As customers go, so go suppliers-and when those customers are the RBOCs, a whole lot of suppliers get going. Thus, while PON technology reached the field in North America long before BellSouth, SBC, and Verizon issued their joint RFP, the RBOC interest catalyzed the current excitement about fiber to the premises, node, and curb. Similarly, some PON systems houses have touted GPON for at least two years; nevertheless, it has taken the newly released RFP from the three RBOCs for pundits to declare GPON the next big wave in U.S. optical access.
Systems vendors have moved accordingly, announcing GPON offerings and rushing to meet the RBOCs’ information needs. One major focus of the RFI, as usual, is cost-does it make economic sense to go to GPON, particularly to its 2.5-Gbit/sec upstream/1.25-Gbit/sec downstream limits? Optical-network terminals (ONTs, alternatively called optical-network units, or ONUs) compose the majority of a PON’s systems hardware, and the optical components represent the largest item on an ONT’s bill of materials.
Component innovations therefore offer the most direct route toward lowering PON costs. Suppliers of optical and electronic components recognize the burden placed on them. However, changing market dynamics may make it difficult for component developers to choose the correct cost-reduction path.
Most ONTs require receivers to handle the incoming downstream traffic and lasers to generate the subsequent upstream signals. Component vendors usually integrate these functions into a single device. When you need only one of each, a component supplier will sell you a diplexer. But when you need a second receiver-for the reception of an RF broadcast video stream at 1550 nm-a triplexer most commonly fits the requirement.
Needless to say, a triplexer generally costs more than a diplexer. That’s one of the reasons IPTV holds such an appeal; the fact that you don’t need to use a separate receiver for video obviates the need for the more expensive optical component. So, with all the attention being paid to IPTV, are triplexers on the way out?
Not any time soon, say suppliers. For example, Chris Pfistner, director of global access business at component and transceiver supplier NeoPhotonics (San Jose), says most of his system customers have IPTV support on their roadmaps. “But when it comes down to reality, a lot of people are looking for an intermediate solution and want to have some sort of video overlay functionality,” he notes.
Teresa Engelhard, vice president of marketing and business development at Xponent Photonics (Monrovia, CA), says several of the customers for her company’s triplexers, diplexers, and optical subassemblies may make such an “interim” design more permanent. Systems houses tell her that RF broadcast may remain the best approach to maintaining the level of video service U.S. customers have come to expect from TV, including the ability to channel surf almost instantaneously. In this scenario, IPTV becomes the medium by which carriers will supply video on demand (VoD) or custom content. That means continued demand for triplexers.
What may change with the move to GPON are the building blocks from which triplexers are made. Almost all diplexers and triplexers today, regardless of the type of PON, use Fabry-Perot lasers and PIN photodiodes. “The interesting part there in terms of components is [with the move to higher-speed GPON] you’re getting to the point where Fabry-Perot lasers and PIN diodes might not have the necessary performance-so you’re right at the point where you might have to shift to DFB lasers and APD [avalanche-photodiode] receivers,” Pfistner cautions. Such a shift to higher-priced lasers and photodiodes would be the last thing cost-conscious systems developers would want to face. Nevertheless, Pfistner adds that Ethernet PON components may face the same questions if systems houses in that camp choose to extend today’s typical 10-km link lengths to the 20 km GPON will support.
Engelhard’s company makes just such an EPON diplexer. It includes a modified DFB that meets the application’s spectral width requirements but offers “cost savings” that she prefers not to detail. Engelhard believes that component suppliers and systems houses will end up sharing the burden of cost reduction, with the system developers assuming an increasing level of responsibility.
“We actually are seeing a reasonable number of requests directly from systems vendors who want to do a discrete design as a way to cost reduce their ONU,” she reports. Engelhard expects this trend to accelerate, and Xponent Photonics has responded with the development of reference designs that it gives away to its customers. The company also has created devices that can be mounted onto a board via the same kind of flex-circuit technology commonly used in the manufacture of such consumer electronics products as cell phones and personal digital assistants (PDAs). This process reduces manufacturing costs.
Conversely, Pfistner believes systems houses will move toward transceivers. “Quite a few systems houses today still buy the optical subassembly and build the transceivers themselves directly on the board,” he admits. However, “we’re clearly seeing a transition away from that to standard form factors, be it SFF or be it pluggable.”
Pfistner also says there is an interest in programmable devices, since the basic optics required for GPON and EPON transceivers do not differ significantly. Such devices would reduce spare counts and potentially simplify designs, particularly for systems houses that envision penetrating multiple markets with the same basic platform. “Of course, the other parameter in the equation is the tremendous cost pressure that we all know is on these optical access components,” Pfistner offers. “So at this point, the jury is still out about how much programmability you can afford in these devices.”
Meanwhile, merchant silicon vendors also have taken notice of the sudden increase in GPON interest. Previously, the PON chip space was small and well ordered. BroadLight (Mountain View, CA) and Freescale Semiconductor (Austin, TX) battled for board space within BPON systems, while Teknovus (Petaluma, CA) and relative newcomer Centillium Communications (Fremont, CA) chased Passave (Santa Clara, CA) for EPON (or, if they were focused on Japan, GEPON) business. But after a slow start, it now seems everyone wants a piece of GPON, with new entrants to the space popping up on a regular basis.
Most of the chip focus has centered on ONT devices. BroadLight offered the first announced GPON chipset, including the BL2000 system-on-chip (SoC) for ONTs. The BL2000, which also accommodates BPON, includes a G.984 MAC, 1.25- and 2.5-Gbit/sec SerDes and CDR circuitry, a packet processor, and software, among other functionality. Not to be outdone, Freescale announced a partnership with Alcatel through which the former’s communications and digital-signal-processor (DSP) expertise and the latter’s GPON skills would combine to create a new SoC of their own.
While Alcatel will be the obvious first customer of the new device, the two companies intend to make the SoC available to the general marketplace, says Suhail Agwani, PON product marketing manager at Freescale. Agwani declined to predict when the chip will reach general availability, although he says his company will issue a statement during this quarter regarding progress on the device, which is still in development.
The wait will prove worth it, Agwani says, because the Freescale/Alcatel device will feature an integrated DSP, which the BroadLight offering currently lacks BroadLight vice president and founder Didi Ivancovsky says the company left the DSP out of its device to maximize cost reduction and because changes in the marketplace toward VoIP will make DSPs less necessary in the future.
However, the duopoly BroadLight and Freescale have enjoyed in the Full Services Access Network (FSAN) end of the market is over. Conexant Systems (Newport Beach, CA) threw its hat into the ONT ring in July with the announcement of its Xenon SoC for BPON applications. While it doesn’t offer as many features as the BPON SoCs from its current competition, combining mainly a MAC/framer and packet processor, the device is already shipping to what Conexant product planning director Rajiv Bighe describes as “a major customer.” The device also represents a first step toward future SoCs, including a GPON/GEPON SoC that could be ready for sampling by the middle of next year.
AMCC (San Diego), meanwhile, has a GPON SoC of its own under development. Leveraging the success of its S2060 Gigabit Ethernet PHY, which has found its way into Asian-produced GEPON systems, the company expects to create an SoC that will work particularly well for business applications. “I think once you move into the business space, that might be the area that we think we can differentiate ourselves in with some of our network-processor functions that we have and some of our higher-ranked communications products,” says Neal Neslusan, director of marketing for transport products at AMCC. The company has not yet announced when the device will be ready, but according to Neslusan, AMCC will demonstrate the technology at upcoming “plug fests.”
Finally, to completely muddy once-clear waters, Passave announced last month that it has jumped the fence between the IEEE and FSAN worlds to offer GPON SoCs of its own. The company plans to use the GigaPASS architecture it created for GEPON for the new PAS5XXX OLT and PAS6XXX ONT chips. The ONT device will incorporate DSP, CDR, SerDes, packet processing, G.984 MAC functions, and a CPU. Both the ONT and OLT devices will support forward error correction.
And just to prove that integration has reached the formerly segregated PON chip space, Bighe says Conexant will take a break from its FSAN efforts long enough to introduce a GEPON SoC, which should begin sampling during this half of the year.
From these suddenly turbulent silicon waters several trends bubble to the surface. First, SoCs have become the coin of the chip realm, although the functions they incorporate will vary. Most of this integration attention will remain focused on ONT devices. According to BroadLight’s Ivancovsky, OLT developers don’t face the cost pressures their ONT compatriots experience. There can also be a wider variety of custom or discrete approaches to the OLT. For this reason, some OLT offerings are based on FPGAs. That said, Ivancovsky reports that a GPON OLT SoC is on the company roadmap; the AMCC sources say the same for their future. Passave, of course, announced an OLT device concurrently with the new ONT SoC.
The need for different approaches to business and residential GPONs offers another common thread. BroadLight has paired with Resolute Networks to pair the latter’s pseudo-wire circuit emulation services (CES) modules with BroadLight’s GPON devices. Together, the devices support one to four T1/E1 ports on the ONT side and circuit emulation aggregation on the OLT side at up to OC-3. As mentioned previously, the Freescale device will require additional DSPs, while AMCC will optimize its chip specifically for this application.
Chip developers do not yet agree, however, on whether a hybrid device that handles both FSAN and IEEE PONs offers an advantage. Bighe says that Conexant firmly believes in the hybrid approach, a philosophy that served them well in the DSL space. The company’s combined GPON/GEPON device may be followed by a BPON/EPON hybrid, depending on whether GPON wipes out the demand for BPON, he adds.
Wave7 Optics illustrated the potential demand for hybrid chips with the recent introduction of its Trident7 Universal OLT. However, not everyone is sure how much demand there would be for a combo chip. Dror Sa’lee, vice president of marketing at Passave, and Mark Abrams, the company’s director of marketing, point out that the global market remains fragmented, with BPON/GPON popular in North America and GEPON the current winner in Japan and Korea. Companies that hope to play globally may find a hybrid device interesting; those who wish to focus on only one market likely would not.
In fact, such a hybrid device could carry unused overhead for manufacturers focused on only one PON architecture. “I think manufacturers will fail if they start adding these features that the user is not using but is paying for,” surmises Freescale’s Agwani. “We’ve seen it fail before and I don’t see any reason why it wouldn’t again.”
Clearly, while both optical- and electronic-component vendors have listened to their customers’ requests for cost reduction, the question of whether systems houses will approach the market with specialized equipment or with platforms flexible enough to play in either GEPON or GPON markets remains unanswered in their minds. For optical-component vendors, that uncertainty forces them to wonder whether transceivers-pluggable, programmable, or otherwise flexible-deserve the most attention or whether they should focus on making it easier and cheaper for systems vendors to roll their own. On the chip side, the value of hybrid devices remains more a question of faith than fact.
It appears that both the technology and the market approach to GPON will have to mature further before cost-optimization can occur in earnest.