Researchers at MIT have created an optical switch so small it can be controlled by a single photon, raising the possibility of creating microprocessors based on the passage of light through a circuit rather than electricity. The technique could allow the development of microprocessors that work in ways similar to current microchips, without most of the size, energy and weight requirements of current versions, according to researchers from MIT, Harvard University and Vienna University of Technology. In a paper published this week in the journal Science (registration required), researchers described building a tiny optical switch that can be controlled by some photons even as it allows others to pass through. Using the tiny switches, it might be possible to create an optical version of the transistor that lies at the heart of modern CPUs, creating computing circuits that could be far smaller and more densely packed into a microprocessor substrate, use less power and generate less heat than electrical processors. All those factors limit the size to which electrical processors can shrink, which will eventually cause the steady increase in CPU power described in Moore's Law to plateau. The switch is made up of pairs of mirrors that are highly reflective, but which allow photons to pass through when the switch is turned on. When it is turned off, only 20 percent of the light passes through. The mirrors act as a resonator that, when set at the proper distance and angle, become transparent to the wavelength of light being used, but which help create an electromagnetic field that keeps photons from bouncing back toward their source when their way is blocked. That allows photons to move in only one direction unless a "gate photon" is fired into the supercooled cesium atoms filling the space between mirrors. Colliding with the cesium atoms, the gate particle throws one electron from one atom into a higher state of energy, turning the mirror opaque and blocking transmission through that panel. The power and design of modern semiconductors is limited partially by the width of the channels that can be carved into silicon to allow electrons to pass through. Intel's smallest commercially produced processors are built on either 22 nanometer or 19 nanometer-side channels. Photons, which are smaller than electrons and can behave as both a particle and a wave, don't have that limitation. Processors based on optical processors could be designed with far more circuits (and therefore computing power) than electron-based processors—and would use only a fraction of the power of electricity, according to Vladan Vuletić, the MIT physics professor who led the research. Optical processing could be a huge advance for conventional computer designs due to improvements in power efficiency, circuit density and resistance. In quantum processors, the optical circuit could make it far easier to keep the qubits—the quantum version of the 1 and 0 bits in conventional computers—in the right position to be used. Optical circuits can also be used to create quantum states that wouldn't be possible otherwise, further expanding the power available. "The beauty of this approach is that it can really do switching at the single-photon level, so your [energy] losses are much smaller," Vuletić was quoted as saying in the MIT news office story announcing the result. "You don’t have to spend a lot of energy for each bit. Your bit is essentially included in a single photon." This particular design is not practical for commercial development because of the need for supercooled cesium atoms and other characteristics acceptable only in experimental design, she said. However, it should be possible to design the same process using microscopic cavities in a chip to replace the cesium-filled mirror chamber, along with quantum dots to fill in as an artificial atom. "There would be extra steps that people would have to take in order to implement the right energy-level structure," Vuletić said. "But in principle, the physics could be translated to a platform that could be cascaded and more easily integrated."   Image: MIT/Christing Daniloff