses of the top silicon and buried silica layers were 220 nm and 3 μm, respectively. By deep ultraviolet (UV) photolithography, the waveguide patterns were transferred onto the photoresist. After development, the unprotected top silicon was removed by coupled plasma etching, which offers the rectangular shape silicon waveguides. Thereafter, 1 μm-thick silica was deposited by PECVD onto the SOI wafer as the top cladding. Figure 7 shows the microscope image (×1000 times) of the mode converter. The light from the fiber is coupled into the Y-branch through the grating coupler. Thereafter, the split modes pass through the phase shifters and interfere in the MMI section. The light output from the grating coupler is on the right side. To be mentioned, the layout of Y-branch 3 dB splitter is modified to satisfy the design rule of 180 nm CMOS technology.

The schematic diagram of measurement setup is shown in Figure 8. Figure 9 shows the picture that the converter is under test. After passing through the polarization controller, the light from the tunable laser source is coupled into the 3 dB splitter by a grating coupler via the single mode fiber (SMF). Since the converted E₁₀ higher mode is hard to directly observe through the single mode grating coupler, it is first filtered through a tapered waveguide. In this way, only the fundamental E₀₀ mode can pass and be collected by another SMF via the vertical grating coupler. Through a fiber power splitter, the obtained E₀₀ mode is coupled into the optical spectrum analyzer (OSA) (90%) and the optical power meter (OPM) (10%), respectively.

To find out the loss of the proposed mode converter, referential waveguides with different lengths are fabricated together with the mode converter. These referential waveguides consist of two identical grating couplers and a straight single mode waveguide. The propagation loss single mode silicon waveguide is confirmed to be ~1 dB/cm. Two grating couplers-induced loss can be deduced by removing the single mode waveguide-induced loss from the measured insertion loss of referential waveguide. Figure 10 shows the grating coupler-induced loss as the function of the optical wavelength. It can be seen that input/output gratings-induced loss within the wavelength range from 1540 to 1600 nm is lower than −25 dB. The minimum value of −25.06 dB happens at 1575.9 nm.

The measured transmission spectrum and conversion efficiency of the mode converter within the wavelength range from 1540 nm to 1600 nm is shown in Figure 11 and Figure 12, respectively. It can be seen that the highest conversion efficiency of 91.5% occurred at 1575.9 nm, and the corresponding mode crosstalk is −10.3 dB. The conversion efficiency in percent (E_p) and dB (E_d) are calculated by Equation (5) and Equation (6), respectively. Here, b is the propagation loss of the E₀₀ mode in the dB scale.

E_p = (1 − 10^0.1b) × 100%

(5)

E_d = 10lg(1−10^0.1b)

(6)

The 3 dB bandwidth is confirmed to be over 60 nm according to the measured E₁₀ mode spectrum in Figure 11. The fluctuation of transmission is due to the overclose distance between the input/output fiber and the grating coupler, which may lead to the instability caused by the Fabry–Pérot cavity. Therefore, the distance between the input/output fibers and the chip surface is supposed to be well controlled to guarantee the accuracy of the measurement. To be noted, the transmission is only characterized at 25 °C due to the limitation of the measurement setup.

Compared with the simulated results in Figure 4, the E₁₀ mode conversion shows good matching to experimental results. However, the measured mode crosstalk deviates from the theoretical expectations apparently. This partly results from the unavoidable fabrication errors. Due to the rules of 180 nm CMOS technology, compromises have to be made in the waveguide layout design. The change in waveguide dimensions compared to the ideal size inevitably affects the phase difference between the two branch arms. In the MMI coupler, the phase difference error will lead to the deviation of mode interference from theoretical expectations. Not all the E₀₀ mode was converted to E₁₀ mode. Partial E₀₀ mode power remains at the output, which deteriorates the mode crosstalk within the interested wavelength range. This effect first degenerates with the increment of the wavelength, then becomes serious again at longer wavelengths.

To demonstrate the proposed mode converter more clearly, performance comparisons between the reported E₀₀-E₁₀ mode converters and this work are shown in

Table 1. Comparison the proposed silicon E₀₀-E₁₀ mode converter with reported works.

Ref.	Footprint (μm2)	Insert Loss (dB)	Crosstalk (dB)	BW (nm)	Structure	Materials
[29]	2 × 2	2.2	−16.2	40	Inversely design	SOI
[30]	0.8 × 1.2	<1.2	/	50	Slots waveguide	SOI
[31]	6.3 × 3.6	2	−12	43	Photonic crystal waveguide	SOI
This work	3.3 × 79.3	0.4–2	>−10.3	>60	PS + MMI	SOI

. All referential experimental results are based on the SOI technology. The inversely designed mode converter shows the lowest mode crosstalk [29], while the mode converter, based on the slot waveguide, has the smallest size and the lowest insertion loss [30]. The photonic crystal waveguide-based converter exhibits balanced performances [31]. In comparison, the proposed work shows the moderate insertion loss and mode crosstalk. Though the largest size, the broadest bandwidth and lower E₁₀ mode insert loss promise favorable potentials in wideband operation.

4. Conclusions

In this work, a wideband silicon mode converter based on E₀₀-E₁₀ is theoretically proposed and experimentally demonstrated. With FD-BPM calculations, the simulated mode crosstalk is less than −29.2 dB in 1540–1600 nm. The standard 180 nm CMOS technology is applied in the fabrication of designed device, instead of the E-beam lithography. The best mode conversion, with an efficiency of 91.5%, is observed at 1575.9 nm. The comparatively wide 3 dB bandwidth of over 60 nm offers the proposed great potential of on-chip multiplexing systems in future.