Another chance for the optical shift

A little over ten years, David Miller of Stanford University argued that optical interconnects for electronic chips would ultimately become necessary:

“Optics is arguably a very interesting and different physical approach to interconnection that can in principle address most, if not all, of the problems encountered in electrical interconnections.”

Miller found problems with dense optical interconnects: for large numbers of connections, the power demand from optical was likely to be prohibitive for some time for on-chip interconnect. But as clock rates exceeded 1GHz, electrical connections would become more power-hungry and problematic. The problem for optical interconnect is that on-chip rates have not proceeded far enough to make them necessary. As clocks have largely maxed out at around 3GHz, the problems caused by skin effect and other issues of high-speed global interconnect have been pushed to one side.

Off-chip, however, interest in optical is picking up again. On-chip you can use parallelism. Off-chip you have to worry about how to maintain skew across bundles of high-speed signals, even using advanced serial protocols. 

Optical backplanes are not new. They had been proposed as a way to improve datarates without consuming huge amounts of power at least five years earlier than Miller’s paper.

At that time, the conventional parallel buses were hitting the wall in terms of performance. Copper over FR4 fibreglass was resulting heavily distorted signals if speeds were pushed too far. In practice, technology saved electrical transmission: a combination of digital predistortion and compensation techniques and the shift to bundled serial channels made it possible to push into the gigabit per second domain. 

The field programmable gate array (FPGA) maker Altera now reckons that the time has come to have another look at optical for data transmission inside electronic systems. Why? Because those electrical tricks are now themselves running out of steam.

An IBM paper from 2005 found that the power advantages of using optical were far from automatic, based on technologies available at the time such as 10Gb/s Ethernet. However, inter-module speeds headed from tens to hundreds of gigabits per second, the advantages of optical for backplane connections, which might have to cross up to 2m, began to win out because of the higher power demands of equivalent electrical connections.

The IBM researchers found that for snoopable connections, such as those between processors in a symmetric multiprocessing system, optical had some clear benefits over electrical signalling.

A lot will depend on how these early devices with integrated optical ports are implemented: the lasers and waveguides chosen will be crucial to determining how efficiently they perform. But Altera is confident it has the right transceiver technology to support it and reckons that optical will be needed to make the most of a new generation of 28Gb/s transceivers the company has designed.

Posted by Chris Edwards

The Low-Power Design Blog is sponsored by Mentor Graphics. The company has focused years of R&D on low-power design techniques and is glad to support a resource that highlights creative methods for reducing the power consumption of electronic systems.