Peering Through the Confusion Over IBM's 160 Gbps Optical Transceiver

After IBM released its initial teaser yesterday touting its demonstration of an optical transceiver with a theoretical 160 gigabit per second (Gbps) throughput speed this upcoming Thursday, press sources seized upon this phrase: "fast enough to reduce the download time for a typical high definition feature-length film to a single second compared to 30 minutes or more."

That may be a tall order, given the fact that the prototype in question is a transceiver and not an endpoint. A transceiver is a device that bridges between layers of the Internet, such as between the physical layer (Layer 1 in the OSI model, the "backbone") and the data link layer (2). So we're not talking about something that goes into your PC, despite what you may have heard on local TV news.

Optical networks are typically synchronous, which means their throughput rate is set according to a clock. There are such things as passive optical data networks, whose technology is being championed by companies such as Hitachi, though as recently as the last CES 2007, Hitachi was talking about Gigabit Passive Optical Network speed as just being able to break the 1 Gbps barrier.

The two standards for high-speed synchronous data transmission over optical fiber are SONET (9.953 Gbps) and IEEE 802.3ae ("10-gig Ethernet," at 10.312 Gbps). But what IBM will introduce on Thursday is not a new optical data network, but instead a single transceiver chip using a typical CMOS structure though atypical materials such as indium phosphide (InP) and gallium arsenide (GaAs). Assuming the network existed for it, IBM's new transceiver could achieve a throughput rate of possibly as much as 16 times the rates of transceivers currently produced by Intel and Fujitsu, the company claims.

So what will be very, very interesting to see is how IBM is able to demonstrably measure throughput speeds without effectively reinventing the optical fiber backbone in so doing. Not that this is completely unfeasible; in fact, if IBM researchers are prepared to demonstrate a 16x advance in optical fiber throughput, then this would be a tremendously more intriguing demonstration than even the little CMOS chipset.

But what will be very, very...very interesting to see is how IBM would be able to sell transceivers based on this technology to telecom clients, given the fact that the company sold its optical transceiver division to JDS Uniphase five years ago. Perhaps a moratorium on IBM's rights to sell transceivers has elapsed, and that would be nice to know.

Yet the fact that no one seems to have asked these questions before settling upon the conclusions offered in the second paragraph of IBM's press release, is more than a bit disturbing. IBM may not be in a position to answer these questions until Thursday at the earliest, but even revealing the nature of the outstanding questions would have shed some light on the real issues IBM researchers are working to address.

At any rate, on Thursday, IBM will reveal whether it truly is prepared to redefine the entire infrastructure of optical networking, or whether someone's finger accidentally hit one too many zeroes on the calculator.