Silicon photonics: a light at the end of the tunnel?

17 Jun 2002

Silicon light sources could replace electronic circuits with optoelectronic ones. Michael Hatcher talks to the key players in the quest for a silicon laser.

From Opto & Laser Europe June 2001

Leigh Canham

Compound semiconductors and silicon are simply not natural bedfellows. The difference in lattice constant between a silicon substrate and a light-emitting compound layer causes distortions within devices that degrade the active layer's electrical conductivity. The faces, quite literally, do not fit. As a result, devices like diode lasers must be kept separate from silicon circuits, thus blocking the path to miniaturization and the development of optical circuits for advanced computer systems.

If silicon could emit light efficiently, there would be no such problems. However, silicon is an awkward customer when it comes to light emission. Its indirect bandgap means that, for light emission to be fairly efficient, recombining electrons and holes must have exactly the same momentum. This effectively makes light emission very unlikely, so silicon usually emits only tiny numbers of photons. In contrast, direct-bandgap materials like gallium arsenide do not have such a strict dependence on momentum and are far more efficient.

Over the past decade, however, researchers have made progress in finding ways to coax light from silicon. Each method relies on the same principle: the law demanding similar momenta of electrons and holes breaks down under certain circumstances - quantum confinement conditions. In essence, this means that, on the nanometre-scale, silicon crystals emit light efficiently.

Leigh Canham's laboratory at the UK's Defence Establishment Research Agency (DERA, soon to be known as QinetiQ) was the first to notice this effect. By etching away most of a lump of bulk silicon with hydrofluoric acid, Canham made porous silicon - a kind of "quantum sponge" comprising a mesh-like network of silicon wires a few nanometres in diameter. Under ultraviolet illumination, these nanowires glowed with visible light.

Following this initial work, Canham's team and another led by Philippe Fauchet at the University of Rochester in New York, US, set about making electrically driven porous silicon LEDs. However, both groups found that poor efficiency conversion generated a tough roadblock. "We got stuck at about 0.1% power efficiency," said Fauchet.

Then, in 1998, a group in Tokyo led by Nobuyoshi Koshida increased the quantum efficiency to about 1%, but, says Fauchet, this is still too low. "It's frustrating because in photoluminescence experiments, the efficiency can reach up to 10%," he said.

Silicon nanocrystals

For high-speed silicon optoelectronics there is a further problem, according to Canham. Even at nanoscale dimensions, silicon retains some indirect bandgap nature, which means that light production is slower than in compound materials. Silicon LEDs could therefore only switch at megahertz frequencies, rather than the gigahertz frequencies needed for high-speed communication. To solve this problem a silicon laser is needed, and it is in this context that last year's observation of optical gain, by Lorenzo Pavesi and colleagues from Trento and Catania, is crucial.

The generation of optical gain is obviously a critical early hurdle to overcome in any laser system. Having investigated the device with pump-probe spectroscopy, Pavesi has come up with a three-level model to explain the origin of the gain, where an interface state exists between the silicon nanocrystal and the silicon oxide. Photons from an argon laser pump silicon atoms from the valence band into the conduction band. Decay from the conduction band into the interface state is very fast, but slow decay from here back to the valence band is very slow, and the result is a population inversion that amplifies the 800 nm probe.

However, Fauchet is somewhat sceptical about the gain mechanism that Pavesi has suggested: "In nanocrystal silicon, I think that excited absorption will always dominate the band-to-band recombination, hence making gain impossible," he said.

Canham, while agreeing that there are some problems with the model, stresses that the key thing now is for other workers to duplicate or improve on Pavesi's experiment. "If Pavesi's achievement can be reproduced, then it would be highly significant. There are some issues regarding the exact mechanism, but the key will be to reproduce those high gain values," he said.

Fauchet and his co-workers have already made an all-porous-silicon microcavity resonator LED with narrowband electroluminescence. This emits over a wavelength-tunable range of 670-750 nm and Fauchet says that these devices could be used in high-quality flat-panel colour displays.

Now, however, Fauchet says coyly that he is planning to do something "along the lines of Pavesi's work, but with a different twist that gives us an advantage". What that advantage is, he is not saying, but he is far less confident about any of the other approaches: "They are far from showing real promise as far as lasing is concerned."

One of the other approaches is employed by Kevin Homewood and colleagues at Surrey University in the UK. They have found a way to coax out silicon light emission with electrical stimulation. Homewood and Karen Reeson engineered local loop dislocations by firing boron atoms into bulk silicon. The resulting structure emits 1.155 µm light at room temperature under forward bias, and Homewood says that he could have a laser based on the same principle within a year.

Silicon nanodots are also gaining ground as candidates for next-generation displays, and, with this in mind, chemical engineers Brian Korgel and Keith Johnston at the University of Texas, US, have shown how to make different-size dots emit in the green, blue and possibly red. Controlling the dot size appears to be relatively straightforward with their novel manufacturing process in which long-chain alcohol molecules bind to the silicon surfaces. By controlling the degree to which these organic "ligands" are present, the desired nanodot size can be produced.

"If you have a lot of ligands, the crystals will stay small," said Korgel. "If you don't have many, they'll continue to grow into bigger ones." The simple manufacturing method is not perfected - Korgel is yet to make a nanodot emitting in the red - but it ought to result in cheap devices. If they can make the switch to electrical stimulation, there should be the prospect of making full-colour displays with the material.

Another potential route to a silicon laser, albeit one emitting in the infrared region, could lie in cascade devices. In these, silicon's indirect bandgap is of little consequence because there is no recombination of holes and electrons and no transfer of momentum. In the cascade structures, simply the transfer of an electron between states provides the light emission. Researchers at the Paul Scherrer Institute in Villigen, Switzerland, recently made some progress on this front. They saw electroluminescence from a cascade structure based on p-type silicon/silicon germanium. Transitions occur within a single band, rather than from one band to another, and this means that the emitted wavelength is in the mid- or far-infrared.

The Swiss team's device emitted at 10 µm, but a familiar problem is thrown up with the SiGe structure - silicon and germanium have a 4% lattice mismatch, and the potential for forming defects is obvious.

Although conversion efficiency in silicon devices has been steadily improving over the past decade, there is still a need to push it higher. Canham says that one way to do this is to improve the size distribution of the nanocrystals and quantum wire diameters. "III-V quantum dot devices performed much better when the size distribution was improved, and I think that the same thing could happen with silicon," he said.

He is less optimistic about the potential of Kevin Homewood's devices, because the reported conversion efficiency was only 0.02%. "The key thing is not to show CMOS compatibility but to demonstrate 10% efficiency and fast operation. What will make semiconductor people sit up and take notice is devices that show 10% efficiency," he commented.

However, as Homewood points out, that figure neglects the degree of edge-emitted light, which, if included, would raise the efficiency to 0.1%. He adds that further improvements in device packaging could increase this by an order of magnitude.

Significantly or not, Canham is no longer directly researching light-emitting silicon but is concentrating on the biomedical applications of porous silicon, having been made chief scientific officer at DERA spin-off pSiMedica. While he says that the performance of porous silicon LEDs is close to that needed for some displays applications, the real driver for silicon optoelectronics is still optical interconnects. That demands improved device efficiency and brightness. (The highest quantified brightness so far is 50 cd/m².)

It would seem that there is a light at the end of the tunnel, but for the moment that light just isn't quite bright enough.