Road to 100Gb/sec…Innovation Required! (Part 3 of 3)

Interconnect, Silicon Photonics

Physical Layer Innovation: Silicon Photonics

So in two previous posts, I discussed the innovations required at the transport, network, and link layer of the communications protocol stack to take advantage of 100Gb/s networks . Let’s now talk about the physical layer. A 100Gb/sec signaling rate implies a 10ps symbol period.

Frankly, this is just not possible on a commercial basis with current technology. Neither is it possible on copper nor on optical interfaces. At this rate the electrical and optical pules just can’t travel any useful distance without smearing into each other and getting corrupted.

So there are two possible solutions.  The first is to use 4 parallel connections each running @25Gb/sec. The second is to use a single channel with a 25Gb/sec symbol rate but to send four bits per symbol. This can be done either through techniques like Pulse Amplitude Modulation (PAM4) or optically by sending four different colors of light on a single fiber using Wavelength Division Multiplexing (WDM) techniques.

It turns out sending four separate streams of data each @25Gb/s is the easiest and most cost effective way to transmit 100Gb/sec for distances less than about 1km links. Let’s look at the different ways you can send four parallel optical signals at a 25Gb/s rate.

There are two different optical technologies that can support 100Gb/sec connections. The first is direct modulation of a Vertical Cavity Surface Emitting Layer or VCSEL. These lasers operate at 850nm wavelength and use multi-mode fiber (MMF). There are four lasers required and each of them is turned on and off at a 25Gb/sec symbol rate which means a 40ps period. This is very challenging to modulate the lasers at this rate and requires switching high currents as a lot of energy needs to be pumped into and out of the lasers.

Kevin D 082514 Fig4


Because of the high currents involved, the devices tend to heat up and it’s challenging to keep the optical energy constant and to build optical cables that will operate reliably for many years. We’ve been able to make VCSEL based optical interconnects work at 100Gb/sec but it is very challenging and we are probably about at the end of the road for VCSEL technology. We don’t see a clear way to reliably modulate the lasers much faster. Moreover, the inter-symbol-interference of the multi-mode fiber also severely limits the symbol rates and the distance within the data center that VCSEL based optical links can support.

There is an alternative technology based on silicon photonics that we think has a much longer technology runway. Silicon photonics can use a single laser that is not modulated at all.  Instead of banging four lasers on and off there is just one laser and it is always on.

This means we can use much lower performance and much lower cost lasers such as Fabry-Perot  lasers operating at 1550nm. This allows using single mode fiber (SMF) which is cheaper and has better optical properties, supports faster signaling rates, and can span much longer distances than MMF. All in all, silicon photonics gives much better performance and distance and a longer runway to continue to driver higher optical bandwidth and thus increase the signaling rate.

But how do you send four 25Gb/sec optical information streams if you have one laser that is on all of the time? This is where the silicon photonics comes in!  Here is a picture of our silicon photonics devices.


Kevin D 082514 Fig5

On the left is the electro-optical transmitter. The low cost laser is flip-chip bonded to the surface of the silicon photonic and the light is passively coupled into the surface of the silicon photonics chip. You have a continuous wave that travels through a passive wave guide on the chip. Using waveguides, this light source is passively split into two and then four separate beams of light travelling in their own waveguides.

Each of these light paths then passes through what is called a Franz-Keldysh optical absorption modulator, which basically acts as an electrically controlled optical photon shutter. It absorbs photons depending on the electrical signal that is applied. Instead of turning four lasers on and off very quickly, we just passively split a continuous wave light source into four beams and then have a photonic shutter to turn the light pulse on and off.

You can think of this as being similar to the old Morse code signaling used by ships in World War II. You don’t plug and unplug the light source, but rather you have a shutter with which you could let the light pass or not to turn the light on and off. Silicon photonics is much more efficient and provides much more runway to continue to improve performance.

The receiver is much simpler. The four light signals are received and waveguides send the modulated optical signals to conventional photo detectors. In the image above, the diagram on on the far right shows the actual electrical and optical eye-diagrams of the four modulated signals after 250m of optical fiber. The top three are electrical eye-diagrams and come after the clock and data recovery and retiming and thus are ultra-clean. The bottom waveform is actually the received optical signal before the electrical retiming and thus has slightly more jitter is visible – but still super clean with a huge eye opening.

So, how do we take advantage of the VCSEL and silicon photonics based technology and deliver to customers in a useful form factor? It turns out it is precisely the same form factor for both technologies: the Quad Small Form-factor Pluggable or QSFP module. This is a module that plugs into a receptacle on a switch or adapter port. It is flexible as you can plug in a copper module for very short reach connections (1-2m), a VCSEL based module for multi-mode fiber for medium length connections (up to about 100m), or a Silicon Photonics based module that gives both long span (up to 2km) and also means that the single-mode fiber plant won’t need to be replaced when you upgrade to even faster signaling rates.



Part 1 of 3 – Transport Layer Innovation: RDMA

Part 2 of 3 – Network and Link Layer Innovation: Lossless Networks


About Kevin Deierling

Kevin Deierling has served as Mellanox's VP of marketing since March 2013. Previously he served as VP of technology at Genia Technologies, chief architect at Silver Spring Networks and ran marketing and business development at Spans Logic. Kevin has contributed to multiple technology standards and has over 25 patents in areas including wireless communications, error correction, security, video compression, and DNA sequencing. He is a contributing author of a text on BiCmos design. Kevin holds a BA in Solid State Physics from UC Berkeley. Follow Kevin on Twitter: @TechseerKD

Comments are closed.