Paper Analysis | The 300mm Silicon Photonics Platform Strikes Back: A 500-Micron Ultra-Compact MZM Takes On the 400Gbps-per-Lane Limit
Reference for Readers
Paper title: A Compact Mach-Zehnder Modulator in 300 mm Silicon Photonic Platform towards 400Gbps/lane Transmission
Authors: Fenghe Yang, Erse Jia, Ying Wang, Xinran Zhao, Weisheng Wang, Haiwen Cai, Wei Chu
Affiliation: Zhangjiang Laboratory, Shanghai, China
Venue: Optical Fiber Communication Conference (OFC) 2026
In-Depth Figure Analysis: From Microstructure to Full-Wafer Measurement
The key to how the team made the device so small is the "slow light" effect. A photonic crystal waveguide raises the group index, strengthening the interaction between light and the electrical signal. Here is a detailed breakdown of the key experimental figures:
Figure 1: Design and Fabrication of the Slow-Light Modulator
Figures (a) & (c), spectrum and group index: Simulations show a lattice constant of Lambda = 245 nm in the one-dimensional photonic crystal (1D PhC) waveguide. At 1310 nm (O-band), it achieves a group index as high as n_g = 16.3. This means light travels slowly inside, so the same phase shift requires a shorter length.
Figure (d), cross-section and doping: Shows the lateral P-N junction design. To balance optical loss and modulation efficiency, the team precisely controlled the doping concentrations (P-type 3.8E17 cm-3, N-type 7.1E17 cm-3).
Figure (e), the 300 mm wafer: This demonstrates commercial volume production capability. Using a 40 nm CMOS-compatible process, extremely small and uniform photonic crystal structures can be fabricated reliably on 12-inch wafers.
Figure 2: Static and Dynamic Characterization
Figures (a) & (b), insertion loss (IL): The measured static insertion loss is only 2.9 dB, and the full-wafer measurement shows a median insertion loss of 2.4 dB. This breaks the conventional impression that photonic crystal waveguides are lossy.
Figures (d) & (e), EO bandwidth: This is the most striking data point. The measured electro-optic bandwidth has a median as high as 94.7 GHz.
Conventional silicon photonics MZMs usually need very long electrodes and complex traveling-wave designs to approach 100 GHz bandwidth; this work reduces capacitance with cascaded P-N junctions, achieving ultra-high frequency response over a very short 500 µm length.
Figure 3: Large-Signal Transmission Experiments (Showing Its Muscle)
Figures (b) & (c), NRZ format: Still achieves a 3.2 dB extinction ratio (ER) at 112 Gbps NRZ.
Figure (f), PAM-8 format: This is the climax of the paper. Without any advanced AI algorithms (such as DNN reconstruction), using only standard FFE equalization, it achieves single-lane 135 Gbaud PAM-8 (equivalent to 400 Gbps) transmission.
This work deserves attention for three core reasons:
A revolution in size: At 500 µm, it is one-fifth the length of a conventional MZM or even shorter. This is a big win for CPO (co-packaged optics), because it allows more channels to be packed around the switch ASIC, achieving higher bandwidth density.
Process stability: Slow-light devices were long considered "fine for lab demos, but they collapse in volume production," because photonic crystals are extremely sensitive to dimensional errors. This result shows that on a 300 mm platform with a 40 nm process, manufacturing stability has improved substantially.
Clean bandwidth without AI assistance: Many solutions claiming 400G/lane today rely on extremely complex, power-hungry digital signal processing (DSP) and AI compensation. This work achieves 400 Gbps at a 2.7 Vpp drive voltage without AI, which means lower power and lower latency.
In summary: Combining a small footprint, high bandwidth, and high manufacturability, this device is an important building block for future 1.6T and even 3.2T optical modules.
Comments