Copper connections are now a major factor threatening the reign of Moore’s Law in the design of very large-scale integration (VLSI) chips. These interconnects not only are costly but also drain a good part of the energy used to power the chips, more than the transistors themselves consume. Above all, they use up a lot of space.
Many research groups around the world are therefore looking for ways to replace copper connections on chips. Approaches include relatively straightforward ones, such as optical communications [see ”Laser on Silicon,” in this section], as well as exotic ones. A really exotic concept, proposed by a University of California, Los Angeles, team led by Mary Mehrnoosh Eshaghian-Wilner, relies on atomic spin, a quantum-mechanical property related to magnetism, and on waves generated when that spin is disturbed.
Atomic spin arises from the magnetic fields generated by an atom’s spinning electrons. Both atomic and electron spin can be thought of metaphorically as rotation, but they are in fact abstract properties and are represented as vectors that can point either up or down—that is, either parallel or perpendicular to magnetic field lines.
In a layer of ferromagnetic material such as a cobalt-ion alloy, the atomic spin vectors all point in the same direction. If, however, the layer is subjected to a subtle magnetic pulse, it will be deflected and start precessing like a top, so that its spin axis forms a cone. The first atom affected by the pulse will pass along energy stored in the precession to the next atom, generating a spin wave that propagates through the ferromagnetic layer.
To demonstrate that such spin waves can transmit data, the UCLA team created a prototype device containing a sender and a receiver. They deposited a 100â''nanometerâ''thick ferromagnetic film of a cobalt-iron alloy onto a silicon substrate, with an insulator layer of silicon dioxide on top. At each end of the device, 8 micrometers apart, they deposited a pair of conducting strips separated by a small gap, one pair serving as a transmitter, the other as a receiver. Applying a voltage pulse to the transmitter pair causes a magnetic-field pulse in the ferromagnetic layer, which in turn disturbs the aligned spins in this layer, creating a spin wave. The receiver pair detects the passing spin wave, because the magnetic field traveling with the spin wave sets up a tiny voltage in the line.
The team reported at a meeting in Italy in May that a 24.5-volt pulse produces a 26-millivolt response in the receiver pair. According to group member Alex Khitun, the team has improved on that result in the meantime. ”There is enough room for scaling down the input voltage and increasing the output voltage,” he says.
Eshaghian-Wilner and her colleagues have presented possible designs for VLSI architectures that would use spin waves to transmit data between processors. The simplest one consists of a number of pairs of conductor bars in a circle on a ferromagnetic film [see illustration, ”Spinâ''wave Bus”]. Each pair of strips can either transmit or receive data, and several data streams can be transmitted simultaneously by creating spin waves of different frequencies.
Because of that ability to transmit at different frequencies through a single ferromagnetic layer, the UCLA researchers argue that spin waves will substantially increase the amount of data that can be processed in some standard systems in microcircuits.