26 March 2012
Can steerable antennas solve supercomputing challenges?
Supercomputers keep getting larger as more computational challenges are addressed. And the larger the supercomputer, the more power it needs: today's goliaths draw close to 10MW.
If the current approach is continued, the power requirements will become unfeasible.
Part of the problem is pushing data around all the different parts of the beast - at all levels from boards upwards - so the right data is at the right place at the right time.
What if a different communications method could be used? A European project is looking to employ direct rf communications using steerable antennas at teraHertz frequencies. There would be no need for switching fabrics and, according to the project, the approach will lend itself to 3d chips, with the interposer acting as the antenna.
But can the project realise its goal of creating, within its 12 year lifetime, a supercomputer with 1billion cores that fits in a volume of 1litre? And if it can, what might such a device enable?