In a lab in Catania, Italy, on a fine day in May 2001, white-coated researchers connected probes to a sliver of semiconductor, turned on the current, and smiled as bright green light poured from the device. Sure, by then the world was already awash in green, blue, and purple light-emitting diodes, all of them fabricated from gallium nitride and other exotic compound semiconductors. But in that lab demo four years ago, the green glow came not from gallium nitride but from silicon. And at the time, most people in the semiconductor industry would have told you that silicon was pretty much worthless at turning electricity into light.
Despite its bad reputation in optoelectronics, silicon is arguably the most important and intensively studied material known to humankind. In the five decades since the invention of the silicon transistor, electronics and integrated circuits made from the material have comprehensively transformed the world, from the way we work and communicate to how we shop and entertain ourselves.
Silicon makes up the microprocessors, memory, and other chips and devices that constitute more than 85 percent of semiconductors sold--worth US $213 billion last year worldwide. But less than $14 billion was spent on optoelectronics, including the lasers that drive data through the optical fibers that crisscross the planet and the countless LEDs that flash from video billboards and on streetlights that tell you when it's safe to cross an intersection.
Silicon's absence from critical optical applications has long bothered semiconductor specialists. If photons could be easily coaxed from silicon, we could do marvelous things. Imagine plugging your office PC into an optical-fiber local area network and pulling files from a distant server at tens of gigabits per second--enormous, high-definition video files popping onto the screen instantaneously. Optical fibers linking the microchips within a PC would accelerate its computing speed as bandwidth bottlenecks from its motherboard's copper wiring disappeared.
The key to that vision is the fabrication of efficient, electrically driven light sources that work at room temperature and are produced using materials and processes compatible with the manufacturing methods currently used to make ordinary silicon memory and microprocessor chips.
Fiber-optic links are now reserved mostly for long-distance telecommunications. But with huge multimedia files hopping from computer to computer, that kind of bandwidth is increasingly needed everywhere, from local networks right down to the links between chips inside computers. And even in the long-distance links, the benefits of fully integrated optics and electronics would be enormous.
At either end of a fiber-optic link are electronics that route the data down the right path and allow countless conversations and data channels to occupy a single line. With lasers built right into the silicon, the electronics could be more closely and efficiently integrated and could cost a lot less.
Compared with the optoelectronics that drive data across continents, the silicon systems that channel, distribute, and store the torrents of bits are cheap. That's because the worldwide microelectronics industry has cumulatively invested trillions of dollars in building up an industrial infrastructure devoted to designing and manufacturing silicon-based microelectronics in high volumes at low cost. The cost savings go right down to the level of raw materials.
Silicon is one of the most common elements on earth, and a silicon wafer, essentially made from sand, costs just pennies per square centimeter. But lasers and LEDs are made of exotic substances called III-V semiconductors--from their columns on the periodic table of the elements. These materials, which include gallium arsenide and indium phosphide, cost anywhere from 30 to 200 times as much as silicon. What's worse, much of the manufacturing infrastructure and knowledge about how to make integrated circuits in silicon is useless for making chips from III-V semiconductors.
To unite the worlds of microprocessors and lasers, we need cheap, integrated optoelectronics made from silicon. And the heart of such optoelectronics will be a laser. Although a silicon laser was always considered a long shot, academic institutions, research labs, and semiconductor companies have spent millions of dollars in pursuit of it. And now, at last, some are close. Intel Corp., in Santa Clara, Calif., for example, announced in January that it had found a way to power a silicon-based laser with a conventional one. The technique allows engineers to integrate the silicon laser on the same chip with such standard computing fare as logic circuits and memory cells, as well as critical optical components such as the modulators that encode electronic bits onto the light beam. But the scheme does not eliminate the rather expensive III-V semiconductor laser; it just makes it cheaper to use. To be rid of the costly III-V compounds altogether, you'd need a silicon chip that turns electricity directly into laser light.
We are almost there. Although the green glowing device my group built in 2001 in our laboratory--which is part of STMicroelectronics NV, the Geneva-based semiconductor giant--did not emit light that was coherent, collimated, and monochromatic (it wasn't a laser, in other words), as a light emitter, it did match the efficiency of conventional LEDs fabricated from III-V semiconductors. Since then, we've been working to make our LEDs more laserlike, and we believe an electrically powered silicon laser--with all that means for computing and communications--is finally within reach.
Silicon is a lousy light emitter. To understand why, you need to know something about its electronic energy structure. In a typical semiconductor, the regular, repeating arrangement of atoms in its crystalline form results in distinct bands of closely spaced energy levels; these are the allowable energy states of the crystal's electrons. In between those bands are gaps where electrons cannot exist. For most practical purposes, only two bands really matter: the valence band, which contains the energy levels normally occupied by electrons, and the band immediately above it [see illustration, " Mind the Gap"]. The upper band is called the conduction band, because electrons energetic enough to reach it become mobile and free to accelerate under the influence of an electric field, thereby constituting an electric current. The difference in energy between the top of the valence band and the bottom of the conduction band is known as the band gap.
Normally, electrons occupy the valence band, but give them the right dose of heat, light, or voltage, and they will jump to the conduction band, leaving behind something called a hole, which is basically the absence of an electron in the crystal lattice. However, this electron/hole pair--an exciton--is a fleeting thing; sooner or later, the electron falls back to the valence band and recombines with a hole. Because energy is always conserved, this recombination of an electron and a hole is accompanied by the emission of a particle, preferably a photon, whose energy matches the difference between the conduction band and the valence band--the bandgap energy.
Energy, however, is not the whole story. Electrons also have momentum, and when an electron/hole pair is created--or destroyed by recombination--both energy and momentum are conserved. In direct-bandgap semiconductors, such as gallium arsenide, it happens that the maximum energy in the valence band and the minimum energy in the conduction band occur at the same value of electron momentum. With these direct-bandgap materials, an electron that has been excited into the conduction band can easily fall back to the valence band through the creation of a photon whose energy exactly matches the bandgap energy. Photons lack momentum, so it's a straight swap: all the energy of the bandgap jump goes into the photon.
That is essentially how any ordinary III-V light-emitting diode works. The key component of an LED is a p-n junction, a division in a semiconductor that separates a region rich in conduction-band electrons (n-type material) from one that is rich in valence-band holes (p-type material). Applying a negative voltage to the n-type side pushes the electrons across the junction and into the holes, and vice versa. They recombine and emit photons. The ratio of generated photons to the electrons injected across the junction is called the quantum efficiency, a key measure of how well a light emitter is working. For high-performance III-V LEDs, the efficiency is around 10 percent.
A III-V diode laser is based on essentially the same principles, but it requires a few extra features. The active area around the junction where the electrons and holes recombine is made smaller, to concentrate the recombination, and the opposite ends of the recombination region are made reflective. Photons bounce between the reflective ends, colliding with atoms and stimulating the emission of additional photons that are in phase with the others in the region. In a laser, the concentrated, active region bounded by the reflective ends is known as the resonant cavity.
Things are not so simple for silicon and many other semiconductor materials. The main problem is that their crystal structure results in what's called an indirect band gap. The minimum energy in the conduction band and the maximum energy in the valence band occur at different values of electron momentum. That means an electron in the conduction band can recombine with a hole in the valence band to produce a photon only if a source of momentum of just the right magnitude, such as a vibration in the crystal lattice--a phonon--is present. The probability that a phonon with just the right amount of momentum will meet an electron/hole pair in a silicon crystal is not very good. In fact, the occurrence of a photon-generating transition in a III-V material is thousands of times more likely than that of such a transition in silicon.
So in silicon, few excited electrons generate photons, most recombinations result in heat rather than light, and the quantum efficiency is terrible.
Faced with those problems, researchers have been pushing two strategies in their quest to get light out of silicon. One scheme is based on a curious effect called quantum confinement. That occurs when an electron/hole pair is physically restricted to a small area, typically less than 30 square nanometers, or 300 times the size of a typical atom. Embedding nanocrystals of silicon within an insulating silicon dioxide layer is one way to make such quantum cages. Within a nanocrystal, the energy levels of the valence and conduction bands differ significantly from those in bulk crystal. In general, the smaller the nanocrystal, the bigger the band gap, opening up the possibility of tuning a device's optical properties by fine control of the nanocrystal's growth during the manufacturing process. Best of all, quantum confinement reduces silicon's momentum problem, increasing the probability that injected electrons will produce photons.
The other idea scientists have pursued is to sidestep silicon's bandgap problems by having another material, embedded within the silicon device, emit the light. That's done by seeding the silicon with lanthanide rare-earth-element ions, which tend to give off light when electrically excited. Some of these, those with atomic numbers from 58 (cerium) to 71 (lutetium), form a group with similar chemical characteristics. The elements' particular electronic configuration is such that if you put them in another material (silicon or silicon dioxide, say), their electronic properties are not much influenced by the host material's quirks (say, low light emission).
Our approach combines both techniques. It has produced light emitters that operate at room temperature with a controllable tradeoff between high efficiency and long lifetime. The device structure looks very much like the metal-oxide-semiconductor transistors that make up the circuits in most microchips [see illustration, " The Silicon LED"]. Atop a region of p-type silicon we built a thin insulating layer of what's known as a silicon-rich oxide. That's simply silicon dioxide with a little extra silicon. Rare-earth ions are implanted in the middle of the oxide layer, and it is heated. The heat causes the silicon to clump spontaneously into crystals a few nanometers across. To finish the device, an n-type silicon layer is added with a metal electrode on top of it.
Applying a voltage to the electrode sets up an electric field that accelerates electrons across the silicon-rich oxide layer. These "hot" electrons collide with the rare-earth ions, kicking them into energy states that lead to light emission. The silicon nanocrystals have two roles. First, they greatly improve the conductivity of the silicon dioxide layer, and that boosts the device lifetime, though it reduces efficiency. Second, instead of emitting light themselves, the nanocrystals act like energy funnels leading to the ions. Hot electrons or emitted photons excite the nanocrystals, which then transfer their excitation to nearby ions, adding to the light emission.
The result is a device that glows brightly at room temperature, with a quantum efficiency of up to 10 percent, comparable to that of state-of-the-art III-V devices.
A great advantage of the technology is that the color of the light emitted depends only on the rare-earth ions used. Samarium glows red; terbium, green; cerium, blue; and erbium, conveniently, yields the infrared used in many telecommunications devices [see photo, " Color Codes"].
The problem with the approach for some uses, at present, is low light output. Although the silicon LED can be as efficient as its III-V competitors, it produces only a fraction of the light you'd get from a commercially available LED. That's because the maximum output power is limited by how densely we can pack the device with rare-earth ions, and that limit at the moment is about 1 quadrillion ions per square centimeter.
We have some power-boosting tricks that we're pursuing, but for many useful applications, they aren't needed. For example, we integrated a silicon LED into a microchip for use as an optocoupler, which is an LED lined up with a photodetector in the same package. An optocoupler converts electronic signals from one circuit into optical ones and then back again. In power control systems, optocouplers electrically isolate the control circuitry from power-switching transistors to guard against short circuits or unwanted feedback.
In an optocoupler we demonstrated last year, we used our new silicon light-emitting technology to integrate the control circuitry, the LED, on-chip light guides, the detector, and the power-switching transistors all on the same chip. For this device, efficiency trumps power. The LED has to shine only brightly enough to communicate accurately with a detector mere millimeters away on the same chip, but we want the current that is required to generate this light to be as low as possible to minimize the chip's overall power consumption.
An efficient integrated LED will find many uses, but for the bandwidth-boosting chip-to-chip and computer-to-computer communications that we and others envision, the light source needs to be directional--more like a laser and less like a lightbulb. Without a laser or something like it, the coupling between the light source and the detector, optical fiber, or any other component may be too weak to reliably get a signal through.
Adding the resonant cavities used in lasers goes part of the way to turning an LED into a laser [see illustration, " Almost a Laser"]. Over the years engineers have designed resonant cavities into LEDs, transistors, photodetectors, waveguides, and switches for fiber-optic communications, to name a few applications.
In semiconductors, a resonant cavity is usually formed by sandwiching the light-emitting region between two stacks of mirrors, called distributed Bragg reflectors. The mirrors are created by alternating layers of two semiconductor materials, one with a high refractive index and the other with a low refractive index. We use silicon and silicon dioxide.
The difficulty in designing a silicon resonant-cavity LED is pumping the light-emitting part of the device with electrons despite its being embedded between two stacks of insulating mirrors. Through a proprietary design, we've managed to do just that. We can pump the silicon-rich oxide and rare-earth ion film without having to electrically contact the mirrors. We've made resonant-cavity LEDs using erbium ions tuned to shine at a narrow band of wavelengths centered, happily enough, at a telecommunications-friendly 1540 nm. As we've made more devices with a greater number of mirrors in the resonant cavity, they've begun to approach a laserlike ideal of a powerful, directional, single wavelength.
Indeed, our resonant-cavity LED device is almost indistinguishable from a vertical-cavity surface-emitting laser, or VCSEL. This type of microlaser is replacing traditional lasers and LEDs in everything from optical fiber communications to optical mice, and it's no wonder, since they're cheaper to build than other kinds of III-V lasers.
To get from LED to VCSEL in a resonant cavity, the light bouncing between the mirrors must manage to excite the majority of the rare-earth ions, creating a state called population inversion. The performance characteristics of the erbium-based, silicon-rich-oxide LEDs we've already built indicate that achieving population inversion is within reach. All we need to do is further reduce the amount of light lost within the structure to imperfections in the mirrors and elsewhere. We're confident we can accomplish that within a couple of years.
So get ready for light-speed links to your PC and even inside it. An electrically driven silicon laser is just around the corner.
About the Author
SALVATORE COFFA is deputy director of STMicroelectronics NV's microcontroller, linear and discrete group, based in Catania, Italy.
To Probe Further
Learn more about STMicroelectronics' silicon LEDs in "High Efficiency Light Emitting Devices in Silicon," by Maria Eloisa Castagna et al., Materials Science and Engineering B , 15 December 2003, pp. 83–90.
For more on optical links in computers, see "Linking with Light," by Neil Savage, IEEE Spectrum, August 2002, pp. 32–36.