Silicon photonics, a technology that’s reportedly been in development for about 16 years, is finally here as a product, and if it wasn’t for the world’s hunger for cloud services (and cloud service providers’ eagerness to provide them), the technology could have remained an interesting science project for many more years to come.
Machine-to-machine traffic in cloud data centers is “exploding,” as Alex Bjorlin, VP of Intel’s Data Center Group and general manager of its Connectivity Group, put it. The systems inside these facilities have to crunch through more and more data, which means more data than ever is being shuffled from server to server.
So much, in fact, that cloud service providers are desperate for better ways to interconnect servers in their data centers than using copper wires. They need to transfer more data faster and use less power doing it, which are all things silicon photonics promises to give them.
It replaces copper wires with optical cables and electrons with light, while using the same manufacturing methods companies use to make processors. Today, these laser-based systems are capable of shuffling 100 Gigabits per second over cables up to 2 kilometers long. You’ll struggle to push 25Gbps over a 10-foot-long copper cable, Kushagra Vaid, who oversees infrastructure engineering for Microsoft’s cloud, said from stage at the Intel Developer Forum in San Francisco Wednesday.
Vaid was speaking alongside Diane Bryant, executive VP and general manager of Intel’s Data Center Group, who announced availability of the chipmaker’s first two silicon photonics products: 100G optical transceivers for interconnecting data center network switches. According to Vaid, silicon photonics is the future of interconnection inside Microsoft’s cloud data centers.
Intel’s announcement is just the first step. The company said the 100G technology will be followed by 400G interconnects for data center switches “within a couple of years.”
The first two transceivers simply plug into switch front plates, but the company envisions future-generation products that are embedded in switch motherboards and eventually integrated with switch silicon itself.
In the same keynote Wednesday, Bryant also announced that the next generation of Intel’s Xeon Phi processor for supercomputers will be designed specifically with artificial intelligence workloads in mind.