Saturday, January 26, 2008

At ISSCC: No magicians for 10GE

10G Ethernet needs some pushing with the costs still relatively high for optical versions and the power consumption still too high for copper. That’s supposed to start shifting in 2008, and the upcoming ISSCC conference in San Francisco may be the place to get the first glimpse of the prospects.

Based on the sketchy abstracts I have on hand, it’s hard to tell what the horizon looks like. Startup Teranetics will describe a 130nm chip that drives 10G Ethernet over 100 meters of UTP, but it burns a fairly hefty 10.5W doing it. And though made in a mainstream 130nm process, it still requires two fairly large blocks--a 55mm-squared analog front end and a 150mm-squared digital processor. What’s more the abstract does not say which flavor of UTP they are using, probably a pricey Cat7.

Startup Aeluros will describe a protocol-independent serial 10G transmitter using a three-tap filter for use in optical cables or backplanes that eats just 165 mW. But it doesn’t sound like it’s for copper cables and there is no word about a full transceiver.

Well, there are a lot of really smart engineers behind these ISSCC papers, but they are engineers after all—not magicians.

2 comments:

Anonymous said...

First of all, IB cables and 10GE CX4 cables are NOT the same. Please look at:
http://www.gore.com/en_xx/products/cables/copper/networking/cx4/cx4_faqs.html#difference

In addition to Base-T requiring black magic to solve the power/size issue, QSFP is expected to address the cost issue of fiber. So far, it delivers.

Steve said...

It's true that first generation 10GBASE-T devices consume > 10W, but these have been sampling since August 2006!

By looking at silicon cycles, you should have a good indicator of the geometries to expect in 2008 for a 2nd generation 10GBASE-T device.

Further, since a large portion of the device is taken up by digital logic, significant power reduction will be achieved (order x2).

RSN.

 
interconnects