Silicon Photonics sounds like a pretty high-level and niche technical topic - and it is.
However, the more you learn about it, its larger importance in the context of 2019’s impending 5G rollout emerges. We spoke to Intel’s VP of its Data Center Group & GM Connectivity Group, Alexis Bjorlin, and Bryan Madden, Intel’s director of marketing for its Network Platforms Group, at this year’s Mobile World Congress about why the company is betting big on silicon photonics.
According to Bjorlin, “what we are oing in the Connectivity Group is focusing on an area of growth called smart intelligent fabrics - which we call highly engineered networks - and then photonics itself.”
What is Silicon Photonics?
If you Google the phrase ‘What is Silicon Photonics’, you’re probably going to come away with more questions than you are answers. Intel themselves define it as “a combination of two of the most important inventions of the 20th century—the silicon integrated circuit and the semiconductor laser.” Like I said, even the simplest description tends to invite more and more questions.
It’s best to start at the beginning.
For pretty much as long as we’ve been transmitting data in the form of electrical signals, we’ve been doing it with copper wiring. It’s simple, it’s cheap, it's easy to get and it gets the job done. Recent years have seen a shift towards optical fiber as a (partial) replacement on large-scale networking projects (like in Australia’s own NBN). However, for less-noticeable yet vital uses - specifically, as an interconnection medium on most integrated circuitry - copper wire is still the standard to beat.
The issue here is that with the arrival of technologies like 5G and the faster data speeds they enable, the maximum throughput that copper allows for has the potential to act as a bottleneck on computing speeds.
How does Silicon Photonics solve that problem?
According to Alexis, “the transition to 5G is going to require up to 100-gigabit per second from the mast-head”
“As the ingress of data at the edge becomes higher and higher - 10x increase, you have a 10x requirement increase in bandwidth to take that data and either - no matter where you’re going to process that data - you can take any of the models where you do core or central-based, you can do edge-processing but no-matter-what the data needs to be transmitted and moved.”
“Not only between physical locations but also within data centers themselves, between the compute elements, the storage elements, between accelerator pools or application-specific computing.”
“For every single element of data that’s being processed, you ultimately need to move it, that’s how the connectivity group comes into play,” she explains.
More specifically, this is where silicon photonics comes into play. With silicon photonics, you can use patterned silicon to transmit data-carrying laser signals. As well as carrying the potential to allow more data to be moved around faster while also consuming less power, silicon photonics can also be easily manufactured at the same mass scale as current silicon based technologies.
How is Intel involved?
Well, according to Alexis, “[Intel] have the world’s only fully on-silicon silicon photonics devices. We manufacture our lasers right on the 300mm silicon platform. We use standard silicon processing technology. So we’re able to manufacture photonics at the scale of traditional semiconductors.”
“Why is that important? It’s important because as you have seen with the tremendous increase in data [and] in data centers being processed....in general, the thing that’s been holding back deployments, scale-outs and build-outs of greenfield data centers, during the last two transitions, it’s been the optics that has held back the transition of the industry.”
Alexis tells us that Intel launched their first silicon photonics-based solution in 2016 and that their second product (launched in 2017) - the Intel 100G CWDM4 transceiver - will be utilized for the company’s upcoming 5G trials.
According to her, “we’re the only silicon photonics solution that can put four 25Gbps wavelengths of light onto a single fiber. So that enables a lower cost-infrastructure because you can use a single bit of fiber and transmit 100Gbps over it”
Bryan Madden says that “the next step is a 400Gbps transceiver and that’ll be coming out later this year. So the end of this year, you’ll see the advent of 400GB modules.”
Who is this technology going to affect?
At present, Intel’s primary audience for silicon photonics seems to be datacom, telecom and cloud-based data center businesses.
Alexis elaborates, saying “one of the reasons that Intel is focused on and that we’re deployed at volume in the cloud service providers is that we’ve focused on specific segments of the market where scale is required. At the 5G deployment will require that scale at high-bandwidth.”
Beyond this, there’s also a lot of potential demand for silicon photonics in the military and aerospace industries. The potential sell here is that silicon photonics allows for computing solutions that take up less space, offer more efficiency and are immune to electromagnetic interference.
For many of the same reasons, there are a lot of potential applications of the technology in the wearables and IoT space.
Touching on the impact that silicon photonics adoption could have on consumers, Bryan Madden says that “looking into the 5G era, latency is a challenge and an opportunity at the same time. So having really-optimized networks that have really fast speeds makes sure that latency is achieved. You’re talking about a 0.5 millisecond latency, [compared to] 30 milliseconds today.”
Specifically, he says “for connected car, it’s going to be very important”
“Every connected car creates 4TB of data a day, that needs to be communicated across the network as well.”
Touching on the long-term vision for the technology, Alexis says “the reason that it’s so important, is that as you continue to go up in speeds ultimately it becomes more and more difficult to escape large switch chips without the use of integrated photonics. So you can start thinking about it as a conversion from electric IO, that will convert over time to photonic IO - and that’s really the holy grail of what we’re trying to achieve here at Intel.”
“Ultimately, we’ll be merging and co packaging or even full integration of optical technologies onto our networking silicon platforms, ultimately that’s the goal.”
Who are they competing with?
There are a number of other silicon photonics companies but none that are in production anywhere near the scale of Intel, nor any with the fully-integrated silicon designs being offered by Intel.
“We really have the most integrated solution,” Alexis insists.
If anything, at this stage, the biggest competition Intel is up against here is the fiber-optic industry, which some analysts say will be worth US$5 billion by 2021.
Alexis argues that “currently, the 4G LTE deployments are serviced with traditional fiber-optic competitors but when you get to the scale and the data rates that are required (by 5G) and the cost structure required…” silicon photonics looks like a pretty appealing way forward.
For a longer look at everything that happened at this year's Mobile World Congress in Barcelona, click here.
Disclosure - Our coverage of MWC 2018 was sponsored by Intel who covered the cost of our flights and accommodation.