LightTrends

 

Embedded Optical Modules -- Not Your Grandfather's Transceiver

(Download Chinese Translation)

Transceiver and semiconductor manufacturers' interest has suddenly spiked in regards to selling optical transceivers inside complex systems to replace the over ten year old SNAP12 transmitters/receivers. The Avago microPOD, shown at various trade shows and used in IBM's Blue Waters HPC, has garnered a lot of attention from both transceiver suppliers and systems designers. As large semiconductor ASICs move to 25GHz, the length that electrical signals can travel across traditional FR4 printed circuit boards to reach other components is shrinking to about 6-inches. In May, 2012, the Optical Internetworking Forum (OIF) formed several working groups to define ASIC-to-optical interfaces for the next generation of data rates at 56 Gbps. Reaches are defined at 10 mm and 50 mm on printed circuit boards.

The front panel interconnect linking together servers, switches, and routers with copper wires hit the optical equivalent of the sound barrier at 10Gbps, where the cable reach decreased and costs increased. Nothing illustrates this situation better than what has happened with Ethernet and the ubiquitous 1GbE, Cat5e twisted pair cable. Transitioning from 1GBASE-T to 10GBASE-T has cost venture capitalists $600 million with more than 10 startups and nearly 8 years later the chant still is: "Wait until we get to 28-nm CMOS in 2014 and it will be so great!" A 1G Ethernet NIC has a street price of $30 where as a 10GBase-T board is about $500. Similarly, Direct Attach Copper (DAC) manufacturers are finding that with each data rate increase, the reach decreases: 7 meters at 10G slips to 3 meters at 14G (FDR) and it will need active signal conditions chips in each cable end to reach 3-5 meters at 25G.

While 10Gbps was the sound barrier at the front panel, 25Gbps represents the sound barrier inside the system on the printed circuit board. 25G signals travel only about 6 inches before losing most of their integrity, requiring an ever escalating number of compensating electronic chips to preserve, reconstruct and retransmit the signals. Retimers, clock/data recovery (CDR), pre-emphasis, and even DSPs are needed to move electrical signals across a circuit board at reasonable distances. The PCB design also changes and requires advanced materials such as Megtron at 200%-500% higher cost than traditional FR4 fiberglass.

High-speed circuit boards are becoming more like analog RF circuits rather than traditional digital electronics and the complexities are escalating exponentially. The engineering skill set needed to design at 25Gs is in the "black art" realm of analog design where the number of engineers and CAE/CAD tools are few. Of course, the design can be accomplished with complex electronics, but at what part count, cost and level of power consumption? Everyone is thinking, "What happens at the next step at 40G-56G? Will there even be VCSELs at 40G-56G or will the industry have to move to silicon photonics and WDM?" At the same time optical technologies are dropping in price rapidly and forming a cross over point.

Today, the industry is at an inflection point, and EOMs, having started out in the late 1990s, with the SNAP12 and POP4 transmitters and receivers, are destined now to be a key design component in high-speed electronic systems. Recent tradeshows and conferences have shown what unlimited R&D budgets for government-sponsored supercomputers can produce with IBM's Blue Waters supercomputers. Server blades on display showed optical interconnects (instead of copper wires) at nearly every interface. Avago microPODs took center stage and competitors and system builders took notice.

Embedded Optical Modules (EOMs) enable placement of optical transceivers next to big, high-speed CMOS chips running at 25GHz in the middle of large circuit boards. Over only a few millimeters high-speed signals can move into the optical domain where reaches can easily span 20-30 meters with no EMI interference, signal loss and lower latency. While high-speed system design always make components move closer together and concentrate heat, once in the optical domain, the components can be located...in the next ZIP code! This changes everything in high-speed system design. Design engineers are taking a close look at using EOMs to address difficult design problems and they are going shopping! This presents transceiver manufacturers with an entirely new market segment opportunity for long-term growth.

However, moving inside the system chassis in not the same technical nor business environment as outside on the front panel, like with traditional pluggable transceivers. The market dynamics and product requirements change significantly, including everything from thermal issues, packaging and connectors, to rigorous vendor and product qualification cycles, which may take over a year to complete! EOMs are more like core chip components in a critical design area than an easily swappable, pluggable transceiver.

To complicate matters more, there are no industry standard, no MSAs, no IEEE committees, no standard packaging, no standard connectors, line rates, configuration... In fact there isn't anything yet defined or standardized! Although, great for innovation, this makes for complicated business decisions for executives and designers. There are very few competing products available today, however, LightCounting found a large number of prototypes and trade show demos where suppliers are testing buyer requirements. This area is wide open to the best technical solutions with few of the traditional MSA restrictions transceiver companies are confined to. Advanced technologies such as silicon photonics and new VCSELs/TIA designs are also being proposed. Also, being a closed photonic, non-interoperable environment, like AOCs, EOMs have distinct product advantages.

Our analysis shows that the business will continue to develop in the HPC area and ripple down to the large, 100G-400G core switches and routers. As many of the open issues are sorted out and prices drop, we forecast the adoption all the way to server boards in data center and eventually to high-speed video systems, HDTVs, PCs, tablets, and smart phones. But the trip will not be an easy one and it is filled with many constraints for optical suppliers -- and CMOS electronics is not giving up easily!

The commercial market is poised to grow from $ 3 million and 40,000 units in 2012 to $39 million and 685,000 units by 2017. Early on a significant portion of the business will be with custom designed EOMs and later develop into the traditional multiple supplier and industry standards transceivers business. A separate forecast for custom-designed EOMs serving HPCs and core switches and routers is forecasted at $11 million in 2012, peaks at $29 million in 2014-2015 and drops off as silicon photonics and optics embedded inside chips is deployed. 25G/28G EOMs will be the big winning product and expected to last a while in the market as most data center protocols converge on the 25G area. CMOS compensating electronics will win designs below 25G and silicon photonics and embedded chip optics at 40G-56G where VCSELs may be not even viable by 2017.

The EOM report describes in great detail the market opportunity, timing, market dynamics and industry challenges. Five applications are analyzed along with the expected product requirements listed for each application segment. The market's trajectory starts with high-performance computers, and moves all the way through the data center switch, server and storage infrastructures to commercial video and��consumer products in HDTVs' tablets and smartphones. EOM uses are profiled that range from mid-board, to mezzanine board collocated with large chips to on-chip EOMs and glass and polymer optical backplanes. Five year forecasts are presented by unit shipments, average selling prices, and revenues with details on how the application opportunities, data rates and expected product requirements will evolve over time. Protocol roadmaps and other key timing markers are presented. EOM products and companies involved are profiled. For more information see www.LightCounting.com/EOM.cfm

By Brad Smith, VP and Chief Analyst, LightCounting LLC.

The LightCounting team will be at the following industry events:

  • IEEE Optical Interconnects, Eldorado Conference Center, Santa Fe, May 20-23, 2012
  • Optinet China, Presidential Plaza Hotel, Beijing, May 30-31, 2012

To set up a briefing with one of our analysts at these industry events, please contact Renee Isley, (Renee@LightCoutning.com).

About LightCounting:
LightCounting, LLC. Is a leading optical communications market research company, offering semi-annual market update, forecast and state of the industry reports based on analysis of publicly available information and confidential data provided by more 20 leading module and component vendors. LightCounting is the optical communications market's source for accurate, detailed and relevant information necessary for doing business in today's highly competitive market environment. Privately held, LightCounting is headquartered in Eugene, Oregon. For more information, go to: www.LightCounting.com or follow us on Twitter at: www.twitter.com/lightcounting.