Original URL: http://www.theregister.co.uk/2013/10/02/boffins_create_bulkprocess_onsilicon_optics/

Boffins create bulk-process on-silicon optics

Managing the on-chip power budget

By Richard Chirgwin

Posted in Hardware, 2nd October 2013 03:02 GMT

A group of researchers from MIT and the University of Boulder at Colorado say they've moved photonics a step closer to integration with both microprocessors and memory.

On-chip photonics offer a number of attractive prospects for chip-makers. Photonic communications generate less heat than electrons moving through copper, and photons following adjacent paths don't generate crosstalk.

Reducing the on-chip heat load is an important objective, because each transistor crammed onto a chip is a heat source. As transistors get smaller, the amount of heat generated in the tiny space of a microprocessor rises, and since you can't (yet) get rid of the transistors, shifting communications to the optical domain leaves a little more elbow room.

However, to make the whole thing affordable on a mass scale, the optics has to be created using the same manufacturing processes as are used to create the silicon components.

That's where the development from the CU / MIT team. Led by CU-Boulder researcher Milos Popovic, the group has created optical modulators that can be manufactured using familiar processes: the silicon-on-insulator CMOS techniques used in some microprocessor fabs, and the bulk CMOS processes used both in microprocessor and memory fabs.

“On top of the energy-efficiency and bandwidth-density advantages of silicon-photonics over electrical wires, photonics integrated into CMOS processes with no process changes provides enormous cost-benefits and advantage over traditional photonic systems,” explained Vladimir Stojanovic of MIT in a statement.

The SIO-CMOS process was able to deliver a modulator that could operate at 5 Gbps (abstract here), with energy consumption of 40 femto-Joules per bit, while the bulk-CMOS process, also running at 5 Gbps, has energy consumption of 160 femto-Joules per bit (abstract here).

The ultimate aim of the DARPA-funded project, also supported by Micron Technology, is to create a complete photonic processor and memory system. ®