Due to the limitation of electronic bottleneck, the traditional switching networks based on electrical devices are challenging to meet the needs of large-scale data exchange in the big data era. As an alternative, emerging silicon-based optical switching networks can effectively increase network capacity and reduce energy consumption. In addition, taking advantage of wavelength division multiplexing (WDM) or mode division multiplexing (MDM) technologies, multiple optical signals can be transmitted simultaneously in one optical link, effectively improving the optical interconnection density and communication bandwidth. However, the current optical switching network usually uses only WDM or MDM, which can only partially develop the parallel advantages of optics. Thus, we propose a novel on-chip optical switching network architecture based on hybrid wavelength and mode division multiplexing (WDM-MDM) technology. By inputting data of different wavelengths and modes into a single optical waveguide to achieve parallel transmission, the switching network's capacity multiplies. In addition, the proposed architecture enables data transmission between all nodes in parallel and supports multicast communication. The proposed architecture is expected to address the challenges of high-capacity switching requirements faced by data center networks by scaling the number of multiplexed wavelengths and modes.

We propose an on-chip optical switching network architecture (Fig. 1) based on hybrid WDM-MDM technology. The architecture uses Benes topology as the core switching unit to achieve non-blocking features. Besides, we design a mode selection/multiplexing (MSM) module-based transmitting module [Figs. 2(a)-2(b)]. All nodes are divided into 2N groups at the transmitting module, with each containing W processor nodes. Each node is connected to a microring resonator (MRR)-based modulator array at the transmitter side to realize electro-optical conversion. The modulated optical signal is then transmitted to the MSM module. In the MSM module, one of the input fundamental mode signals will be converted into a higher-order mode signal. Then these two different mode signals will be multiplexed into a multi-mode waveguide and transmitted to the input port of the hybrid WDM-MDM switching unit of 2×2 (Fig. 3). The switching unit contains two passive mode multiplexers (demultiplexers) and two double-ring MRR-based single-mode optical switching units of 2×2. Each input port in the hybrid WDM-MDM switching unit of 2×2 includes two modes and W different wavelengths, so it supports data exchange of any combination of 2W input data channels. Finally, we use Benes topology to cascade the proposed switching unit of 2×2 to form a large-scale hybrid WDM-MDM optical switching network.

We use the ANSYS Lumerical Solutions simulation platform to conduct device-level and system-level modeling and hardware parameter optimization for a 2-mode, 4-wavelength-based scale switching network of 16×16. The width of the single- and the multi-mode waveguide is 0.45 μm and 1 μm; the radii of the MRRs are 9.96, 9.98, 10, and 10.02 μm; the gaps 1-5 in CMR and SMR are 0.25, 0.2, 0.2, 0.47, and 0.2 μm, and the coupling length and gap in mode conversion region 1 are 15 and 0.20 μm, respectively. Figure 4 shows the transmission spectra and filed intensity distribution simulation results of the CMR switch, SMR switch, and asymmetric mode converter. Then, we analyze the transmission spectrum of the 16×16 switching network by choosing I₁ as the input port and measuring the transmission spectra of 16 output ports (Fig. 6). Among the 16 switching links, I₁→O₁₆ has the maximum insertion loss of about 9.30 dB, and the path I₁→O₁ has the minimum insertion loss of about 8.95 dB. The difference between the maximum and minimum insertion loss is less than 0.5 dB, which indicates that the switching network has excellent fairness. The input signal of I₁ is also modulated to four operational wavelengths, and the output signal is observed at the corresponding output port (Fig. 7), which verifies that the switching architecture is capable of multicast communication. In addition, we simulate the eye diagram with a single channel of 25 Gbps data rate and obtain the extinction ratio, rise time, and fall time results (Fig. 8). The results show that the detected eye diagram performance is better from I_1-4 to O_1-4, which means that reducing the number of MRRs turned on can effectively reduce the negative influence on the network performance. Finally, we compare the proposed architecture with traditional single-wavelength single-mode (SWSM) and multi-wavelength single-mode (MWSM) optical switching architectures. The results show that the proposed architecture can effectively reduce the insertion loss without increasing the wavelength cost, and this advantage becomes more obvious as the size of the network expands (Fig. 10).

We propose a scalable WDM-MDM-based on-chip optical switching network architecture and design an MRR-based wavelength/mode selection/multiplexing module and a WDM-MDM optical switching module of 2×2. As a proof of concept, a switching network architecture of 16×16 with 2-mode, 4-wavelength, and Benes topology of 2×2 has been simulated. Finally, the maximum insertion loss performance of the proposed architecture and the traditional SWSM-based and MWSM-based switching architectures are simulated and analyzed, and the superiority of the proposed architecture is proven. The proposed architecture can further increase the scale of wavelength multiplexing and mode multiplexing and reduce the insertion loss through structural optimization, which can help realize large-scale and high-capacity optical data center switching networks.