Get email delivery of the Cadence blog featured here
Recently, Cadence held the fourth Photonics Summit and Workshop over two days. I attended the summit. The workshop was hands-on designing an RF photonics front-haul implementation. In fact, that was the focus of the whole summit, which was titled The Convergence of RF and Photonics. One reason that this combination is increasingly important is 5G.
RF and photonics is especially so in the US where the dominant 5G technology will be mmWave, at least for now, since the US military has so much of the mid-range spectrum that is primarily being used for deploying 5G in Asia and Europe. Of course, 4G spectrum can be reallocated to 5G, but since 4G is basically out of capacity already that is only a minor increase in spectral efficiency. So mmWave is the only way to expand capacity extensively. But mmWave goes about 2-300m in air, won't pass through walls, trees, or even hands. So there will need to be literally millions of small scale basestations. it is not practical to put an entire basestation with all the DSPs and processors on every light post, so the proposed solution is to put as little as possible in those basestations, not much more than the antenna, then feed the raw RF data to a much bigger basestation to handle the actual processing. The link from the small scale basestations to the bigger one is known as the fronthaul (analogous to the link from the bigger basestation to the internet routers, known as the backhaul). Since this is somewhat analogous to sending everything to a cloud data center for processing, this is known as Cloud-RAN or C-RAN. RAN stands for Radio Access Network and is a description of the radios and antenna that handle cellular connectivity (as opposed to things like Google Maps that operate over the RAN but are not considered part of it). The fronthaul is expected to be fiber, hence this will be a huge market for photonics in general, and RF in particular.
A caveat: Several of the summit presenters didn't give permission for their presentation to be videoed nor their slides published, so I won't be writing about any of them.
The opening keynote was by Intel's Yuliya Akulova. Her talk was titled Hybrid Laser Platform: The power of optics with the scalability of silicon.
Of course, Intel's biggest focus is the data center, since that is where they make a lot of their money. She opened with a bit of history and the graph above. On the bottom x-axis, it shows the speed per lane going from 10gbps to 200gbps. On the top x-axis, is the year the technology was (or is expected to be) deployed. The left y-axis shows the distance in terms of what is being linked, and the right y-axis shows the distance. The colors show the gradual shift from copper to multi-mode fiber, to single mode in the data center. The top green bar shows that single-mode fiber has been dominant for a long time once you reach the kilometer-scale between different data centers.
A limited factor in data centers is the faceplate space. The trend has been to go to more complex signaling (several bits per clock) and multiple lanes per connector. The next generation is targeted at 1.6Tbps using integrated optical I/O. Of course, the graph above has no y-axis, but you don't need me to tell you that the amount of data being transmitted is increasing fast. I'm sure you've seen lots of other graphs.
Here's Intel's plan for the 1.6Tbps with integrated optics. Note that the little red chips are the integrated switch chips giving 1.6Tbps, and in the middle is the switch chip delivering 25.6Tbps (16 times 1.6Tbps since there are 16 switch chips).
So her takeaway #1 was:
After a brief introduction to silicon photonics, Yuliya got to the real heart of her talk, Intel's hybrid laser platform. This is all done in Intel's fabs on 300mm equipment, so leveraging the scalability of all the semiconductor manufacturing infrastructure. The hybrid laser supports wafer-scale fabrication although with extra flexibility for optimization. They have multi-wavelengths and/or multi-channel arrays with low-loss coupling into SiPh waveguides. She talked about III-V lasers, but primarily I believe these are InP (indium phosphide).
One potential issue with IIi-V/Si wafer-bonded lasers is reliability. But she had the numbers and graphs. Here are a couple of datapoints:
There was a lot more product data. One interesting feature that she talked about was on-chip optical amplifiers. These have an on-chip input waveguide, the amplifier, and produce a much amplified optical signal (without turning it into electrical and back again).
Her wrap-up about the lasers was takeaway #2:
Finally she took a look at the tools for photonic design to get to takeaway #3 on the topic:
Her final summary:
Sign up for Sunday Brunch, the weekly Breakfast Bytes email.