University researchers in the U.K. will study how to make future microprocessors faster by replacing some of their internal connections with radio links.
Researchers at the University of Bath have received new funding to investigate how the emerging field of spintronics can be applied to in-chip communications, they announced Friday.
Electronic circuits control the flow of electrons from one place to another, but they only exploit one property of the electrons: their charge. Spintronics, or spin-based electronics, attempts to exploit another of the electrons' properties, their "spin." A quantum property, spin can be either "up" or "down". By measuring or modifying the spin, the property can be used to transmit, manipulate or store information.
Spintronic techniques are already used in some hard disk drives as a way to increase the density of stored information. Spin control also plays a role in the storage of information in MRAM magnetic memories, and in the manipulation of data in quantum computers, another emerging application of spintronics.
However, the researchers at Bath are interested in applying spintronics to the transmission of information, not its storage or manipulation.
The transistors used in today's microprocessors could run at speeds of up 100GHz if it weren't for the wires connecting them to one another, according to Alain Nogaret, a lecturer in the department of physics at Bath.
As processors run at higher and higher frequencies, wires present an obstacle to electrical signals, rather than an unobstructed path, so the signals quickly fade away, even over short distances: "The limit is not the transistors, but the losses in the electronic signals between transistors or clusters of transistors," Nogaret said.
"One way to cut these losses is to send these signals through microwaves," he said. In that way, the signal loss at 100GHz can be cut to just a couple of decibels per centimeter from 115 decibels per centimeter along a wire, he said.
Nogaret and his team hope to generate those microwaves by applying a theory he published in Physical Review Letters last year, entitled "Electrically induced Raman emission from planar spin oscillator," in which he predicted that radio signals are emitted when the spin of an electron trapped in a magnetic field resonates with that field.
It's almost the reverse of the way that magnetic resonance imaging machines work, he said.
To do that, the team will need to develop ways to reliably and precisely deposit magnetic layers on to semiconductor wafers, in order to manufacture the tiny transmitters.
They will also have to boost their power. "You need to have enough power to transmit signals reliably. Our devices have a power of 1 nanowatt [one billionth of a watt] but it needs to be 100 times that to overcome the [thermal] noise," he said. The group plans to build the transmitters in clusters of ten or so, benefitting from a phenomenon called superradiance to get the necessary 100-times boost.
Rather than have the transmitters broadcasting in all directions, potentially interfering with one another, Nogaret's group will etch microwave guides onto the wafers, carrying the signals directly to where they are needed.
Once the research is complete, Nogaret predicts that it will take another five to 10 years before the technology appears in production chips.