11:00 – 11:30
Toward Energy Efficient Cloud Networking for Post-Moore’s Era
Fotini Karinou, Senior Researcher, Microsoft Research

The trend of rapidly increasing intra-DC traffic due to AI and disaggregation driven workloads call for low latency  and high bandwidth networks that are ultra-power efficient. In this talk we will discuss how innovation in optical technologies and cross-layer cloud network architectures could help overcome those challenges in the post-Moore’s law era.

11:30 – 12:00
Optical networks for disaggregated data centers and distributed deep learning
Georgios Zervas, Professor of Optical Networked Systems, University College London

Data Centers of any size and form are made of a range of diverse technologies and aim to support a plethora of applications in a cost-effective and sustainable manner. This talk will report on the challenges and opportunities for data centres and the key role that optical technologies and networks should play in the years ahead. I’ll delve into the importance of co-design across technologies, networks, and computing systems before reporting our latest results on optical switching, network architectures and control for disaggregated and distributed large scale deep learning systems.

12:00 – 12:15
Next Generation PON technologies Beyond 10
Derek Nesset, Optical Access Research Director, Huawei Technologies (Munich Research Centre)

The deployment of fibre access systems (FTTH) for delivering broadband services is accelerating in the UK. Passive optical networks (PON) are a key technology helping to deliver that growth. GPON has been widely deployed and some UK network operators are deploying the next-generation XGS-PON offering 10Gbps line rate. However, demand for even higher bandwidth PON systems is expected for FTTH and other services and the ITU-T has very recently standardised 50G-PON. This presentation will review the 50G-PON standards and the progress of technology demonstrators and carrier trials. Furthermore, we will take a look into the future concerning PON technologies beyond 50G.

12:15 – 12:30
AI-enabled fibre-optic access networks for smart cities
Elias Giacoumidis, Principal Application Engineer, VPI Photonics

The collaborative R&D project KIGLIS envisions artificial intelligence (AI) enabled fiber-optic communications for smart cities. It targets a considerable increase in data rates with almost identical hardware costs using AI methods. Dynamic bandwidth allocation and specific latency requirements will enable multiple access to passive optical networks (PON).

We present the performance limitations of different PON solutions based on direct-detection and coherent reception for up to 100 Gb/s and 31.5 dB power budget. We demonstrate effective electro-optical converted signal processing optimization using machine learning (ML) techniques. We show that sufficient receiver sensitivity and dispersion tolerance can be achieved by combining a low-chirp optical transmitter in the O-band and simplified receiver-side feed-forward equalization (EQ) in a PAM4 configuration. In C-band, ML-based-assisted EQ can reach the stringent power budget PON requirements when both coherent and direct-detected systems are employed.

 

Back