The bandwidth demands of hyperscale data center operators have been increasing tremendously over the last years. First prototypes of 1.6 Tb/s (8 channel, 200 Gb/lambda) modules have been released recently. As the explosive bandwidth demands driven by new applications such as (generative) artificial intelligence (AI) show no sign of stopping, this panel aims to highlight the requirements for next generation 1.6 Tb/s and beyond connectivity, i.e. aiming at 3.2 Tb/s, from the data center operators’ perspective. Also, it shall provide clarity on the desired module specifications by discussing recent advances in electronic and especially photonic integration.
Topics will include the following questions:
- What are the needs of hyperscale data center operators in cloud and AI infrastructure?
- How much parallelism (number of wavelengths, spatial channels) do we need?
- What are ultimate and practical limits of symbol and per lambda rates?
- How will nonlinear crosstalk and residual CD be solved?
- What power consumption per module can be handled? This panel will comprise experts from hyperscalers, switch and module vendors as well as chip suppliers sharing their views on next generation intra-data center networks. Interaction between speakers and audience through Q&A is highly encouraged in a controverse panel discussion.
Juthika Basak, Nokia Corp., United States
James Chien, Marvell, United States
Stephan Pachnicke, Christian-Albrechts Universität zu Kiel, Germany
Ben Lee, NVIDIA, United States
Xiang Liu, Huawei, China
Radha Nagarajan, Marvell Technology, United States