Related Experiment Video
Updated: May 14, 2026
06:36
Optrode Array for Simultaneous Optogenetic Modulation and Electrical Neural Recording
Published on: September 1, 2022
Spatiotemporal Locality-Aware Adaptive Hybrid Optoelectronic Interconnect for Reconfigurable Array Processors
Bowen Yang1, Yong Li1, Rui Shan2
1School of Electronics and Information, Northwestern Polytechnical University, Xi'an 710072, China.
Sensors (Basel, Switzerland)
|May 13, 2026
Summary
This study introduces an adaptive framework for hybrid optoelectronic networks-on-chip (HONoCs) to overcome energy-delay bottlenecks. It enhances performance by intelligently managing electrical and optical resources for data-intensive applications.
Area of Science:
- Computer Engineering
- Integrated Circuits
- Optical Interconnects
Background:
- Electrical networks-on-chip (NoCs) face energy-delay bottlenecks due to RC-delay constraints in scaled reconfigurable array processors (RAPs).
- Hybrid optoelectronic NoCs (HONoCs) struggle with medium selection, balancing optical setup overheads against static thresholds that lead to inefficient resource utilization and congestion.
- Traffic heterogeneity and non-stationary loads exacerbate performance issues in current HONoC designs.
Purpose of the Study:
- To develop an adaptive switching framework for HONoCs that addresses the limitations of static medium-selection strategies.
- To improve energy efficiency and reduce latency in data-intensive applications on scaled RAPs.
- To provide a novel approach for dynamic resource management in hybrid on-chip interconnects.
Main Methods:
- Introduction of the Temporal-Spatial Locality Index (TSLI) to categorize data flows (Electrophilic, Photophilic, Hybrid-sensitive).
- Proposal of Cross-layer Congestion Entropy (CCE) for unified electrical and optical resource state assessment.
- Development of an Adaptive Medium Selection State Machine (AMSSM) and Weighted Multi-dimensional Medium Matching (WMMM) for dynamic mode switching and fine-grained channel selection.
- Implementation of Predictive Optical Path Provisioning (POPP) to mitigate setup latencies.
Main Results:
- Achieved 22% higher saturation throughput in 8x8 mesh HONoCs.
- Reduced the energy-delay product (EDP) by 38%.
- Decreased average latency by 57% under non-stationary traffic conditions compared to static threshold methods.
Conclusions:
- The proposed adaptive framework effectively manages hybrid electrical and optical resources in NoCs.
- The TSLI, CCE, AMSSM, WMMM, and POPP mechanisms offer a robust solution for energy-delay bottlenecks.
- This research provides a foundational paradigm for efficient, high-performance on-chip interconnects in future data-intensive applications.
