Intel is getting ready to launch its silicon photonics interconnect technology, which it intends to deliver an increase in bandwidth not just for data centre networks but also within servers themselves by replacing copper connections with fibre optics.
At an event in London attended by V3, Intel's business development director for Silicon Photonics, Jeff Demain, said that the technology is set for official launch in the near future. He declined to specify a date, but hinted that more would be revealed at the next Intel Developer Forum (IDF) in San Francisco in September.
Demain said that Intel was focusing on two main form factors for silicon photonics; a pluggable one such as the MXC connector it developed in partnership with Corning, and an embedded one that can be mounted on the motherboard to provide high-speed optical links directly from the processor.
"This is the way optics will go long term," Demain said. "As bandwidth goes up, driving a signal from wherever your electrical device is on the board to the edge of the board to go into the optics becomes a challenge, and 25Gbps is a very common next bandwidth point. At this speed, even going nine inches is a challenge, so now we make it possible to put the optics right next to the driving device."
Intel will be launching both form factors, according to Demain, showing off an optical module (pictured) designed to drop into a motherboard connector like a processor does. This module interfaces to a standard optical fibre rather than the alternative of using wave guides embedded in the motherboard, which Intel regards as a less flexible option.
The optics module itself is protocol agnostic, and can be used as a conduit for optical distribution of PCI Express (PCIe), the QuickPath Interconnect (QPI) used to link Intel processors with the outside world, or Ethernet.
In fact, one of the usage cases Intel is promoting silicon photonics for is to replace existing Ethernet transceiver modules in data centre switches, a move it has already begun with the launch of the 100G CLR4 Alliance in April.
However, with its MXC connector set to offer a cost-effective way of carrying up to 1.6 terabits over distances of hundreds of metres, data centres can even do away with the top of rack (TOR) switch and flatten out the network topology, Demain said.
This fits in with Intel's Rack Scale Architecture initiative, which aims to disaggregate the server into separate pools of storage, compute and memory that can be allocated as necessary to meet the changing requirements of workloads, and upgraded separately without having to rip and replace the entire rack.
Intel believes it has an advantage in the developing optics market because of its chip manufacturing expertise. Demain said that Intel's differentiator is hybrid silicon technology, which enables it to combine the light generating or "lasing" material with the on-chip silicon circuitry in a single manufacturing process.
"We're going to print optical devices in silicon now, which means that we have all that high volume capability available, so it's a new era of optics where the technology gets everywhere. It gets into the car, it gets into the server, it gets into the TV," he said.
Everything we think we know about the imminent Apple iPhone 9, iPhone 11 and iPhone 11 Plus launches
All the latest rumours about Apple iPhone Displays, CPUs, launch dates and even prices
Nvidia brings Turing microarchitecture into the high-end gaming segment
Did you make the shortlist for the UK's most respected IT event?
Latest Tesla news: Tesla share price continues to fall after Saudi Arabia's sovereign wealth fund is linked to investment in rival
SEC 'probe' takes its toll on Tesla as new research suggests that Tesla loses $6,000 on every $35,000 Model 3