Two fully functional optical memories on single chips have been fabricated by researchers in Japan. The devices use bistable optical cavities to store the bits, and allow multiple bits to be controlled simultaneously by the same waveguide. The researchers hope that, in future, such a memory could be used for optical logic operations to increase the speed of computation.

Today, optical fibres are the material of choice for transmitting data, thanks to their lower signal attenuation compared with copper wires and their much higher bandwidth. Currently, however, optical signals have to be converted into electronic ones for processing, and then once more, to convert the output back to an optical signal. Such conversions consume energy and time, and fail to utilize the biggest advantage of optical transmission – that photons do not interfere with each other, meaning that several signals with different frequencies can travel down one fibre simultaneously in a process known as "multiplexing". Photonic signals have to be "demultiplexed" before an electronic processor can deal with them, and so optical processors are of interest to many researchers.

Random memories

A key element in any processing unit is the random access memory (RAM), in which data are stored temporarily while the computer runs a program. A modern electronic RAM usually stores each bit of memory as the charge on a capacitor, and various optical equivalents have been proposed. In 2012 Masaya Notomi and colleagues at NTT Laboratories in Kanagawa, Japan, designed a four-bit RAM made from a photonic crystal – a periodic optical nanostructure comprising a network of holes that allows some wavelengths of light to propagate while blocking others. Inside the photonic crystal were four identical cavities that had two possible refractive indices – a pulse of light at the cavity's resonant frequency would allow a switch between the two indices, while light at a different frequency would reveal the cavity's state without disturbing it. By designating the two states as 0 and 1, the researchers created a readable and rewritable memory. However, each of the cavities had to be controlled by a separate waveguide.

Now, the same researchers have made the cavities much smaller and non-identical, allowing them to introduce multiplexing. They created two different types of optical RAM – one made from silicon and the other from indium phosphide and indium gallium arsenide phosphide. In each RAM, multiple cavities were arranged lengthways, with a single waveguide passing all of them. The researchers used computer modelling to work out exactly how to move specific holes in the photonic crystal such that each cavity had a slightly different resonant frequency. They were then able to send a "write" pulse down the waveguide containing the frequencies of whichever bits they wanted to switch and only those cavities would respond.

Stable lifetimes?

The silicon RAM contained 105 working cavities, with all the resonant wavelengths falling between 1540 nm and 1570 nm, at an average spacing of just 0.23 nm, all of which was fabricated on a silicon crystal just 1 mm long. Unfortunately, the cavity states were stable for less than 10 ns – too short for a viable optical memory. However, the lifetime of the bits in the indium-phosphide-based RAM was, in principle, infinite. Because indium phosphides are less well established in industry than silicon, the technology for manufacturing indium-phosphide components is less precise, and so Notomi and colleagues could only produce a 28-bit memory. However, they believe this provides a better blueprint for future research. "Our final goal is to produce better indium-phosphide systems by improving the fabrication accuracy," says Notomi.

Martin Hill of the University of Western Australia in Crawley describes the paper as "a nice piece of work on a difficult area of photonics". But he also points out that, at present, the switching speed of the optical cavities is lower than the switching speed of electrical transistors, and says that before the device becomes useful as a product, the researchers need a way of making the switching frequencies more predictable and reproducible.

The research is published in Nature Photonics.