Comms chips

Inphi Pitches OC768

Indium phosphide technology will reign supreme in the coming generation of optical networks, when they hit speeds of OC768 (40 Gbit/s). So says Inphi Corp., a startup that came out of stealth mode this week (see Inphi Targets 40-Gig Chips).

Inphi is designing and developing physical-layer communications chips for OC768 networks, which require serial outputs running at full line rate. At such fast speeds, silicon technology is left panting by the roadside, and new materials must be called on.

Despite the startup's relative youth -- it was founded in November 2000 -- it appears to be furlongs ahead of its competitors. It has already unveiled prototype 4-to-1 multiplexer and 1-to-4 demultiplexer chips, which are expected to start sampling before the end of the year. It also plans to ship two further products in the same time frame.

In contrast, Vitesse Semiconductor Corp. (Nasdaq: VTSS), which has been working on indium phosphide technology for longer, has yet to unveil any products. So far it has only made a technology announcement about the development of a VCO (voltage controlled oscillator), which is a building block of a timing circuit, which is in turn a building block of a complete chip (see Vitesse Touts InP Technology).

There are a couple of startups developing indium phosphide chips including Gtran Inc. and Velocium, which has partnered with Hitachi components subsidiary OpNext Inc., but they also appear to be behind Inphi in their developments (see Startup Develops Fast InP Process and OpNext, Velocium Team on OC768 ).

Some chip makers, such as Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC), have exploited an alternative material, silicon germanium (SiGe), in order to get OC768 chips to market more quickly. But Loi Nguyen, Inphi's founder and CEO, is dismissive of SiGe.

"[SiGe] burns too much power; it burns about three times more power than indium phosphide," he notes. Inphi's multiplexer and demultiplexer chips consume less than 900 mW of power, he claims.

There's a second reason that he believes SiGe won't make it. Although SiGe chips can squeak by at 40 Gbit/s, they will really struggle to reach speeds of 50 Gbit/s, he says.

Yes, strange though it may seem, 40 Gbit/s networks will actually need chips that run at 50 Gbit/s. That's because OC768 is likely to make extensive use of forward error correction (FEC), which allows the data to go farther between regeneration points (see beginner's guide on Digital Wrappers and Forward Error Correction (FEC)). Standard FEC runs at 43 Gbit/s, while "superFEC" has an overhead of 25 percent.

Inphi has demonstrated the first commercial chips that operate at speeds of 50 Gbit/s, according to Nguyen.

So how did the startup accomplish so much in so little time? It's all down to the experience of the technical team, says Nguyen. Both he and CTO Gopal Raghavan have tens of years experience in designing high-speed indium phosphide circuits for the aerospace industry, at HRL Laboratories.

Another factor that has allowed the company to move fast is the fact that it’s a fabless outfit, so it does not have to commission expensive equipment lines and refine manufacturing processes. Instead it's using the indium phosphide process developed by Global Communications Solutions Inc. (GCS), which, interestingly, is also the foundry used by Gtran (see GCS Does InP HBT R&D). Vitesse, on the other hand, has worked from scratch and uses its own foundry (see Vitesse an Indium Giver).

Inphi was bankrolled by Mayfield Fund and Tallwood Venture Capital. It raised $12 million in round A and is looking to raise a second round early next year.

— Pauline Rigby, Senior Editor, Light Reading
flanker 12/4/2012 | 7:43:28 PM
re: Inphi Pitches OC768
OC 192 port prices are dropping like a rock, and DWDM channel count is maxxing at 320. So where is the business model for OC 768/STM 256?
iptwister 12/4/2012 | 7:43:22 PM
re: Inphi Pitches OC768 Your point that prices are dropping on OC-192 is reasonable.

Here is my perception: OC-192 pricing is dropping due to a temporary supply/demand imbalance in inventory. Capital Spending is way down (demand) and therefore supply of OC-192 equipment is way up (supply).

In general, I believe strongly in OC-768/40gig, because it is more efficient.

In the long run, and the medium run, companies will want the lower operating cost of denser hardware.

Certainly there is a capital cost involved in upgrading to OC-768, so this must be borne by the cash-strapped telcos, which they may not do now.

But, having been an exec at a telecommunications company, I can tell you that one-time capital expense costs are generally moderately trivial relative to the value of lower unit-costs. This was, however, before the capital markets collapsed :-)

In summary, I believe OC-768 is viable in 1Q2003 because of the lowered cost of operations that it seems to offer.

Sign In