1. Introduction

High capacity chip-scale communication in the high performance computer and data center is one of the critical issues to be solved in the big data era [1, 2]. The traditional electrical interconnection comes across intrinsic problems due to the large power consumption and limited bandwidth [3, 4]. Especially for the intra-chip optical interconnection, where the high level of integration and high complexity are needed, such problems will be very urgent with the arrival of the limitation of Moore Law. Optical interconnection, featured with large bandwidth, low power consumption and multiplexing compatibility, is the promising candidate to solve such a problem. On the other hand, the development of silicon photonics provides a route for the chip-scale optical interconnection. In the past decade, various kinds of silicon based modulators [5, 6], polarization related devices [7, 8] and photodetectors [9, 10] had been reported. Many on-chip photonic links based on the silicon devices had also been reported [2, 11–16]. However, most of those proposed schemes were based on the single wavelength [11–15], wavelength division multiplexing (WDM) [2] or mode division multiplexing (MDM) [16]. To further increase the transmission capacity, other kinds of multiplexing techniques can be introduced. As one of the important multiplexing methods, polarization division multiplexing (PDM) had been explored extensively in optical fiber coherent communication [17]. Although some PDM based transmitter [18] and receiver [19, 20] have been demonstrated. However, most of these reported schemes were designed for fiber transmission, which are different from the proposed intra-chip interconnects. It is very meaningful to introduce the PDM technology into chip-scale interconnection systems to further enhance its transmission capacity.

We therefore propose and demonstrate a PDM photonic integrated circuit (PIC) suitable for chip-scale optical interconnection. The PIC is composed of light coupler and splitter, two micro-ring modulators (MRMs), polarization rotator splitter (PRS) based multiplexer/demultiplexer and two germanium photodetectors (Ge PDs). Data transmission using 20 Gb/s PDM signals are successfully demonstrated with clear eye diagrams, based on the proposed circuit. Under 10⁻⁹ bit error ratio (BER) level, the power penalty is about 1 dB between single and dual polarization cases, while the deviation for two polarization tributes is less than 0.5 dB.

2. Design and fabrication

The schematic and configuration of the proposed PDM PIC is shown in Fig. 1. The external light is firstly coupled into the PIC through a grating coupler (GC), and then divided into two paths by a multimode interferometer (MMI). In each branch, the electrical signal, such as from the central processing unit (CPU)/memory, is loaded onto the optical carrier as fundamental transverse electric (TE₀) mode through the MRM. In the multiplexer part, one of the TE₀ modes is converted into transverse magnetic (TM₀) mode while the other remains unchanged. The two signals with orthogonal polarization states are combined through the multiplexer and then transmit in a bus waveguide. At the receiver side, the multiplexed signals are firstly demultiplexed and then detected by two Ge PDs separately. The recovered electrical signals from the PDs are then sent into the corresponding memory/CPU. Under this scenario, the communication between the CPU and memory is realized. For proof of concept demonstration, the CPU and memory are not integrated in the proposed PIC.

 

Fig. 1 Schematic of the proposed PDM PIC. The inserted pictures are the details of (a) silicon MRM, (b) Ge PD and (c) PRS based multiplexer. MUX: multiplexer; DEMUX: demultiplexer.

Download Full Size | PDF

The silicon modulator used in the proposed PIC is based on the micro-ring structure due to its small footprint and low power consumption. The structure of the MRM is shown in Fig. 1(a). The input light from the straight ridge waveguide coupled into the ring with 10 μm radius. Both widths of the straight and ring waveguides are 520 nm, and the gap between them is 230 nm. Along the ring waveguide, the boron and phosphorus doping form the P + and N + regions, respectively. The doping concentration of the P and N regions are ~$2\times {10}^{18}{\text{cm}}^{-3}$ and ~$1\times {10}^{18}{\text{cm}}^{-3}$, respectively. P + + and N + + regions with both ~$1\times {10}^{20}{\text{cm}}^{-3}$doping concentration are formed to realize ohm contact on both sides of the waveguide. The RF signals are applied onto the micro-ring through the coplanar waveguide (CPW) electrode with ground-signal-ground (GSG) structure.

The polarization multiplexer and de-multiplexer are realized by the PRS. It is based on bi-level taper TM₀-TE₁ converter [7], with structure and detailed parameters shown in Fig. 1(b). The TE₀ signal from port 1 transmits across the straight waveguide and maintaines TE₀ at output port 3. For the TE₀ signal injecting from port 2, it will be converted first into first-order transverse electric (TE₁) mode through the adiabatic directional coupler. Then, the TE₁ mode is further transformed into TM₀ signal using a bi-level taper, which has a 90 nm slab. Therefore, both TE₀ and TM₀ light will be obtained at port 3. If the polarization multiplexed signals are injected from port 3, the PSR functions as a polarization demultiplexer.

In the detection part, the Ge PDs transform the demultiplexed optical signals into electrical ones. The gain peaking technology is adopted in the Ge PD to extend its bandwidth [9, 10], by introducing an inductor into the equivalent circuit of the typical Ge PD to form a RLC circuit. In effect, the increased inductance will counteract part of the capacitance at high frequency, and thus the S₂₁ curve will fall off slowly or has a little lifting at high frequency, which will result in an improved bandwidth. The circular spiral on chip inductor with 60 $\mu m$ out radius and 540 pH inductance value is specially designed to realize the large bandwidth, and the structure of such a gain peaked Ge PD is shown in Fig. 1(c). The compact MRMs used in the PDM PIC restrict the total bandwidth of the photonic link, which could be enlarged if the Mach-Zehnder interferometer (MZI) modulators were used instead. The width, length and depth of the Ge region are 5, 10 and 0.5 μm, respectively. The bottom P + doped silicon region with ~$1\times {10}^{19}{\text{cm}}^{-3}$ doping concentration together with the top N + + doped Ge region with ~$1\times {10}^{20}{\text{cm}}^{-3}$ doping concentration form the vertical PIN junction. The CPW electrode connected with the Ge and Si regions is used for high frequency electrical signal transmission.

The GC with 70 nm etching depth and 620 nm period is used for light coupling. The 1 × 2 MMI, which has a 2 μm width and 7.7 μm length, divides the input light into two paths. The whole PDM PIC containing all the above devices is fabricated on the silicon on insulator (SOI) wafer with 220 nm top silicon and 2 μm buried oxide (BOX). In our demonstration, although only straight waveguide is utilized for transmission, it will be quite meaningful if the bend waveguide is taken into account. Simulation shows that the minimum bending radii for TE and TM are 5 and 10μm, while the transmittance can be larger than 99%. The PIC is fabricated at the Institute of Microelectronics (IME) in Singapore. The top view microscope image is shown in Fig. 2(a) and the details of the grating coupler, modulator, multiplexer and photodetector are shown in Figs. 2(b)-2(e).

 

Fig. 2 (a) Top view microscope image of the fabricated PDM PIC; the detail microscope images of the (b) grating coupler, (c) modulator, (d) multiplexer and (e) photodetector.

Download Full Size | PDF

3. Device characterization

Before characterizing the performance of the proposed PIC, all the separate devices are measured based on the reference structure. The direct current (DC) and alternating current (AC) characteristics of the MRM are first measured and the results are shown in Fig. 3. When the thermal heating voltage is 8.35 V, the spectrum shifting covers a full spectrum range (FSR) of 9.5 nm in Fig. 3(a), which corresponds 1.14 nm/V tuning efficiency. The bandwidth of the MRM is measured using the lightwave component analyzer (LCA) and the measured S₂₁ curve is shown in Fig. 3(b). Under 2V reverse biased voltage, its bandwidth is estimated to be about 13 GHz. The relative smaller bandwidth is limited by the large parasitic parameters and large Q factor, and it can be improved by using the interleaved PN junction or redesigning the micro-ring's Q factor [5].

 

Fig. 3 (a) DC spectrum shift versus thermal heating voltage; (b) measured S₂₁ of the MRM under 2V reverse biased voltage.

Download Full Size | PDF

Then the performance of the PRS is characterized using the reference structure with edge coupling input/output light, and the measured transmission spectra are shown in Fig. 4. The spectra have been normalized by deducting the coupler profile. It can be observed from Fig. 4 that the average transmission losses are less than 0.5 dB for both through and cross ports of the PRS. In Figs. 4(a) and 4(b), one specific polarized light is injected from port 3 (denoted in Fig. 1(b)) and two orthogonal polarization components are measured at the through (Fig. 4(a)) or cross ports (Fig. 4(b)), respectively. Under such a condition, lower than −15 dB crosstalk for through port and lower than −13.5 dB for cross port over the wavelength range from 1530 to 1570 nm can be observed. In Figs. 4(c) and 4(d), TE₀/TM₀ polarized light inputs from port 3 and the TE₀ components are measured at both through and cross ports, obtaining larger than 15 dB extinction ratio over a full C-band. The crosstalk can be further improved by cascading polarization cleanup filters [7].

 

Fig. 4 Measured spectra at the (a) through and (b) cross ports when input two orthogonal polarized light from port 3 (denoted in Fig. 1(b)). Measured spectra from both through and cross ports when input (c) TE₀ and (d) TM₀ polarized light from port 3.

Download Full Size | PDF

The PD is also characterized based on the reference structure, and the measured results are shown in Fig. 5. Figure 5(a) shows the dark current and responsivity with respect to the reverse biased voltage. Under 3 V reverse biased voltage, the dark current and responsivity are estimated to be about 0.9 μA and 1 A/W, respectively. The bandwidth of the Ge PD is measured using the LCA with careful calibration and the result is shown in Fig. 5(b). It can be observed that the bandwidth is about 43.7 GHz when the reverse biased voltage is 3 V.

 

Fig. 5 (a) Measured dark current and responsivity of the referenced Ge PD. (b) Measured S₂₁ curve of the Ge PD when the reverse biased voltage is 3V.

Download Full Size | PDF

The link performance of the proposed PIC is then characterized, with experiment setup shown in Fig. 6. A CW light from the tunable laser at 1550 nm is coupled into the PIC through the GC, and a polarization controller (PC) is used to maximize the coupling efficiency. Two 10 Gb/s data streams from the bit pattern generator (BPG) with pseudorandom bit steams (PRBS) length of 2³¹-1 are applied onto the MRMs after being amplified by the drivers. Meanwhile, the bias and driving voltages of the MRM are 2 and 0.5 V, respectively. In the detection parts, the converted electrical signals are first amplified by the microwave amplifier and then injected into the digital communications analyzer (DCA) and error analyzer (EA) for analysis. Meanwhile, the reverse biased voltages from the digital source meter are applied into the MRMs and Ge PDs through the bias-tee. Although the ring resonators are very thermal sensitive. The thermal caused variation is avoided by adjusting the thermal voltage during the measurement. On the other hand, the feedback controller can also be introduced to solve this problem.

 

Fig. 6 Link performance experiment setup. Amp: microwave amplifier.

Download Full Size | PDF

The measured eye diagrams are shown in Fig. 7(a). Taking the single input TE₀ light for example, when only MRM2 (denoted in Fig. 2) loading with electrical signal, the TE₀ light carry signal while the TM₀ light maintains continuous wave (CW). Therefore, only PD1 detects signal (denoted as single case TE₀ to TE₀) and PD2 detects nothing (denoted as single case TE₀ to TM₀). Similarly, when inputting TM₀ polarized light (only MRM1 loading electrical signal), only PD2 detects signal (denoted as single case TM₀ to TM₀) while PD1 detects nothing (denoted as single case TM₀ to TE₀). Compared with the single input case, the conclusions are the same when two orthogonal polarized signals are adopted (PDM case), excepting the additional degradation of the eye diagrams. The degradation can be attributed to the polarization crosstalk resulting from the PRS.

 

Fig. 7 (a) Measured eye diagram (time scale: 40 ps/div) when inputting single polarized and PDM signals. (b) Measured BER results when inputting single polarized signal and PDM signals.

Download Full Size | PDF

In order to further characterize the proposed PIC quantitatively, the BER measurement is performed and the results are shown in Fig. 7(b). Under 10⁻⁹ BER level, the power penalty is less than 0.5 dB between the two orthogonal polarized signal for both single and dual input cases. Such a penalty mainly caused by the non-equal transmission loss between the two unsymmetrical polarizations branches of the PRS. According to the measurement results shown in Fig. 8, for the reference 1.06 mm long single mode strip waveguide, the difference between the transmission losses of the TE₀ and TM₀ is less than 0.7 dB/mm. Therefore, the PDM signals can transmit stably and have little influence on the power penalty. For the same polarized light, the power penalty is about 1 dB between single and PDM cases.

 

Fig. 8 Measured transmission spectra of the straight waveguide for TE₀ and TM₀ modes.

Download Full Size | PDF

All the measurements are performed with several different chips, and no obvious differences are observed from chip to chip. The total power consumptions of the modulator and PD are estimated about 130 fJ/bit and 4.4 pJ/bit, respectively.

4. Conclusions

We proposed and demonstrated an on-chip PDM PIC suitable for chip-scale large capacity optical interconnection. Such a PDM PLC is composed of modulation, multiplexing/demultiplexing, transmission and detection parts. The link performance of the proposed PDM PIC is fully characterized. Clear eye diagrams are observed for both single and dual polarization input cases. The proposed PDM PIC can be further combined with other multiplexing techniques to meet with the large bandwidth demand in future chip-scale optical interconnection.