The good optics of silicon photonics: Light sailing serenely down a fibre
15 years in the making. Out of the labs and into the real world: The 100Gbps Internet connection
Column Why on earth would Cisco want to push fibre to your desk?
It's a good question. Even modern video editing houses are content with 10Gbe to workstations, and most of us aren't modern video editing houses. Even Wi-Fi can reliably deliver speeds at or near the gigabit. No, you don't need fibre to your desk. But don't tell me you don't fancy it anyway, just for the geek of it.
And that's the real reason Cisco has done this. It's not one that fits onto spec sheets or use cases - not today's use cases, at least. It has to know why and you have to know how a piece of solid-state physics is creating quite a battleground in the data centre, one where Cisco is locking horns with Intel, and how some long-term bets may be about to pay off. But in the end, it's all down to the photon versus the electron.
Raw power in IT - raw useful, pay-the-bills power - often gets limited by bandwidth, more so than computation or storage. We have become astonishingly adept at creating enormous oceans of data and near-infinite armadas of CPUs to sail them with, but getting the one into and out of the other keeps hitting the barriers of boring old physics.
Electrical signals need lots of power to overcome noise at high speeds, and soon dissipate in the losses of cabling. Wireless is worse. Light, however, sails serenely down a fibre, indifferent to interference and with a tiny fraction of electronics's watts per bit. A little photon goes a long way.
The standard equation has been to have electrical signalling in a concentrated area, converting to light for longer distances. Conversion is relatively expensive, needing optical components that compared to chips are complex, fiddly and exotic.
It doesn't scale well. Electrical signalling scales much better. The same magical silicon Moore's Law that has given us seas of gigahertz cores for tuppence has done much the same for the technologies that drive very fast local data buses and networks.
But silicon has another trick - it can create and detect light. Why not replace the expensive, fiddly, bulky optical stuff with silicon circuits? That's the idea behind silicon photonics.
Moving to the real world
Both Intel and Cisco like silicon photonics. A lot. They and others think it can hugely increase connectivity and bandwidth while reducing cost and power. Which it can - but it has been a long time coming.
Intel is a pioneer. It first paraded its silicon photonics in public around 15 years ago, when it predicted that we'd all be using light, not copper, to connect our circuits together not just from box to box but inside, replacing a lot of the copper wiring between chips. Unlimited terabits for teeny wattages in very small spaces, the company promised, just as soon as it ironed out a few small production challenges.
Cisco intros desktop switches, one with USB-C to power your laptopREAD MORE
You may have noticed your 2021 workstation is not a sea of light inside (Gaming rigs with their Blackpool illuminations do not count. Yuck.) That's because those production challenges were - and are - very challenging indeed. While silicon can indeed handle light like a boss, it has to be built into configurations that are very different to those for swiftly wrangling electrons.
Light components on-chip are much, much bigger than transistor circuits- it's like having to build a multistory car-park on the same production line as you're building the cars. And ordinary chips go through hundreds of stages requiring different levels of heat - the hottest first, with layer after layer of ever more subtle, ever cooler processes following. That's not how you make silicon light components. Try making concrete in the same smelter as iron.
Nonetheless, Intel kept at it, neither closing the photonics effort down or selling it off, which if you know Intel shows how tempting the rewards must be. Now more than a decade after first going public, it's still making major announcements and doing rather well actually selling the stuff.
The market for silicon photonics is all in the data centre, where there's the highest premium on bandwidth, density and power. The numbers aren't huge - by revenue, silicon photonics transducers, the core components, bring in around $350m, predicted to grow ten times in five years. Intel has around 60 per cent of that market, with Cisco in second place at around 30 per cent on the back of buying specialist companies Luxtera in 2018 and is still trying to snaffle Acacia. This is an active market.
But while data centres are great places to be, the market outside is even more tempting. Intel's vision of light buses inside computers seems as far away as ever, because interchip connections and the capabilities of SoC inside single packages have kept up with demand. We just don't need to go to lightspeed there. But outside the box?
What silicon photonics can enable is the 100Gbps internet connection. That's not theoretical - Intel has a silicon photonics transceiver that does 100Gbps down 10 kilometres of fibre. The effect of such speeds - at latencies which can approach the speed of light, you're not going to get better with Einstein in charge, is that data centres start to feel like they're plugged into your expansion bus.
Local and remote get very blurry. Fancy a fully immersive VR session, live on a tropical beach for an hour in the middle of a British winter? A front-row seat at the loudest, brightest, most banging gig you can imagine? Tell me that doesn't sound like a good idea right now.
Silicon photonics and related areas in fibre physics and optical processing are hot like never before. Data centre demands alone make so, and prove that the technology is out of the labs and in the real world. That real world is hungry for data, for new experiences, new ways to use all our newly-affordable high performance video and audio devices. It gets what online means, what online can do, and what a revitalised hyperspeed internet can do.
All the parts are here, it just needs a spark - and that's what Cisco is saying with its fibre to the desk switch. It's saying that if you build it, the geeks will come first, and where the geeks go, the world will follow.
It worked with the PC. It worked with wireless. It worked with online. It'll work for the next connected age. ®