Briefing: Silicon Photonics

Part 2: A system vendor's perspective

• Silicon photonics as a technology has its challenges
• Its biggest impact could be to shake up the industry's optical component supply chain
• Silicon photonics will not displace VCSELs

An interview with Alcatel-Lucent Bell Labs' Martin Zirngibl, domain leader for enabling physical technologies, on the merits and potential impact of silicon photonics

Martin Zirngibl admits he is skeptical when it comes to silicon photonics. "There is a lot of hype around silicon photonics but there are also some real advantages," he says. "We have a strong silicon photonics programme inside Bell Labs and I tell my folks: If you prove me wrong, I'm going to be very happy."

The skepticism stems from the technology's limitations. "There is no Moore's Law in photonics, you cannot cascade many photonic elements," says Zirngibl.

Photonic components are also analogue. Once several devices are cascaded, the signal loss accumulates. This is true for photonic integration in general, not just silicon photonics.

Another issue is that the size of an optical component such as a laser or a modulator is dictated by the laws of physics rather than lithography, used to make ever-smaller transistors with each generation of CMOS process. Zirngibl compares optical transmitters and receivers to cars: they improve with time but the fundamental size does not change.  

"Silicon photonics could form an ASIC-like model and break the supply chain" 

A consequence of shrinking feature size with semiconductors is that chip performance gets better with integration. Integration in photonics, in contrast, involves compromise and a tradeoff in optical performance.

However, the advantages of silicon photonics are significant. The technology can benefit from the huge investment made in the semiconductor industry. "CMOS foundries exists with 8- and 12-inch wafers," says Zirngibl. These mature processes are extremely well controlled, producing high-yielding devices. "If you match any component with that type of process, you have instant high volume and instant scalability," says Zirngibl.

Silicon photonics may require something different but if it can use these CMOS processes, the result is a free ride on all this investment, he says: "That is the real advantage.”

For Zirngibl, the impact of silicon photonics will more likely be on the industry supply chain. An optical component maker may sell its device to a packaging company that puts it in a transmitter or receiver optical sub-assembly (TOSA/ROSA). In turn, the sub-assemblies are sold to a module company which then sell the optical transceiver to an equipment vendor. Each player in the supply chain adds its own profit.

Silicon photonics promises to break the model. A system company can design its own chip and go to a silicon foundry. It could then go to a packaging company to make the module or package the device directly on a card, bypassing the module maker altogether.    

"Silicon photonics could form an ASIC-like model and break the supply chain," says Zirngibl. "This worries the large module makers of the world."

"The problem with coherent is that it needs a lot of optical stuff"

Zirngibl stresses that such a change could also happen with traditional optical components. A system vendor could adopt a similar strategy with indium phosphide chips, for example. But the issue is that indium phosphide does not share the mature processes or the scale of the semiconductor industry, and as such an ASIC model is harder to achieve.

"If you can use CMOS processes for optical components then, all of a sudden, optical could become an ASIC-like supply chain," says Zirngibl. "It could cut out a lot of the module and package vendors."

That is what Cisco Systems has done with its CPAK module based on silicon photonics. "Cisco broke the supply chain model by doing an internal development of a module, they don't rely on anyone else," he says.

Challenges

Silicon photonics faces several challenges. One is silicon has no optical source. “A regular CMOS process will not product a light source." Companies are pursuing several approaches as to how best to couple a III-V source to silicon.

Another issue is that the optical performance of a silicon photonics design must match that of alternative solutions. "At the end of the day in photonics it is always about performance," says Zirngibl.

A 1dB or 2dB worse insertion loss compared with an alternative photonic design may be acceptable but it has to roughly match. "If it does not, even if the device is for free, the fact that you have a performance degradation will make you pay somewhere else [in the system]," says Zirngibl.

"We once tried access; there is nothing more cost-sensitive than fibre-to-the-home (FTTH) and we wanted to push silicon photonics for access," says Zirngibl. FTTH is highly cost-sensitive and is a volume market. But the resulting design had a 5dB worse performance than a free space equivalent. "We didn't have the slightest chance to get in: a 5dB insertion loss in access means a split ratio of 1:16 instead of 1:32  and a 3-4km reach instead of 20km."

One application where optical performance is key is long-distance transmission using coherent technology. Coherent offers significant benefits: 100 Gigabit per channel, reaches of several thousand kilometers, spectral efficiency, and the ability to correct in the digital domain for many of the transmission impairments.

"The problem with coherent is that it needs a lot of optical stuff," says Zirngibl. A coherent line card has a high power consumption and uses lot of expensive optical components. Companies are looking at silicon photonics as a way of reducing cost while shrinking the size to fit within a pluggable transceiver. The tradeoff is reach; instead of a span of 1000km-plus, achieving a few hundred kilometers would be more likely.  

"For interconnect, VCSELs are not going to be displaced"

Companies such as Oclaro, and Finisar and u2t Photonics have announced developments involving indium phosphide to achieve a compact-enough design to fit within a CFP2 pluggable module.

"Silicon photonics has a modulator that can be driven with a low voltage, and that could be driven using CMOS, a real advantage," says Zirngibl. "Unfortunately, the modulator has a lot of insertion loss, so you have to solve it elsewhere."

At OFC/NFOEC 2013, Alcatel-Lucent, working with the CEA-Leti foundry, presented a long-distance laser design using silicon photonics. "We do wafer bonding on silicon - you marry indium phoshide with silicon photonics," says Zirngibl. "If you match a process that allows you to do a light source with 8-inch or 12-inch wafers, you have something that could be a winning solution."

Short-reach connections

One important question that impacts the potential silicon photonics opportunity is when does the crossover from electrical to optical occur?

If the link distance is sufficiently short, it makes sense to stay in the electrical domain. This is because going optical inevitable requires electrical-optical and optical-electrical conversions over a link. "If it is very short distance, it will always be electrical,” says Zirngibl. The issue with electrical is that as signal speeds increase to 25 Gig, losses accumulate very quickly with distance and the signal fades.

"We believe that this crossover from electrical to optical is 1 meter at 100Gbps," says Zirngibl, with the 100 Gigabit being four 25Gbps lanes.

Accordingly, for any distance above 1m, optical interconnect will be used for 100 Gig signals between boards and between systems. “The electrical I/O goes to the end of board where you have a VCSEL interconnect and goes to the next line card, where there is another VCSEL interconnect," says Zirngibl.

In such a design, getting the optics closer to the processor makes sense. "A good case for a processor with almost an optical I/O," says Zirngibl. Companies such as Arista Networks and Compass-EOS are already doing this. "The problem is that it is pretty ugly, cables coming out of the processor, and how do you slide in and out a card?" he says. "What would be really cool is a VCSEL and printed optical waveguides."

This is an area that still needs some work, he says, but there are companies developing optical PCBs such as Vario-optics.

Zirngibl believes one promising application for silicon photonics is for a coherent receiver at 100 Gig. "That is when you will see it [silicon photonics] first," he says. "There is demultiplexing, no light source is needed and you can do the detection on silicon photonics." 

For short-reach interconnect, Zirngibl believes silicon photonics will not displace VCSELs.

"VCSELs are by nature an incredibly efficient, low-cost solution," he concludes. "For interconnect, VCSELs are not going to be displaced."

Part 1: Optical interconnect, click here

Part 3: Is silicon photonics an industry game-changer? click here