The Transistor Laser

Ultrafast transistors that output optical and electrical signals open new computing frontier

12 min read

For years we've been hearing about all the fantastic things computers will be able to do once they process data with light instead of electricity. The mysteries of the universe will be unlocked. A golden age of limitless computing power and bandwidth will usher in a techno-utopia.Don't believe the hype.

Setting aside the question of whether an all-optical processor would even be desirable, optical computing schemes lack the photonic equivalent of the most fundamental computing element, the transistor. That device--first demonstrated in 1947, when John Bardeen and Walter H. Brattain stuck two cat-whisker wires onto a germanium base and showed power gain from one wire, called the emitter, to the other, called the collector--spawned the US $300-billion-per-year semiconductor industry. The transistor makes possible our digital lifestyle: cellphones and PCs, digital cameras and MP3 players, medical imaging systems and set-top boxes, supercomputers and the Internet, and more.

The transistor has given us much during the last six decades. Now it has given us light, and with that, the potential for much speedier broadband communications in both telecommunications networks and within and between chips--not in some remote sci-fi future, but perhaps within the next decade.

Our team at the University of Illinois at Urbana-Champaign, has at its disposal an extraordinary prototype transistor that can switch on and off more than 700 billion times per second, faster than any other transistor in the world. On a hunch two years ago, we inspected in greater depth some samples of this transistor, which are made from indium phosphide and indium-gallium-arsenide, the same sort of semiconductor compounds used in today's light-emitting diodes and laser diodes.

We detected significant light in the base of these transistors and immediately began to engineer a new, more powerful kind of device, a transistor laser. Our transistor puts out both electrical signals and a laser beam, which can be directly modulated to send optical signals at the rate of 10 billion bits per second. With some further modification, the transistor laser will eventually send a staggering 100 billion bits per second or more. And it will likely do so at room temperature.

We've had light-emitting diodes and lasers for more than 40 years, transistors for almost 60. But until now, we haven't had a single device that can take an electrical input and simultaneously output both an electrical signal and an optical signal.

The emergence of the transistor laser has taken a long time, but it's not because of lack of interest--but because of a dearth of ideas. Researchers have measured light emission from transistors before. In the early 1980s, a research group at the California Institute of Technology, in Pasadena, led by graduate student Joseph Katz even fabricated a few experimental devices they called translasers. Using a wire, they integrated a transistor with a laser diode to fashion a device that could produce both electrical signals and laser beams, though not both simultaneously.

In 1992, researchers at the Interuniversity MicroElectronics Center, in Leuven, Belgium, built an indium-gallium-arsenide bipolar transistor that emitted light when cooled to liquid nitrogen temperatures. Some groups have even reported that certain transistors emit light at room temperature. However, a transistor working as a transistor has never before shifted its operation into stimulated emission of light--that is, into generating an output laser signal--while simultaneously delivering an electrical signal with gain.

Now, based on our recent work at the University of Illinois, there is such a dual-output device, and a sound reason to envision a future where high-speed computing meets the next generation of broadband communications.

The ability to send and receive signals at the equivalent of three DVDs worth of data--100 billion bits--per second could turn those herky-jerky teleconferences between offices in Tokyo and New York City into high-resolution events. Grandparents in Montreal could watch a granddaughter in a school play in St. Louis through a video cellphone. Supercomputer grids crunching the test data from the world's most advanced particle accelerators might produce results in minutes instead of days. And if you think Internet searching is fast now, wait until servers with transistor-laser-equipped microprocessors pluck answers to the most obscure queries from the hundreds of billions of video, audio, and text files expected to be available online in the next few years.

We're talking instant gratification for the culturally insatiable, real-time entertainment on demand: you'll order Enrico Caruso singing an aria from Rigoletto, the third episode of "The Honeymooners," and the full text of Gravity's Rainbow, and download it all in less time than it takes you to speak requests into your set-top box.

To grasp the profound shift in computing and communications that the transistor laser could make possible, consider how the transceivers in today's optical-fiber communications networks work. First, a computer sends an electrical signal to an optical transmitter, where the signal is converted into pulses of light. The transmitter contains a laser and an electrical driver, which uses the source data carried in the electrical signal to modulate the laser beam, turning it on and off to generate 1s and 0s that travel on the beam through glass fiber. At the end of the fiber, a photodetector reads and converts the data encoded in the photons back into electrical data.

Now imagine implementing a similar process inside a chip, at the transistor level. We can use transistor lasers to convert electrical signals into optical signals and vice versa. In future electro-optical processors, transistor lasers could help route photons through circuits made of waveguides specially designed to take advantage of light's high speed and practically lossless efficiencies over short distances.

Instead of using relatively slow wires to connect chips stacked together in packages, we could use transistor lasers as optical interconnects. These would let data flow instantaneously to and from memory chips, graphics processors, and microprocessors, supercharging weather forecasting and online banking, security checks and telesurgery, airline reservation systems and video games, just about any application.

And then there are the applications we can't even imagine yet. After all, when one of us--Nick Holonyak Jr.--invented the first practical light-emitting diode in the early 1960s, no one could have guessed that it would wind up in traffic lights and key-chain fobs, revolutionize the traditional lighting industry, and become the basis of a global optoelectronics industry worth billions [see photo, " Laser Leaders"].

That industry today is growing at a phenomenal pace. Reed Electronics Research, in Oxon, England, projects that the worldwide market for LEDs will grow from $5.9 billion this year to $9.8 billion in 2008; the market for laser diodes will rise from $4.8 billion to $7.9 billion. The transistor laser that our group is developing could have an equally profound technological and economic impact.

To understand why our transistor laser is uniquely suited to emitting both electrical and optical signals, you need to understand both the workings of the transistor and the basic light-emitting diode, the workhorse of optoelectronics. And to understand the LED, you need to know a little about band gaps and crystals.

In a semiconductor crystal, the arrangement of atoms results in distinct bands of closely spaced energy levels; these determine the energy states of the crystal's electrons. Generally, only two bands matter: the valence band, which contains the energy levels normally occupied by electrons, and the band just above it, called the conduction band. Electrons energetic enough to reach the conduction band are free to accelerate under the influence of an electric field, thereby constituting 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, with typical energies ranging in wavelength from the infrared through the visible.

Normally, electrons occupy the valence band, but jolt them with the right dose of heat, light, or voltage, and they will hop to the conduction band, leaving behind a hole, which is basically the absence of an electron in the crystal lattice.

This electron-hole pair is ephemeral, however; 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 release of energy. In the case of an LED or a laser, a photon is released, whose energy matches the difference between the conduction band and the valence band--the band-gap energy.

But in addition to energy, electrons also have momentum. In indirect-band-gap materials such as silicon and germanium, the minimum energy in the conduction band and the maximum energy in the valence band occur at different values of electron momentum.

Because of that, 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 the right magnitude, such as a vibration in the crystal lattice--a phonon--is generated and assists in conserving momentum in the process. This is a low-probability effect, and defects and heat-generating phonons do the work, yielding very little light. Indeed, so few photons came out of Bardeen and Brattain's primordial device that they probably wouldn't have detected it even if they had tried.

In contrast, the semiconductor compounds we use in our transistor laser, gallium arsenide and indium-gallium-phosphide, readily produce photons. In these materials, which come from the III�V columns of the periodic table, the maximum energy in the valence band and the minimum energy in the conduction band occur at the same value of electron momentum. These III�V compounds are known as direct-band-gap materials because an electron that has been excited into the conduction band can easily fall back to the valence band through the creation of a photon (of little momentum) whose energy matches the band-gap energy.

Photon emission is at the heart of every light-emitting diode. The simplest of semiconductor devices, a diode consists of two terminals and a single junction, called the p-n junction, between them. The p-n junction 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 into the region populated with holes. They recombine and emit light.

The basic structure of our transistor laser can be thought of as two back-to-back diodes separated by a thin connecting layer, a base layer. Called a bipolar junction transistor (BJT), it is one of two distinct families of transistors, the other being field-effect transistors [see illustration, " Anatomy of a Transistor Laser"].

The BJT is a direct descendant of Bardeen and Brattain's point-contact transistor and is so named because the main conduction channel uses both electrons and holes to carry the main electric current. It also shares the same names for the three terminals found on the point-contact transistor: emitter, base, and collector. Two p-n junctions exist inside the BJT: the collector-base junction and the base-emitter junction.

When voltage is applied to the base-emitter junction, injected electrons from the emitter diffuse across the base. The base is thin enough that most of the electrons can pass through to the collector before recombining with holes in the p-type base. (Indeed, some electrons must recombine with holes or the transistor doesn't work.)

The collector-base junction is usually reverse-biased to collect electrons and block holes. Bias is a dc voltage applied between two points to control a circuit. Forward bias occurs when the p-type material is positive with respect to the n-type material and the junction conducts; in reverse bias, the p-type material is negative with respect to the n-type material, and the junction does not conduct appreciably. Because the base-collector junction is reverse-biased to prevent the flow of holes, the electrons coming from the emitter through the base sweep right into the collector, often resulting in gain measured at the collector of as much as 100 times the current in the base.

For several years now, the group working with one of us, Milton Feng, has been developing a superfast bipolar junction transistor consisting of two diode regions made of at least two different semiconductors--in this case, indium phosphide and indium-gallium-arsenide--placed back to back. Called a heterojunction bipolar transistor (HBT), it switches on and off at the world record pace of 710 billion times per second, or 710 gigahertz, and has a breakdown voltage of more than 1.7 volts.

The avalanche breakdown field of a semiconductor refers to a field strong enough to liberate huge numbers of electrons and holes from the atoms of the semiconductor crystal. Finally, if enough charge carriers have been freed from the atoms, the current through the semiconductor surges, burning it out.

A high breakdown field is desirable because it means that a semiconductor device made from the material can withstand higher voltages over small dimensions. Feng's HBT tolerates a relatively high voltage, and that high voltage, along with high current, translates into greater power density for the device. In that device, as much as 10 000 amperes per centimeter squared go into the base, while the collector puts out a current density of 1 000 000 A/cm2.

The base current density of 10 000 A/cm2 is 10 or 100 times as dense as that needed for a state-of-the-art high-speed laser, which pulses on and off at perhaps 10 GHz, or with difficulty to over 20 GHz, at a current density of under 1000 A/cm2. Because Feng's HBT is made of direct-gap III�V materials, we speculated that at these normally destructive current densities, light was probably being generated instead of heat. This light would broadcast and remove the energy lost in recombination and not generate excessive heat.

That idea turned out to be on target. In the summer of 2003, we found infrared light that shot out in all directions. Our HBT is both an electrical and an optical signal source driven by the base current. We call it a three-port device: the emitter is grounded with one input being the base, the reverse-biased collector outputs an amplified electrical signal, and the base itself outputs an optical signal when electrons and holes recombine there.

Discovering the light was only the first step in our efforts to create a useful device. First, we used a slightly different combination of III�V materials. Instead of the indium phosphide and indium-gallium-arsenide used in Feng's record-breaking transistors, we use indium-gallium-phosphide and gallium arsenide, because these materials are more readily available, and easier, quicker, and cheaper to work with. To create a laser beam that we could modulate to carry signals, we had to modify the structure of the transistor both to favor light emission and to intensify it into a laser beam.

It won't be long before we have commercial-grade transistor lasers

A diode laser is based essentially on the same principles as an LED, but it requires a few extra features. The opposite ends of the recombination region must be made reflective to create a resonant cavity that aids in the stimulated recombination and emission of photons, ultimately forming a laser beam. The active area around the junction where the electrons and holes recombine is made smaller to concentrate and improve recombination--the area becomes a quantum well, first introduced into a diode laser by one of us, Holonyak, in 1977.

Quantum wells are regions so flat that they are almost two-dimensional. Electrons and holes confined in one of these thin layers behave quantum mechanically: their energy levels become constricted to certain values, or quantized. Limited to these levels, the electrons and holes become more likely to combine and emit photons in a narrower spectrum.

Because our HBT had gain to spare, we traded some gain for more light generation during recombination in the base region by trapping electrons in a quantum well. In our device, the quantum well is a layer of indium-gallium-arsenide no more than 10 nanometers thick. Inserted into the HBT base region, the quantum well acts like a special recombination center that governs the flow of charge from emitter to collector. It grabs some of the electrons as they blast through the base on their way from emitter to collector, in the process decreasing the gain by almost 90 percent but increasing the recombination of electrons and holes in the base [see again, "Anatomy of a Transistor Laser"].

The product of this enhanced recombination process is a much stronger directed light signal emerging from the base and a complementary electrical signal emerging from the collector. To turn this light into a laser beam, we have to modify the edges of the transistor. Like jewelers cutting facets on a diamond, we cleave the crystal to make the opposite ends of the recombination region reflective, creating a resonant cavity. Photons bounce between the reflective ends, stimulating the emission of additional photons that are in phase with the others generated in the region.

When the light-emitting transistor begins operating as a laser at a near-infrared wavelength of 1006 nm, the spontaneous signal scattered about in the crystal shifts to an intense directed signal, a coherent laser beam that we can toggle on and off 10 billion times per second. The point at which lasing begins, called the lasing threshold, depends on several factors, including current and ambient temperature.

In experiments we conducted two years ago, we had to operate the light-emitting transistor at �73 0C. For our most recent experiments, we operated the device at room temperature. We discovered that the lasing threshold current varied according to temperature, 36 to 40 to 44 milliamps for 20 to 25 to 28 0C, respectively. Operation at these temperatures strongly suggests that it won't be long before we have commercial-grade transistor lasers capable of working under real-world conditions.

When transistor lasers move into commercial fabrication within the next few years, manufacturers won't be cleaving crystals by hand to create mirrors for the resonant cavity. Instead, they will probably opt for more precise plasma-etching techniques to create the mirrored surfaces that bookend the resonant cavity, and they will build more advanced, multifaceted mirrors called Bragg reflectors internally and externally to further enhance the lasing effect inside the device.

Different wavelengths of light are possible, too. The light in our original device was infrared, but more recently we've been able to coax visible light from a primitive transistor made from the same kinds of III�V semiconductor material used for red and yellow LEDs of the sort you find in traffic lights--aluminum-indium-gallium-phosphide.

While we are not very far away from being able to mass-produce transistor laser devices that could be part of an IC-- we can already produce 100 of these devices on a wafer--there are a huge number of issues to keep researchers busy over the coming years. Systems engineers will have to figure out how to integrate transistor lasers into devices and determine what kinds of signals they want to route where. Circuit designers will have to come up with ways to exploit fast transistors that output signals in two different modes simultaneously. And materials scientists will have to find ways to cut the high cost of III�V crystals to make such devices commercially viable on a mass scale.

There is much work ahead, but unlike the host of self-assembling, blue-sky nanotechnologies currently being touted as the next great thing in optoelectronics, transistor lasers do not need an entirely new fabrication infrastructure for further development or even to go into production.

We aren't alone in our quest. The U.S. Defense Advanced Research Projects Agency is funding transistor laser research at the University of Illinois, which is leading a team that includes Columbia University, the Georgia Institute of Technology, and Harvard University. We believe that research into light-emitting transistors will spread to other groups around the world. Any group building high-speed HBTs could turn its attention to developing these devices, and others working with semiconductor lasers will no doubt join the fray as well.

Like the fundamental work that went into creating the point-contact transistor, then the junction transistor, and eventually the LED and the laser--and ultimately led to the booming semiconductor industry--our work is just at the beginning of what could be a new era in computing and communications. Feels like old times, except that we can see the light--and put it to work.

We thank Richard Chan and Gabriel Walter for their contributions to the transistor laser.

About the Authors

Nick Holonyak Jr. is the John Bardeen Endowed Chair Professor of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign, the winner of the 2003 IEEE Medal of Honor, an IEEE Fellow, and the inventor of the LED.

Milton Feng is the Nick Holonyak Jr. Chair Professor of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign and a Fellow of the IEEE.

To Probe Further

For a more detailed look at the transistor laser, see the following papers:

"Room Temperature Continuous-Wave Operation of a Heterojunction Bipolar Transistor Laser," by M. Feng, N. Holonyak Jr., G. Walter, and R. Chan, Applied Physics Letters , Vol. 87, no. 13, 26 September 2005.

"10-GHz Power Performance of a Type II InP/GaAsSb DHBT," by D.C. Caruth, B.F. Chu-Kung, and M. Feng, IEEE Electron Device Letters , Vol. 26, no. 9, September 2005.

"From Transistors to Light Emitters," by N. Holonyak Jr., IEEE Journal of Selected Topics in Quantum Electronics , Vol. 6, no. 6, November/December 2000.

This article is for IEEE members only. Join IEEE to access our full archive.

Join the world’s largest professional organization devoted to engineering and applied sciences and get access to all of Spectrum’s articles, podcasts, and special reports. Learn more →

If you're already an IEEE member, please sign in to continue reading.

Membership includes:

  • Get unlimited access to IEEE Spectrum content
  • Follow your favorite topics to create a personalized feed of IEEE Spectrum content
  • Save Spectrum articles to read later
  • Network with other technology professionals
  • Establish a professional profile
  • Create a group to share and collaborate on projects
  • Discover IEEE events and activities
  • Join and participate in discussions