Variation-Aware Photonic System Modeling and Design

Optical interconnects provide high bandwidth, low power and low latency compared to traditional electrical interconnects. It is promising that the optical interconnects should replace the electrical links in short reach applications, such as data centers, inter-chip, and intra-chip communications.


distances

However, the optical interconnect systems are very sensitive to temperature fluctuations and fabrication process variations. In the photonic integrated circuit (PIC) era, conventional temperature stabilization techniques (e.g. the thermoelectric cooler, TEC) are unaffordable and the process variation issues become more severe. Therefore, variation-aware designs are needed for integrated photonic systems. We tackle this problem in different abstraction levels and approaches:

  • Microring Reconfiguration: Microring resonators are widely used in optical network-on-chips (NoC). However, the microring structure is highly sensitive to thermal and process variations: its resonance wavelength will deviate from the carrier wavelength, which will cause function failure. We proposed four power-efficient tuning methods to compensate for the wavelength mismatch:

  • tuning approaches


  • TiO2-based athermal optical devices: We utilize the negative temperature-optical coefficient of TiO2 to offset the positive temperature-optical coefficient of Si, to achieve temperature-insensitive optical devices.
  • Variation-aware adaptive tuning: The performance of optical device varies significantly due to fabrication process variations and thermal variations. The power budget of the optical link can be set based on the worst case to guarantee the yield, but wastes much energy. We propose an energy-efficient adaptive tuning technique that allocates just-enough power for each fabricated link using on-chip self bit-error-rate test (BERT).
  • Compact models for microring modulator: Accurate compact models are developed for carrier-injection microring modulators, including static and dynamic, optical and electrical properties. The model can predict the modulators' quality factor, extinction ratio, and eye diagrams for different device geometries and driving conditions. The models are implemented in Verilog-A and are compatible with SPICE simulators.
  • Spatial pattern analysis of wafer-scale test data of microring modulator: Significant spatial patterns are identified from the optical and electrical test data of silicon microring modulators with heaters. The spatial patterns are related to fabrication process steps and would be helpful in process diagnose and uniformity improvement.
  • Dynamic laser power scaling (DLPS): Laser power consumption is a significant portion of the total power consumption of an optical link. The laser is kept at the same output level even when there is no or few link traffic. Based on the observation that the required laser power decreases when the link data rate decreases, we propose a dynamic scaling technique that decreases the laser power or turns off the laser when there is few or no network traffic.

Selected Related Publications:

Current students: Rui Wu, Yuyang Wang

Past students: Dr. Chong Zhang, Dr. Fan Lan, Dr. Yan Zheng, Dr. Jock Bovington