Monitoring the spectrum and polarization state of light can provide key information about light–matter interactions, revealing nanostructures of a material [1] and its chemical composition [2]. Measurement of the wavelength-dependent state of polarization (i.e., Stokes spectrum) is required for the studies of vibrational circular dichroism [3,4], Rayleigh scattering [5], vector magnetograph [6], and so on. A spectropolarimeter is used to measure the Stokes spectrum in these applications, which, in some fields, is also called spectroscopic ellipsometer [7]. Among other applications, such devices play a critical role in the pharmaceutical industry for chiral separation and analysis of racemic drugs: many drugs are chiral compounds and marketed as racemates, whose chirality can be measured via circular dichroism [8].

Over the last decade, the demand for compact, cost-effective, and low-power spectropolarimeters has increased dramatically. Recently, some miniature architectures of such devices have been demonstrated [9–12]. However, these works were still exploring the possible improvements of traditional free-space optical components. Not surprisingly, this approach is limited to the decimeter scale footprint. A new paradigm is needed to reach qualitative changes. A chip-scale spectropolarimeter would be highly desired (particularly for portable, biomedical applications), which has yet to be reported. In this work, we will propose a chip-level spectropolarimeter in silicon photonic integrated circuits (PICs).

The proposed device includes four spectrometers and one polarimeter. Some high-performance on-chip spectrometers have recently been demonstrated on silicon PICs, such as the Fourier transform spectrometer (FTS) [13,14] and arrayed-waveguide grating spectrometer (AWGS) [15]. The silicon FTS typically consumes a significant power (at the watt scale [13,14]) for the thermal tuning of waveguide delay. Such a high power raises concerns about reliability and scalability for a lab-on-a-chip system. The power consumption of the AWGS is relatively low, but requires a large number of photodetectors, which complicates the measurement system and takes a large footprint [16]. The silicon FTS using arrayed Mach–Zehnder interferometers (MZIs) [17,18] also has such a problem. The recently demonstrated digital silicon FTS using arrayed MZIs plus an on-chip thermo-optic switch fabric achieved a single photodetector solution with a lower power consumption, but at the cost of a larger footprint and increased control complexity [19].

Here, we will propose a structure of a serially coupled double microring resonator (SDMR) to realize the chip-level spectrometer. With the advantages of small size, high tunability, and low-power consumption, microrings (MRs) are an excellent choice for a wavelength filter [20]. But, the inverse relationship between the size and free spectral range (FSR) limits its application in spectroscopy. The structure of the SDMR can well solve this problem. Because of the Vernier effect, the FSR of the SDMR can be significantly extended without decreasing the diameter of the MRs. More importantly, the temperature required to cover the entire FSR can be drastically reduced.

Compared to spectrometers, integrated polarimeters have been much less investigated. Only a few miniature full-Stokes polarimeters were demonstrated recently [9,21–27]. The capacity of on-chip probing a different dimension of photons than intensity and phase opened up immense opportunities for communications, quantum information, astronomy, and biomedical and chemical sensing. Our recent work [28] has proven that an optimal polarimetric frame with a minimum number of photodetectors can be achieved in silicon PICs, offering excellent performance comparable to conventional free-space solutions but with significantly improved compactness and robustness. Nevertheless, despite their broadband operation, none of these integrated polarimeters can capture the wavelength dependence of Stokes parameters.

In this paper, we propose and experimentally demonstrate, for the first time, a chip-scale spectropolarimeter in silicon PICs, which encompasses both functionalities of a full-Stokes polarimeter and four spectrometers. Analysis of an arbitrary state of polarization is realized using a 2D nanophotonic antenna and an on-chip interferometric circuit. With adopting the SDMR, the proposed device simultaneously achieves a high resolution (1 nm) and a broad bandwidth (50 nm) in the Stokes spectrum. The efficacy of the proposed spectropolarimeter is demonstrated by characterizing the chirality of a cholesteric liquid crystal (CLC) slab. The whole device, including an array of photodetectors integrated on the same chip, takes a compact footprint of 0.6mm2 and a mean power consumption of only 360 mW.

Figure 1 shows a schematic of the proposed device. It is designed based on a standard 220-nm-thick silicon-on-insulator (SOI) wafer with a 2 μm buried oxide layer and 3 μm oxide cladding. A surface polarization splitter (SPS) is used to split an arbitrary state of polarization into two orthogonal linearly polarized components (Ex and Ey) and couple them into different waveguides. A polarization analyzer (PA) in an interferometric circuit then converts the two orthogonal E-field components into four intensity channels. The spectrum of each intensity channel is measured using a chip-level spectrometer. The four spectral measurements capture the full information of wavelength-dependent polarization, from which we can eventually retrieve the Stokes spectra of the input light via a matrix operation.

The SPS makes use of a nanoantenna structure, consisting of a 2D array of sub-wavelength cylindrical holes on a Si substrate. The nanoantenna is designed so that both orthogonal linearly polarized components of the light, either from an optical fiber or free space, are coupled into the fundamental TE mode of the planar waveguides. Simultaneously, the SPS decomposes each orthogonal component equally into two paths in opposite directions, as shown in Fig. 1.

The PA circuit consists of a 3-dB broadband directional coupler (BDC) [29] and three Y-junctions [30] for 3-dB power splitting/combination. Taking the outputs of the SPS, the PA projects the Stokes vector of the incoming light into four intensity channels through interference operation: I1 and I4 from direct detection of 22Ex and 22Ey, respectively; I2 from the interference between i24Ex and 24Ey; and I3 from the interference between 1+i4Ex and 1+i4Ey. Here, we denote the incoming polarization by a wavelength (λ) dependent Stokes vector: S(λ)=(S0(λ),S1(λ),S2(λ),S3(λ))T, where (⋆)T means the transpose of the matrix (⋆). Defining a wavelength-dependent intensity vector: I(λ)=(I1(λ),I2(λ),I3(λ),I4(λ))T, we find the relationship between I(λ) and S(λ) given by where MS(λ) is the synthesis matrix of the PA. The BDC [29] used in our design has a wide bandwidth in excess of 100 nm. Thus, the synthesis matrix MS(λ) is practically wavelength insensitive in the spectral range considered in this work and can be written by the following expression:

Following the PA circuit, four silicon dual-microring resonator spectrometers (Si-DMRSs) are used to measure the spectra of four intensity channels, respectively. Each Si-DMRS consists of an SDMR and a germanium on silicon photodetector (Ge-PD). The MRs in the SDMR have slightly different FSRs. Because of the Vernier effect, as shown in Fig. 2, the cascaded architecture can achieve a largely extended FSR without using ultra-small MRs which are challenging for fabrication on a wafer scale. The extended FSR of the SDMR is given by [31] where FSR1(2), D1(2), and ng1(2) are the FSR, diameter, and group index of the single MRs, respectively; the subscript 1(2) indicates the first (second) MR. We have ng1≈ng2 when the diameters of the two rings are very close. According to Eq. (3), we can increase the extended FSR of the SDMR by decreasing the difference of the diameters. A metal heater is used on the top of each MR to individually vary their temperatures. Tuning the heating powers (HPs) applied to the MRs, the wavelength of each intensity channel, Ii(λ), can be continuously swept and then detected by a Ge-PD. More details about the design of the SDMR are shown in Appendix A.

Figure 3(a) presents a packaged prototype of the proposed SP. Details of the fabrication and packaging processes are described in Appendix D. The fabricated silicon photonic chip sits in the center of the printed circuit board (PCB). Its footprint is ∼1mm×0.6mm. The optical micrograph of the fabricated chip is depicted in Fig. 3(b). The footprint of each Ge-PD on the end of the right side is only 30μm×150μm. The chip includes 16 electric I/O ports. The connections between these I/O ports are presented in Appendix B. Figures 3(c)–3(e) present the scanning electron microscope (SEM) images of the SPS, BDC, and SDMR (silicon layer only), respectively. The effective area of the SPS for optical coupling is near 22μm×22μm, which can be adjusted by varying the number of grating periods. More details about the design and performance of the SPS are given in Appendix A. The mean period and diameter of the holes of the fabricated SPS are 558 nm and 345 nm, respectively, which agree well with the designed parameters. The diameters of the two MRs in each SDMR are 26 μm (MR1) and 22 μm (MR2), respectively.

Before our experiment with the full-Stokes spectropolarimeter, we first characterized a single Si-DMRS integrated with a Ge-PD on the same chip. Figure 4(a) shows the optical micrography of the fabricated Si-DMRS. A Ge-PD design without doped Ge or Ge-metal contacts [32] was adopted in our device. Because the germanium lattice is not disturbed by dopants or metal contacts, it allows for better performance in background loss, bandwidth, and dark current. The Ge-PD was measured to have a responsivity of 1.12 A/W and dark current of ∼15nA at −4V reverse bias, at 1550 nm wavelength. More information on the structure and performance of the Ge-PD is provided in Appendix A. Figure 4(b) shows the transmission spectrum from the drop port of the fabricated SDMR with a resonance wavelength near 1561 nm. A bimodal filter shape is observed, and its 20-dB linewidth is near 0.9 nm. A sweeping step of 1 nm was chosen in the following experiment. An extended FSR of 50 nm is measured in Fig. 4(c).

The center wavelength as a function of HPs applied to MR1 and MR2 was calibrated for each channel using a tunable laser. More experimental details about the calibration are provided in Appendix B. The calibration result is shown in Fig. 4(d). The tuning efficiency is ∼10mW/nm and ∼11mW/nm for MR1 and MR2, respectively. Thanks to the Vernier effect, the maximum HPs required to cover the entire extended FSR for MR1 and MR2 are only ∼70mW and ∼100mW, respectively. Figure 4(e) shows the measured transmission spectra of the Si-DMRS swept across the entire extended FSR (1530 nm to 1579 nm) with a step of 1 nm.

Figure 5 compares the spectrum of a broadband source recorded by a commercial optical spectrum analyzer (OSA) as a reference (solid black) and that measured by the Si-DMRS (dotted lines), where a good agreement is observed. To verify the stability of the proposed device, we perform several measurements within a week using the same HP calibration; results are shown in Fig. 4(d). The measurement results show excellent agreement over 6 days, indicating a very stable operation of our device. The measurement mean squared error between our device and reference is only 0.03.

The spectropolarimeter’s performance was tested using a CLC slab [33]. The schematic of the CLC slab is presented in Fig. 6(a). It consists of chiral molecules with a mechanical twisting power, which imposes a macroscopic helicoidal self-organization. As a result, the local average orientation of the long molecular axis is periodically rotating from layer to layer, forming a natural molecular helix (i.e., structural chirality). With a proper choice of the molecular mixture parameters, the CLC slab acts like a spectral resonant filter (e.g., Rocking filter) in a desired spectral range, which only left-handed (or right-handed) circular polarization can pass through. The most complex behavior occurs at the edges of the resonant wavelength range where polarization sensitive reflection and strong polarization rotations (along with strong dispersion) are present. To demonstrate the efficacy of the proposed spectropolarimeter, we fabricated a CLC sample with an edge of the resonant range near 1550 nm. More details about the fabrication of the CLC sample are given in Appendix C.

With a linear polarization input (i.e., S1/S0=1), the Stokes spectra of the CLC sample were measured by both a benchtop instrument and our device. Details about our experimental setup are provided in Appendix B. All the four Si-DMRSs were calibrated following the procedure described in Appendix B. The calibration results are also depicted in Appendix B. The wavelength-dependent synthesis matrix of the PA, MS(λ), was calibrated using four known independent polarization states.

Figure 6(b) shows the measured normalized Stokes parameters S1/S0, S2/S0, and S3/S0 obtained by the benchtop instrument (dotted line) and our device (solid line). Excellent agreement is observed in the measurement results between our device (solid lines) and the benchtop instrument (dotted lines). The resonant range of the fabricated CLC material is below 1.5 μm. In the resonant range, only right-handed circular polarization can pass through the CLC; as shown in Fig. 6(b), S3/S0 evolves toward −1, while S1/S0 and S2/S0 approach 0 as the wavelength decreases. While in the non-resonant range (beyond 1.58 μm), the CLC material does not change the input polarization state. Therefore, as seen in Fig. 6(b), S1/S0 increases gradually from 0 toward 1 with the wavelength, while S3/S0 increases from −1 to 0 in the non-resonant range.

Compared to the benchtop instrument, our device shows a mean squared error of 0.04 in polarization measurement [Fig. 6(b)]. It can be reduced by (1) optimizing the architecture of the PA, e.g., a broadband optimal PA was recently proposed [34]; (2) improving the thermal stability of the SDMR, which limits the accuracy of our spectrometer; and (3) reducing the noise of the on-chip Ge-PD.

The entire spectropolarimeter, consisting of an SPS, a PA, and four spectrometers with Ge-PDs, has a compact footprint of ∼1mm×0.6mm. In spite of the compactness, our device remains a high performance with a high resolution (1 nm) and broad bandwidth (50 nm) of Stokes spectrum, which, however, has still not reached its limits. For example, according to Eq. (3), we can obtain a bandwidth of 100 nm if increasing the diameter of MR2 to 24 μm. In addition, the spectral resolution of the proposed device can be further improved by another order of magnitude (to 0.1 nm) by decreasing the cross-coupling coefficient between the two MRs without introducing significant losses (as shown in Appendix A). Other solutions for resolution improvement include the use of an integrated Fabry–Perot cavity [35]. The performance of the PA can also be improved by adopting the optimal measurement frame as shown in our previous analysis [28]. However, increasing the number of measurement channels does not necessarily improve the performance because it reduces the power received by individual channels and thus the signal-to-noise ratio. With a proper design, an optimal performance can be achieved with four measurement channels, the minimum required for the reconstruction of full-Stokes vectors, in the presence of both additive Gaussian noise and shot noise [28].

Because of the employment of the Si-DMRS, the energy consumption is significantly reduced. Our spectropolarimeter only dissipates near 3.6 J of energy to complete one measurement of the Stokes spectra, a few orders of magnitude lower than traditional solutions. The tuning efficiency can be further improved by adding thermal isolation trenches near the MRs (>10 times) [36] and by increasing the sweeping frequency of the HP (>100 times). Due to the limitation of our setup, the sweeping frequency was only 5 Hz in our experiment. However, the thermal response time of the MR is lower than 4 μs, indicating that a sweeping frequency of 250 kHz is possible [37], which can also be improved by applying microheaters embedded in silicon waveguides [38]. Assuming a higher sweeping frequency of 5 kHz for a larger number of spectral sweeping steps of 1000 (versus 50 in our current experiment), the total energy consumption of the proposed spectropolarimeter is estimated to be only ∼72mJ. In this case, one measurement of Stokes spectra can be accomplished within 0.2 s.

The resonance wavelength of our SDMR is sensitive to the temperature. Thus, the thermal stability determines the wavelength accuracy, which can be understood by the relation between the wavelength drift (Δλ) of the spectrometer and the local temperature deviation (ΔT): where neff(λ0,T0) and ng(λ0,T0) are the effective and group indices of the MR waveguide, respectively; λ0 is the measured wavelength and T0 is the local temperature of the MR; and αL is the linear thermal expansion coefficient of waveguide. For example, assuming ΔT within ±0.1°C, the wavelength accuracy is about 7 pm according to Eq. (4), which can meet requirements of most practical applications.

Achieving an integrated spectropolarimeter on a silicon photonic chip paves the way toward fast, affordable full-Stokes spectroscopy. To decrease the cost and size of the device, traditional solutions come with a reduced number of spectroscopic components, and consequently, compromised measurement speed and Stokes spectral resolution. By contrast, our solution allows for simultaneous achievement of high speed and high resolution as all the Si-DMRSs can be monolithically integrated with little increase in footprint and cost. Our device is fabricated using industry-standard silicon photonics foundry processes, indicating an easier path toward mass production using established large-wafer manufacturing facilities. The operating frequency range can be readily extended to the visible and mid-infrared regions by using other complementary metal oxide semiconductor (CMOS) compatible material platforms (e.g., silicon nitride-based platform and germanium-based platform) but the same architecture. Leveraging the economies of scale and advantages of silicon PICs integration, the proposed spectropolarimeter has a vast potential for application in the fields of Internet of Things, pharmaceutical analysis, astronomy, and so on. Moreover, an array of our devices can be used to analyze a light beam with complex spatial modes such as optical vortex [39].

Acknowledgment. We thank Guan Xun for welding the electrical cable onto the PCB, and Jonathan St-Yves for the layout design of the photodetector. We also thank Simon Levasseur and Nathalie Bacon for their technical support.