It's curtains for you, copper: IBM boffins push the LIGHT FANTASTIC
Use silicon photonics to continue chasing Moore's Law
IBM last month claimed a breakthrough in photonics – the practice of using light pulses rather than electrons to quickly send signals in chips.
Big Blue claimed its boffins had successfully designed and tested a fully integrated wavelength multiplexed silicon photonics chip which would be capable of turning out 100Gbps optical transceivers.
Commercial applications for photonics are obvious and include the usual suspects – cloud computing, Big Data and a new-breed of HPC data centres – but how far away are IBM and competing vendors from genuinely making silicon photonics chips a widespread commercial reality?
Photonics refers broadly to the application of light to perform practical functions. As far as we’re concerned, that really means moving information about more efficiently by using pulses of light.
All of us are familiar with some common optical telecoms and fibre optic data technologies. Some of us may even have worked with early semiconductor and laser diodes.
But the silicon photonics tech which has turned the head of IBM and others refers specifically to “speed-of-light” computing, built natively into chips to accelerate data transfer. The goal is to produce commercially viable optical links capable of connecting racks, boxes and individual chips together with low-power, single-die optical transceivers.
In July last year, IBM announced a $3bn research initiative to push the limits of chip technology. Twelve months on, IBM says its own flavour of silicon photonics chips will use four colours of light travelling within a single optical fibre to transmit data in and around any given computing system.
To illustrate the sea change it believes might be coming, IBM reckons new transceivers will be capable of sharing 63 million tweets or six million images in one second, or downloading a high-definition film in just two seconds.
The big difference between IBM’s latest statements and the lab research it has been conducting over the last decade is the promise of tangible product. Engineers have apparently now tested the company’s ‘CMOS Integrated Nano-Photonics Technology’ – fully-integrated wavelength multiplexed photonics on a single silicon chip, using sub-100nm semiconductor technology, paving the way for the manufacture of 100Gbps optical transceivers.
Replacing copper wiring with silicon photonics is a pragmatic way to use pulses of light to move large volumes of data at rapid speeds while overcoming the limitations and high-costs of traditional interconnects.
It promises to remove data bottlenecks inside of systems and between components, whether in the same enclosure or miles apart.
Silicon photonics doesn’t reinvent the wheel so much as develop smaller, shinier wheels, capable of going further and faster.
IBM doesn’t even claim to be pioneering optical interconnects within data centre environments. Recent announcements have been quick to acknowledge vertical cavity surface emitting laser (VCSEL) technologies already in play, which use multimode optical fibre to carry optical signals.
However IBM claims that demand for greater distance and data rates between ports in cloud and other applications are driving down costs for single-mode optical interconnect technologies, which hurdle the bandwidth and distance limitations of multimode links.