Intel is revamping its strategy around the data center by going beyond the Xeon chip and into silicon photonics transceivers. The company announced Monday at the European Conference on Optical Communications (ECOC) that samples of its silicon photonics transceivers targeting 5G wireless infrastructure and data centers are available now, with production set to start in the first quarter of 2019.
The company notes that global data center IP traffic is increasing significantly. In 2016, global data center IP traffic was 6.8 zettabytes, and that will triple by 2021 because of all this data generated by humans and the Internet of Things (IoT).
The choke point becomes copper wire, the standard for Ethernet connectivity. Copper wire can only effectively transmit about eight to 10 meters, said Eoin McConnell, director of marketing for the connectivity group in Intel’s data center group. Fiber optics can go as far as 10 kilometers.
He cited the example of one Facebook data center that is the size of eight football fields and said 10 meter connectivity just won’t do in that environment.
“Hyperscale data centers need to think about their networks. Connectivity can constrain or enable the performance and they need to look at how photonics can deliver on their SLAs,” he told me.
Ethernet plays just fine within the server racks and cabinets, and Intel is not targeting that layer. Instead, it’s going after the fiber-optic switches that sit on top of the racks to communicate with other servers, other data centers, the internet, and cellular networks.
In addition to networks, Intel is looking into how can silicon handles the increased capacity of forthcoming 5G networks.
“As we look out into the 5G horizon and whether it’s increasing the number of spectrums or the number of antennas to handle data, it will drive a significant need in increased capacity from service providers,” said McConnell.
McConnell acknowledged this isn’t exactly an empty market. There are multiple fiber-optic vendors producing these modules.
“The thing we have a difference around is how we do this from a volume manufacturing perspective, in assembly and test,” he said.
Intel can mass produce the connectors automatically, while other vendors have to hand-assemble the components.
Intel’s 100 gigabit silicon photonics transceivers, which meet the bandwidth demands of 5G wireless, fronthaul applications, but the company is not stopping there. Earlier this year, the company demonstrated its 400 gigabit silicon photonics capabilities, and samples of its 400G silicon photonics products are expected to be available next quarter with volume shipment of the modules in the second half of 2019.