In Search of Optical Networking's 'Holy Grail'
As I basked in the afterglow of last month's Optical Fiber Communications (OFC) 2019 and reviewed my notes, one phrase came up a number of times that I don't think I've heard with such frequency in the past: Holy Grail.
It seems that the optical vendors, and perhaps systems vendors as well, are looking for the solution that will give it eternal life in the face of one of the most dynamic industry environments the industry has ever seen.
Business models are changing, and technologies both relatively "old" (coherent) and "new" (silicon photonics) are rising in importance. At the same time, it appears the optics industry is on the cusp of a major inflection point. Microsoft's Brad Booth noted during his presentation on the "Beyond 400G" panel that optics is on the verge of being adopted "all the way down to the server ... I'm talking an order of magnitude more optical connections."
Similarly, Cisco's Mark Nowell, citing LightCounting data, noted that in the next five years, the 400G Ethernet Market is expected to be 20x that of 100G. I can't help but visualize the scene in Indiana Jones and the Last Crusade when Indiana Jones enters the room full of potential grails guarded by a Knight Templar. Choose the right path, and ride the wave of broadening optical adoption to everlasting life. Choosing the wrong one and you age rapidly into oblivion, turning into a wrinkly, dusty corpse. (Instead of the Knight Templar guarding the grail room, I'd envision Arista's Andy Bechtolsheim -- he seems to be continuously reborn in the networking industry.)
It's also not hard to see why the industry may even be more “thirsty” to find the grail than in years past. Vladimir Kozlov of market research firm LightCounting shared data on one panel that indicated that for the first time "in many years" the optical transceiver industry declined in 2018, despite strong volume demand and robust spending by hyperscale players.
Kozlov attributed the decline in part to collapsing CWDM4 pricing and a lack of new products that typically carry higher ASPs. Optical transceiver vendors woke up to 2019 a bit hungover and parched. Making that headache throb more was the lingering effect of the ZTE ban and ongoing "trade war" with China.
So what is the Holy Grail for the optics industry? Arlon Martin, senior director of marketing at Samtec, stated that, to the Ethernet Alliance, putting 100G in an SFP transceiver used to be "the grail," because this implied that 100G electrical lanes and 100G optical lanes had been achieved. While 100G in an SFP hasn't yet been delivered, vendors noted that the key attributes that of "grail" – 100G optical and electrical lanes – are on the cusp of high-volume ramps. We heard San Jose, Calif.-based Credo say the major technical challenges around the 100G SerDes were solved, and we heard multiple vendors, including Cisco and AOI, say that 100G optical lanes were ready for prime time. Knight Templar Bechtolsheim stated that he expects first 100GSERDES switch silicon in 2020.
Silicon photonics and vertical integration
Juniper, however, claimed to have found a different Holy Grail. In a session with investors and analysts, Juniper CTO Bikash Koley presented a slide which boldly stated, "Juniper Achieves the Holy Grail," which pictured a glamour shot of a polished wafer populated with transceiver chips. Koley boasted to analysts and investors about Juniper's differentiated silicon photonics technology that initially came to the company via its acquisition of Aurrion in 2016.
Juniper said it has "matured" the technology and noted that only it and Intel have the transmitter technology that doesn't need discrete lasers (both bond a small amount of indium phosphide on to a silicon cavity). Further, Juniper said that, unlike Intel, it has integrated both transmit and receive functions on the same chip, which further allows them to test all transceivers at "wafer scale" to make sure they all work before dicing up the wafer. Juniper will be shipping its silicon photonics-based transceivers by the end of Q2; I confirmed that Juniper's first offering will be LR-4, not CWDM4, for now.
A related topic is "embedded" (either onboard or co-packaged) optics, something Koley stated, "It's not that far out in the future, a couple more years." It would seem that Intel, Cisco and Juniper are in a Holy Grail "boat race," and they all appear to be drinking a brew with varying concentrations of silicon photonics, vertical integration and outsourced manufacturing. Embedded optics seems to be the end game that all of these companies are working towards.
Bill Gartner, VP & GM of Cisco's optical business, acknowledged during an OFC media lunch that Luxtera's longer-term roadmap motivated his company's recent acquisition of the silicon-photonics player. Cisco was also after Luxtera's embedded optics expertise and its "wafer scale approach to automated the manufacturing process" and "delivering on the on the promise of silicon photonics better than anyone else," Gartner said.
Meanwhile, Broadcom appears to be stumbling around, unable to decide which chalice to drink out of, or whether to drink from everybody else's.
On the "Beyond 400G" panel, Claire Szu Ma, senior manager of AOI Taiwan, noted that the co-packaging of optics presents some challenges, such as not offering a "pay-as-you-go" for the end customers that modules current offer (as systems would come fully populated with optics.) Additionally, she noted the implications of onboard or co-packaged optics means that transceiver vendors "will possibly become component vendors" (i.e., sell lasers or engines), "or a switch vendor will be more like a transceiver vendor."
Ma's observation is no small, offhand comment for a vertically integrated module vendor like Applied Optoelectronics. Seeming to confirm this trend, Lumentum made a significant announcement earlier in the week of the show, that it was divesting its datacom transceiver business to Cambridge Industries Group (CIG).
Meanwhile, II-VI is off in the corner, pouring some Finisar-branded datacom transceiver "spirits" into a beer bong; investors must have thought II-VI was drinking the wrong water, with the stock selling off 19% on the day the deal was announced. However, II-VI has a history of successfully selling components (and a strong track record of deploying capital with good returns), so time will tell from which cup the company is drinking.
While the most significant degree of disruption appears to be in the data center and short reach applications, vendors in the DCI, metro and long-haul markets have big strategic choices to make as well. The advent of 400G ZR pits component and IC vendors like Inphi and Acacia against optical systems like Ciena, Huawei and Infinera. Cignal AI's Andrew Schmitt predicted that by 2022 or 2023, the "modularization" of coherent technology would eventually begin to cannibalize the optical hardware market.
It appears that no matter where one sits in the optical ecosystem, much change is afoot. So, unfortunately, when it comes time to pick one's grail, vendors will have to choose wisely. As Professor Henry Jones warned in the movie, "In this sort of race, there's no silver medal for finishing second."
— James Kisner, Consulting Analyst, LightCounting