Coherent Crosstalk Noise Analyses in Ring-Based Optical Interconnects

Luan H. K. Duong, Mahdi Nikdast, Jiang Xu, Zhehui Wang, Yvain Thonnart, Sébastien Le Beux, Peng Yang, Xiaowen Wu, Zhifei Wang2015 · DOI

TL;DR

This paper tackles the critical issue of crosstalk noise in optical interconnects for chip-scale communication. By developing detailed analytical models for both coherent and incoherent crosstalk at the optical-circuit level, the authors provide a framework to evaluate the real-world performance of different ring-based optical network-on-chip (ONoC) architectures, revealing that architectural design significantly impacts signal-to-noise ratio (SNR).

What Problem Does This Paper Address?

As computing systems demand ever-increasing bandwidth and lower latency, traditional electrical interconnects are reaching their limits. Optical interconnects, particularly those using microresonators (MRs) in wavelength-division multiplexing (WDM) configurations, offer a promising solution. However, these MR-based optical networks are susceptible to crosstalk noise, which arises from unintended signal leakage between different wavelengths or optical paths. This noise degrades the signal-to-noise ratio (SNR), ultimately impacting bit error rate (BER) and achievable data bandwidth. While previous work has touched upon crosstalk, this paper aims to provide a comprehensive, circuit-level analysis of both coherent and incoherent crosstalk in WDM ONoCs, specifically focusing on ring-based architectures, and to establish a methodology for comparing their realistic performance.

Methods

The authors developed analytical models for the fundamental optical elements used in ONoCs, primarily microresonators (MRs). They modeled the power transfer function of MRs using a Lorentzian profile, accounting for coupling efficiencies and intrinsic power losses. Crucially, they extended this to model both coherent and incoherent crosstalk noise. Coherent crosstalk arises from signals with similar optical propagation delays, while incoherent crosstalk occurs when these delays exceed the laser's coherence time. These models were then applied to two specific ring-based ONoC architectures: Corona and SUOR. The analysis systematically evaluated signal power loss and crosstalk noise at the optical-circuit level, considering various operational states of optical components like injectors, modulators, and detectors. The Signal-to-Noise Ratio (SNR) was used as the primary metric for performance evaluation.

Datasets

Not discussed in the paper. The paper uses parameters derived from existing literature and projections for future optical interconnect technology (circa 2020) for their analytical models.

Key Insights and Ideas

  • Architectural Design is Paramount: The study highlights that the specific architectural design of an ONoC, even within the same ring topology, can lead to vastly different worst-case SNR values – up to a 50dB difference between SUOR and Corona.
  • Coherent Crosstalk Dominance: In certain configurations (like Corona's data channel and broadcast bus), coherent crosstalk, often generated by inactive modulators, can be the dominant noise source, significantly impacting SNR.
  • Q-factor Impact: The quality factor (Q-factor) of MRs plays a critical role in SNR. While higher Q-factors generally improve SNR, the benefit plateaus beyond a certain point (around 14,000 in their analysis), suggesting diminishing returns.
  • Passing Loss Sensitivity: SUOR's SNR is shown to be much more sensitive to MR passing loss compared to Corona. This is attributed to SUOR's use of on-chip lasers, leading to more localized crosstalk generation that can overwhelm signals in specific configurations.
  • A Unified Comparison Platform: The developed analytical models provide a valuable tool for comparing the realistic performance of different ONoC designs, moving beyond ideal theoretical limits to account for the practical impact of crosstalk.

Key Results

The paper quantifies the worst-case SNR for different structures within Corona and SUOR. For instance, at a Q-factor of 9,000, the worst-case SNR for Corona's control arbitration was -51dB, while for Corona's broadcast bus, it was 13.38dB. The analysis also showed that small Q-factors (below 1000) severely degrade SNR. Furthermore, the study presented ideal and actual performance metrics, including bandwidth and energy per bit, demonstrating how crosstalk-induced SNR degradation affects these values. For example, Corona's data channel, with an ideal bandwidth of 160 Tb/s, sees a reduction to 114.19 Tb/s when considering worst-case SNR and error correction coding.

Limitations and Future Work

The authors acknowledge that their analytical equations are independent of specific data, and the numerical results are illustrative. They used a fixed die size for comparison, which might not be representative of all future designs. Future work could involve extending these analyses to other ONoC topologies beyond ring-based structures, exploring the impact of different WDM schemes, and investigating more advanced error correction codes to mitigate crosstalk effects. The paper also suggests that their framework can be used to compare the realistic performance of various optical interconnect networks by considering BER and data bandwidth degradation.

Summary updated

Abstract

Recently, optical interconnects have been proposed for ultra-high bandwidth and low latency inter/intra-chip communication in multiprocessor systems-on-chip (MPSoCs). These optical interconnects employ the microresonators (MRs) to direct/detect the optical signal. However, utilized MRs suffer from intrinsic crosstalk noise and power loss, degrading the network efficiency via the signal-to-noise ratio (SNR). In this paper, both coherent and incoherent crosstalk in wavelength-division multiplexing (WDM) networks are discussed and systematically analyzed. We carefully develop our analytical models at the optical-circuit level, and apply them to two ring-based networks: SUOR and Corona ONoCs. The quantitative results have demonstrated that the architectural design of the ONoCs determines the impact of crosstalk on the SNR. Even though SUOR and Corona are both ring-based ONoCs, the worst-case SNR can be differed up to 50dB. Our analyses of the worst-case SNR can be utilized as a platform to compare the realistic performance among different optical interconnection networks via the degradation of BER and data bandwidth.
BibTeX
@inproceedings{duong2015coherent,
  author = {Luan H. K. Duong and Mahdi Nikdast and Jiang Xu and Zhehui Wang and Yvain Thonnart and Sébastien Le Beux and Peng Yang and Xiaowen Wu and Zhifei Wang},
  title = {{Coherent Crosstalk Noise Analyses in Ring-Based Optical Interconnects}},
  booktitle = {Design,},
  year = {2015},
  doi = {10.7873/date.2015.0823},
}
Figure