Polarization-Insensitive Silicon Photonic Variable Optical Attenuator

Abstract

We propose and demonstrate a polarization-insensitive silicon photonic variable optical attenuator. The designed device uses a two-dimensional apodized grating coupler as a surface-normal coupling interface, which has the advantages of low-cost fiber packaging and polarization insensitivity. For optical attenuation, PIN diodes are inserted into each waveguide to act as optical absorbers. The compact device, featuring a footprint of 250 × 850 μm², exhibits a fiber-to-fiber insertion loss of 6 dB. Under a 3 V bias voltage, wavelength-dependent attenuation of 18 dB at 1295 nm and 26 dB at 1315 nm is achieved. Systematic characterization across diverse input polarization states confirms polarization-dependent loss below 0.5 dB under arbitrary polarization states, validating the device’s robust polarization insensitivity for wavelength-division multiplexing systems.

1. Introduction

With the accelerated evolution of information technologies, the exponential growth in global communication traffic has driven remarkable advancements in optical communication systems. In wavelength-division multiplexing (WDM) architectures [1,2], inter-channel signal intensity variations emerge as a prevalent issue that compromises receiver performance. This technical challenge necessitates the implementation of optical amplifiers and variable optical attenuators (VOAs) for precise power equalization [3,4,5]. Beyond channel balancing, VOAs serve critical functions in receiver sensitivity characterization, network diagnostics, and system validation through programmable attenuation simulation.

Recent research has explored diverse VOA implementations including optofluidic [6], liquid crystal [7], microelectromechanical systems (MEMS) [8], polymer-based [9], and planar waveguide configurations [10,11,12,13]. In addition to the traditional Mach–Zehnder Interferometer (MZI) structure of VOAs, mirror structures and multi-channel VOA arrays have also been derived to achieve high attenuation and reduce power consumption [14,15]. Among these, silicon-on-insulator (SOI) waveguide VOAs demonstrate particular promise due to their CMOS compatibility and superior integration potential. Conventional electro-absorption VOAs employing PIN junctions leverage the plasma dispersion effect in silicon. However, these devices inherently suffer from polarization-dependent performance limitations originating from the intrinsic birefringence of submicron silicon waveguides. While theoretical designs suggest polarization-independent operation through dimensional optimization, practical implementations remain susceptible to polarization sensitivity due to etching-induced sidewall roughness and manufacturing tolerances. Furthermore, fiber–chip coupling interfaces introduce additional polarization-dependent losses (PDLs), necessitating complex polarization management schemes incorporating beam splitters and rotators that increase system complexity and cost [16,17,18].

Two-dimensional gratings couplers (2D-GCs) can couple any polarized light and separate two orthogonal polarized wave components into two waveguide TE modes to achieve polarization diversity operation. However, in order to suppress second-order back reflection, traditional 2D-GCs are designed to interface with slightly tilted single-mode fibers. This tilted coupling scheme results in different coupling spectra of two orthogonal polarizations, leading to a parasitic PDL. Although there are several methods to reduce the PDL [19,20], a decrease in coupling efficiency (CE) always comes at an expense. Unlike the traditional 2D-GC, the two-dimensional apodized grating couplers (2D-AGCs) adopt a four-port design and perfect vertical coupling scheme, which eliminates the PDL in theory and breaks the design trade-off between CE and PDL [21,22]. Therefore, they are attractive for polarization-insensitive devices and applications [23].

In this paper, we present a polarization-insensitive VOA architecture incorporating a vertically oriented 2D apodized grating coupler. The device features four symmetrically arranged PIN diodes with unified DC electrodes, enabling wavelength-independent attenuation. Experimental characterization demonstrates 6 dB insertion loss with 18 dB attenuation at a 3 V bias, achieving a 0.5 dB average PDL at the operational wavelength, representing significant progress in polarization-insensitive photonic attenuation technology.

2. Device Design

Figure 1 presents the schematic configuration of the proposed polarization-insensitive VOA, integrating a pair of 2D-AGCs, eight parabolic mode converters, and four symmetrically distributed PIN diodes. The 2D-AGC splits incident fiber-coupled light into four identical quasi-transverse electric (TE) waveguide modes through polarization-selective coupling, eliminating transverse magnetic (TM) components via mode orthogonality. Identical waveguide geometries ensure polarization-independent transmission by maintaining equivalent optical paths for all channels. At the output port, the four optical signals recombine through reciprocal 2D-AGC coupling, achieving polarization-insensitive fiber–chip interfacing. Parabolic mode converters enable adiabatic mode conversion between grating couplers and single-mode waveguides, simultaneously minimizing the device footprint and insertion loss. Carrier injection through forward-bias PIN diodes induces plasma-dispersion-mediated absorption in silicon waveguides, generating wavelength-independent attenuation while preserving polarization insensitivity through structural symmetry. Due to the symmetric design of the grating structure and the waveguide, this integrated scheme is capable of operating under any polarization state and achieving polarization-independent performance.

The 2D-AGC serves as the polarization-insensitive coupling interface, functioning as polarization splitter and power splitter for orthogonal polarization states when the light is launched from a perfectly vertical aligned fiber. This bidirectional architecture enables simultaneous light separation and recombination through structural centrosymmetry, as illustrated in Figure 1b. To achieve high coupling efficiency, a combined method with hill-climbing optimization and particle swarm optimization (PSO) was employed to determine the optimal design parameters for each grating layer through a simple multi-parameter scan. The design was then validated using the three-dimensional finite-difference time-domain (FDTD) algorithm. This hybrid optimization approach enabled an efficient search for the optimal coupling efficiency. The optimized structure parameters are shown in

Table 1. Geometric dimensions of the 2D-AGC.

Parameters	G1–G6	G7	G8	G9	G10	G11	G12
Pitch P (nm)	450	460	460	450	450	440	440
Hole span S (nm)	270	280	273	265	256	250	240

.

Numerical optimization of the grating achieved 74.6% coupling efficiency at a 1304.1 nm central wavelength with polarization-independent characteristics, maintaining upward reflection (UR) below −13 dB near the operation wavelength, as depicted in Figure 2a. The inset i shows the cross-sectional light field distribution of the grating coupler, clearly showing the process of light transitioning from the fiber mode to the grating diffraction mode. It is evident that, apart from a small amount of light being reflected into the fiber or leaking into the substrate, the majority of the light is effectively coupled into the waveguide. Moreover, to observe the power distribution at the four ports, we simulated the light intensity distribution at a coupling polarization angle of 45°, as shown in inset ii, which clearly illustrates the grating behavior of polarization and power splitting.

Parabolic mode converters enabled 98% transmission efficiency across the 110 μm device length through adiabatic mode transformation between the grating couplers and single-mode waveguides. The attenuation mechanism employs four symmetric ridge-type PIN diodes (380 nm waveguide width) with coordinated forward-bias injection. Doping profiles were engineered to achieve optimal carrier injection through via-connected electrodes, as shown in Figure 1c. Carrier density modulation induced plasma dispersion absorption in the silicon waveguides, demonstrating monotonic attenuation enhancement with increasing bias voltage, as shown in Figure 2b. This unified architecture maintains polarization insensitivity through balanced optical paths and symmetric carrier distribution.

The simulations revealed the VOA’s polarization-independent performance characteristics, with Figure 2c showing the 3.75 dB insertion loss and 0.06 dB PDL at 1302.78 nm peak wavelength demonstrated under zero-bias conditions. The voltage-dependent attenuation analysis in Figure 2d shows the monotonic attenuation scaling from 3.75 dB to 19.47 dB with 0–3 V bias application, which achieved a 19.47 dB dynamic range at maximum bias while maintaining a sub-0.1 dB PDL across the operational spectrum. The adoption of a symmetrically designed grating structure and optical waveguides theoretically eliminates the PDL, thus enabling polarization-insensitive light transmission.

3. Results and Discussion

The device fabrication employed standard CMOS processes with integrated annular alignment markers featuring a 125 μm inner diameter designed to match standard single-mode fiber dimensions, ensuring precise optical alignment. Strategic openings patterned above the grating structure reduced the silicon dioxide layer thickness to enhance light transmission efficiency. Figure 3a displays the micrograph of the polarization-insensitive VOA with a compact 250 × 850 μm² footprint. Figure 3b is a scanning electron microscopy (SEM) image of the 2-D AGC. In the layout design, square-hole grating scatters were employed. However, under SEM observation, the actual size of the etched holes was found to be smaller than the designed dimensions, and the shape of the holes was more elliptical, exhibiting distinct rounded corners. This discrepancy can be attributed to the optical proximity effects that occurred during the photolithography process. Figure 3c shows the SEM image of the cross-sectional view of the optical attenuator. The post-fabrication measurements revealed an oxide opening depth of 4.379 μm, a total surface-to-waveguide thickness of 6.85 μm, and an actual cladding thickness of 2.471 μm, which deviated significantly from the design specifications.

Figure 4a shows the measurement setup for the optical spectrum and PDL of the device, where a tunable O-band laser (N7778C, Keysight Technologies, Santa Rosa, CA, USA) integrated with a polarization synthesizer (N7786C, Keysight Technologies, Santa Rosa, CA, USA) and an optical multiport power meter (N7745C, Keysight Technologies, Santa Rosa, CA, USA) are mainly employed. Light coupling was achieved through normal-incidence alignment of a single-mode fiber to the grating coupler, with real-time power monitoring ensuring optimal positioning. Systematic spectral characterization under varying bias voltages was carried out by automatic wavelength sweep.

The measured fiber-to-fiber IL of the VOA is 6 dB, of which the loss of the mode spot converter is 0.2 dB, the coupling loss of the 2D-AGCs is about 5.2 dB, and the rest is the loss of the waveguide. The voltage-dependent attenuation characteristics in Figure 4b demonstrate a maximum attenuation of 18 dB at the central coupling wavelength with a bias of 3 V, following the spectral envelope of the grating coupler response. It should be noticed that there is a discrepancy between the measured and simulated GC coupling curve. Compared with the simulation results, the measured central wavelength is red-shifted by about 8.5 nm and the spectra are reshaped. To understand this, we ran a calibrated simulation of the GC based on the SEM-adjusted critical dimensions. The simulation results of the coupling loss for back-to-back GCs are shown in Figure 4b, which aligns with the measured spectra of the VOA. The calibrated coupling loss per grating is 2.2 dB, which is near to the measured result of 2.6 dB. One may notice that the attenuation is saturated at 3 V and the spectral shape changes significantly. However, when the voltage reaches 3.2 V, the spectrum recovers to normal. This phenomenon may be attributed to an internal physical mechanism combined with plasma absorption and the Mach–Zehnder interference effect. Due to the fabrication non-uniformity between the four attenuation arms, there should be random phase differences and they can be accumulated or amplified by the increased bias voltages. This uncertainty may cause a spectral reshaping at specific wavelengths or an attenuation discrepancy between simulation and measurement, just as shown in Figure 4c,d.

The PDL characterization in Figure 4e demonstrates exceptional polarization insensitivity, achieving an average PDL of 0.5 dB near the operational wavelength. Through the polarization synthesizer, different polarization states can be achieved by controlling the Stokes parameters S1, S2, and S3. Three random polarizations defined by Stokes parameters were selected [SOP1: (0.44, −0.88, 0.18), SOP2: (0.59, −0.36, 0.73), SOP3: (0.91, −0.29, 0.31)] as well as SOP1′, SOP2′, and SOP3′, which are orthogonal to each other, respectively. This analysis revealed wavelength-dependent extinction ratio variations with a polarization-state sensitivity of less than 0.5 dB, as depicted in Figure 4d. The observed attenuation consistency across polarization states indicates robust polarization insensitivity, with residual fluctuations attributable to grating structural imperfections and Fabry–Pérot reflections within the waveguide cavity.

Table 2. Performance comparison of VOAs (^a: measured; ^b: simulation).

Ref.	Type	Length (μm)	Attenuation (dB)	PDL (dB)	Driving Conditions
[14]	Si-TO	125	35.5	-	10.8 mW
[15]	SiO2-TO	1.5 × 104	43	0.7 a	200 mW
[26]	Si-EO	30	2.6	-	1.5 V
[27]	MEMS	-	20	0.15 a	-
[28]	Si-TO	50	30	-	50 mW
[29]	LCPGs	-	50	0.3 a	20 V
[30]	Si-EO	1 × 104	20	-	470 mW
[31]	Si-EO	5000	4	0.2 b	50 mA
[32]	Si-EO	1380	60.11/59.08	-	33 V/3.2 V
This work	Si-EO	150	18	0.5 a	3 V

shows the performance comparison between our design and other similar VOA works realized using different methods. Compared to bulky polarization-insensitive VOAs based on LCPGs and MEMS, our silicon photonic solution exhibits the advantages of a small footprint and low power consumption. Notably, our design exhibits the best measured PDL performance among the silicon-based VOAs, by integrating a vertically oriented 2D-AGC with symmetrically arranged p-i-n diodes. As a proof-of-concept device, there is a lot of room for performance evolution. Future efforts should focus on optimization of IL, attenuation, and PDL. First, an inverse design methodology could be applied to the grating structure to systematically enhance coupling efficiency and minimize IL through precise geometric parameter optimization [24]. Second, using a non-linear and stepwise segmented compact taper can effectively reduce size and minimize losses [25]. Third, the attenuation characteristics may be improved by strategically extending the effective interaction length of the VOA or using a higher doping concentration. Finally, the implementation of thermal compensation mechanisms could address phase mismatch issues induced by environmental fluctuations, thereby enhancing the consistency between experimental measurements and simulation results.

4. Conclusions

This work successfully demonstrates a polarization-insensitive VOA operating at the 1310 nm communication band. The device employs fully vertical 2D-AGCs as polarization-diverse fiber–chip interfaces, achieving polarization-independent operation while significantly reducing packaging complexity and cost. With an ultra-compact footprint of 250 × 850 μm² and a PIN diode attenuation region merely 150 μm in length, the VOA delivers 18 dB optical attenuation at a 3 V bias voltage, accompanied by a PDL as low as 0.5 dB. These attributes position the proposed architecture as a competitive solution for next-generation polarization-robust photonic subsystems in fiber-optic communication networks.