It's curtains for you, copper: IBM boffins push the LIGHT FANTASTIC
Use silicon photonics to continue chasing Moore's Law
Do the math
Metrics and benchmarks will ultimately convince more forward-thinking data centre managers than marketing alone. IBM claims its engineers in New York and Switzerland have demoed a “reference design” with a range of up to 2km.
It says this chip provides transmission and reception of high-speed data using four coloured lasers, each operating as an independent 25Gbps channel. The plan is for a transceiver that wavelength multiplexes these channels on-chip, to deliver the 100Gbps aggregate bandwidth over single-mode fibre.
There remain, however, some hurdles to overcome and more than a few doubts over the timelines for viable commercial adoption of silicon photonics. Back in February we reported that “quality issues” relating to performance at target temperature ranges had forced Intel to push back production runs of its own silicon photonic modules until 2016.
It decided to manufacture commercial parts on a 300nm production line, rather than the 200nm development production line.
Like IBM, Intel has been driving R&D for photonic chips for well over a decade, and had demonstrated 100Gbps modules at its own developer forum back in April 2013. Plans to ship optical processor and storage interconnects to early adopting HPC data centres at the back end of this year may be temporarily on hold, but Intel strongly believes that its photonic pipeline is an inevitable evolution from copper.
As Intel fellow Mario Paniccia insisted in his 2013 TedTALK: “It will allow us to keep up with Moore’s Law and we will not be limited by internal network speeds.”
Silicon photonics certainly seems like an ideal step for the modern data centre to make. If proved capable and reliable in real-world environments, then higher-speeds, extended reach, more efficient heat dissipation and smaller form factor would all be compelling improvements over copper for HPC applications.
Last year, the Yole Development group published its Silicon Photonics 2014 report, in which it forecast that the silicon photonics market would grow at a CAGR of 38 per cent, from $25m in 2013 to $700m in 2024. Those figures already sound extremely cautious.
Commercially, the technology may be embryonic, but in the labs of IBM and Intel it is regarded as a robust and necessary solution that will become a reality – and soon. Data centre limits of density, power, heat and weight will surely force the issue.
Elsewhere, Ovum’s Daryl Innis believes that despite the excitement generated by silicon photonics over the past few years, it continues to be criticised because it is taking a long time to commercialise.
It has also not been clear how it will deliver lower costs – whereas he feels IBM’s design “attacks these issues head-on”.
IBM reckons its CMOS Integrated Nano-Photonics Technology will finally deliver a “cost-effective” silicon photonics answer, using standard fabrication processes at an existing silicon chip foundry.
We’ll have to see, though, whether IBM’s development was just another step on along the development path or whether things really are nearing the commercial phase for photonics. ®