New Ge-Si Photodetector Reaches 336 Gbps for AI Data Centers

This new photodetector can send data 3 times faster than previous models, reaching 336 Gbps. This is a big step for AI data centers.

Researchers led by Dr. Wei Chu have demonstrated a reconfigurable germanium-silicon (Ge-Si) photodetector capable of achieving data transmission speeds of 336 Gbps. Published recently in Advanced Photonics Nexus, the study addresses the bottleneck in AI-driven data centers by utilizing a low-loss integration strategy known as fan-out wafer-level packaging (FOWLP).

Performance MetricAchievement
Transmission Rate336 Gbps
Packaging StrategyFan-out wafer-level (FOWLP)
Core MaterialGermanium-Silicon (Ge-Si)
Key Structural ComponentHeterogeneous Redistribution Layer (RDL)

Engineering for Signal Integrity

The transition to higher bandwidths in server-to-server communication is currently obstructed by physical limitations in electronic-photonic integration. Traditional manufacturing methods frequently suffer from signal degradation and physical space constraints, preventing the density required by modern machine learning workloads.

The approach employed by Dr. Chu’s team relies on a heterogeneous redistribution layer designed to mitigate these losses:

  • Grounded Coplanar Waveguides (CPWG): Used within the RDL stack to maintain signal fidelity across high-speed interconnects.

  • FOWLP Integration: By moving away from conventional assembly, the team reduced the signal path complexity, effectively scaling interconnect density.

  • Reconfigurable Architecture: The photodetector’s adaptability allows for stable signal detection across diverse communication environments, addressing the stability-to-speed trade-off often found in high-speed communication systems.

Background and Context

The rapid expansion of large language models necessitates a fundamental rethink of how data centers move information. As of May 19, 2026, the reliance on optical interconnects is total, yet the physical manufacturing processes for these components have struggled to keep pace with the power and speed requirements of newer hardware clusters. By optimizing the link between photonic detection and electrical routing, this work attempts to bridge the gap between legacy packaging constraints and the requirements of future high-performance computing clusters.

Read More: Amazon and Argos cut Final Fantasy VII Rebirth price to £43.99

The study confirms that the use of an RDL stack allows for high-quality signal detection without the typical trade-offs in energy efficiency.

Frequently Asked Questions

Q: What is the new Germanium-Silicon photodetector and what speed can it reach?
Researchers have made a new Germanium-Silicon photodetector that can send data at a very fast speed of 336 Gigabits per second (Gbps). This new technology was announced recently.
Q: How does this new photodetector help AI data centers?
This photodetector helps AI data centers work much faster by sending data between computers at high speeds. It solves a problem where data flow was too slow for AI tasks.
Q: What new technology is used in this photodetector for faster speeds?
The photodetector uses a special packaging method called fan-out wafer-level packaging (FOWLP). It also uses grounded coplanar waveguides within a special layer called a redistribution layer (RDL) to keep the signal strong.
Q: Why is this photodetector important for the future of computing?
As AI models get bigger and need more data, faster ways to move information are needed. This new photodetector helps bridge the gap between old packaging methods and the needs of future high-speed computers.