Kent Choquette thought the type of laser he’d spent his career developing was doing pretty well in 2017. That device, the vertical cavity surface-emitting laser, or VCSEL, had seen remarkable success since it was first introduced commercially in 1996, and by 2017 an estimated one billion of the tiny devices had been sold, mostly for use in optical interconnects for data centers.
“We were really proud of that,” says Choquette, a professor of engineering at the University of Illinois, Urbana-Champaign. But that success was soon dwarfed. “Literally every year since, we put out at least a billion more.”
2017 was the year that Apple decided to incorporate VCSELs in its iPhone X to run its Face ID and proximity applications. The lasers send out a pattern of dots of infrared light, and the phone’s camera captures the returning light and measures how long it took to come back to construct a 3D image of a face. Other manufacturers began incorporating VCSELs as well, and, as the smartphone market boomed, so too did the VCSEL market.
VCSELs long ago established dominance as lasers for low-power, short-distance applications, pushing out the traditional edge-emitting lasers, which produce light along the plane of the lasers and emit a differently shaped laser beam. The design of VCSELs allows them to be manufactured easily. They’re built by depositing a series of layers on a semiconductor wafer, first alternating layers of materials with different refractive indexes to form a mirror, then a cavity and an active region with multiple quantum wells, then another mirror on top. They’re tiny and inexpensive, and because they emit light out of the top surface, they’re easy to integrate with other components or to be used as arrays. Depending on the application, their power consumption can be 10 to 100 times lower than edge emitters
In fact, says Jung Han, a professor of computer and electrical engineering technological innovation at Yale University, their advantages are so great that they overcame a significant disadvantage for use in data centers—the lasers have the wrong wavelength. Most of the VCSELs in data centers are based on gallium-arsenide (GaAs) and emit light at 850 nm. The fused silica fibers over which the lasers transmit data between racks of computers in a data center are not fully transparent at that wavelength, because they’re optimized for standard telecommunications wavelengths between 1,300 and 1,550 nm, where VCSELs are still under development. With 850-nm VCSELs there’s a fair amount of lost signal in the fibers, but for the 30 to 100 or so meters between racks it’s an acceptable loss. “They just go with the wrong wavelength because it’s a lot cheaper than edge emitting,” Han says.
That’s not to say that people don’t want to improve the lasers. There’s work to develop new VCSELs built of different semiconductors to reach other wavelengths, both longer and shorter than the near-infrared range that currently dominates. Researchers are also trying to make the lasers more powerful and more energy efficient, and they are exploring new designs to accomplish that. Meanwhile, applications beyond data centers aim to take advantage of VCSELs’ appealing qualities.
“It’s the applications that are driving the VCSEL research,” says Choquette. Manufacturers of virtual reality and augmented reality glasses would like to use different semiconductors to create tiny, cheap lasers emitting red, green, and blue for their displays. Automotive companies use VCSELs to power the lidar systems that autonomous vehicles use to detect their surroundings. Most current lidar systems use VCSELs emitting at 905 nm, but longer wavelengths, say around 1,550 nm, experience less interference from sunlight, and also pose less danger to the human eye and so can be used at higher power to achieve greater distances. Consumer products that need to sense their surroundings or communicate with others—a next-generation robot vacuum cleaner, for instance—could also benefit from having access to more eye-safe wavelengths. The right wavelengths in the shortwave infrared (SWIR) spectrum, roughly between 1,300 and 4,000 nm, could be used to detect certain biomolecules, while others can trigger targeted drug release or have cosmetic applications. To lower energy use, data center operators would like more efficient lasers. “And you want to get higher power basically for everything,” Choquette says. “There’s no single application that would say, ‘Please don’t give me that much power.’”
There’s even interest in creating VCSELs in the ultraviolet (UV) region. The Defense Advanced Research Projects Agency, for instance, has a program, Laser UV Sources for Tactical, Efficient Raman (LUSTER), seeking to develop a compact laser emitting between 220 and 240 nm. Such lasers could be used for detecting chemical and biological weapons, as well as for decontamination, precision manufacturing, and medical diagnosis, but most UV lasers are too large, complicated, and expensive for use outside of the laboratory. Using DARPA funds, Han has tried since 2014 to build a near-UV VCSEL. “It turns out it’s very, very hard,” he says. “To this day, nobody has achieved a near-UV VCSEL below 400 nm.”
Oxide-confined VCSEL arrays. Photo credit: Kent Choquette, University of Illinois Urbana-Champaign.
Han has had more success with blue-emitting VCSELs based on gallium-nitride (GaN). He’s even founded a startup company, Ganvix, to develop and market blue VCSELs, though he says it may be three to five years before they can really make an impression in the market.
One of the main challenges for VCSELs in the blue region has been to make the distributed Bragg reflectors (DBRs) that act as mirrors in the devices. DBRs require materials with very different refractive indexes, which can be placed in alternating layers, to create the mirrors. To achieve that, Han and others developed a method to tune the index of GaN using an electrochemical process to stud the material with pores. The holes fill with air, which has a refractive index of 1, thus changing the index of the semiconductor.
One challenge for GaN VCSELs is their low efficiency. Wall-plug efficiency—how well the laser converts electrical power into optical power—doesn’t often get much above 10 percent. Last year, Tetsuya Takeuchi, a professor of materials science and engineering at Meijo University, managed to increase that to 20 percent. The VCSEL in question used alternating layers of GaN and aluminum-indium-nitride to form the DBR, but controlling the growth of the GaN layer, and thus the quality of the mirror, was difficult. Takeuchi developed a process to measure reflectance from the GaN during growth, thus allowing him to more precisely control the size of the cavity. He used a different measurement technique to monitor the thickness of an indium-tin-phosphide electrode in the VCSEL and a space layer made of niobium pentoxide. This precise calibration of the different components of the VCSEL led to the efficiency improvement.
Tien-chang Lu, professor and chair of the Advanced Nanophotonics Lab at National Yang Ming Chiao Tung University in Taiwan, says advances like Takeuchi’s have given blue VCSELs performance that is almost as good as GaAs-based VCSELs. The question that remains, he says, is whether they can achieve mass production. “Maybe there will be no universal techniques for blue, green, and UV VCSELs,” Lu says.
Another technique for making the DBRs involves using dielectric thin films. Such mirrors are bad at heat dissipation, which can reduce the performance of the VCSEL, but they may be a way to achieve UV VCSELs, Lu says, while Han’s porous approach is better for blue VCSELs.
At a slightly longer wavelength, green, VCSELs continue to lag. The problem is that, while mirrors exist, gain—the increase in power that makes a laser—is very weak. Out at 1,550 nm, VCSELs have the opposite problem. Those lasers are based on indium phosphide which have reasonable gain. Meanwhile, gallium antimonide VCSELs reach wavelengths of 2,100 nm and above. “But none of these material systems support very simple, high-quality mirrors that can be easily combined with the active region to make efficient, high-performance VCSELs,” Han says.
He’s been tackling that problem by taking the pore-etching idea from GaN and applying it to these materials. The attempt, he says, has been successful. In fact, he predicts that indium phosphide-based VCSELs at 1,550 nm—a wavelength useful to long-distance fiber optic communications—could be ready for mass production within just one or two years.
Photonic crystal coherent VCSEL arrays. Photo credit: Kent Choquette, University of Illinois Urbana-Champaign.
Even while some researchers work to extend VCSELs in visible and SWIR wavelengths, others continue to improve the GaAs devices in the 750 to 980 nm range. Dieter Bimberg, founding director of the Center of Nanophotonics at Technical University Berlin and head of the Bimberg Chinese-German Center for Green Photonics, is looking at improving the efficiency of VCSELs as a means to reduce the power demands in data centers. Generative AI and blockchain for cryptocurrency use enormous amounts of computing resources, which drives a huge demand for electricity, and the growth is unsustainable, Bimberg says. “You cannot always continue asking for larger bit rates. Because if you want to get a bit rate which is larger by a factor of two, you must increase the power of the laser by a factor of four,” he says.
Bimberg found he could increase the efficiency of VCSELs by etching holes into the top of the lasers. He fills the holes with gold, and the metal acts as a heat sink and decreases electrical resistance, reducing heat in the cavity and avoiding the problem of thermal rollover, in which power output drops even as current increases, to much larger currents. More holes provide more cooling, and decreasing the temperature makes the laser operate more efficiently and at higher bit rates than with the classical design. It’s also possible to etch holes and not fill them, creating extra apertures to let the light out. Two apertures in an asymmetrical layout will produce output polarized in two ways, allowing so-called polarization multiplexing to double the bit rate. Three or four apertures and the bit rate can be tripled or quadrupled, while still fitting the beam into the same fiber.
Instead of improved energy efficiency, it might be useful to get higher single-mode output from a VCSEL. A brighter VCSEL could, for instance, allow a lidar signal to travel farther. One method being explored to squeeze out more light is the creation of multijunction laser diodes. Standard diodes use a positive-negative junction. Feeding in voltage excites an electron and separates it from a hole, and when the two recombine, the extra energy is released as a photon. A multi-junction diode allows that process to happen several times.” At the expense of more voltage, the same electron can generate multiple photons,” Choquette says. “So, you can get much higher power. You can get quantum efficiencies exceeding 100 percent.”
Lu says that if multijunction diodes can be developed for GaAs VCSELs, the same design could also increase the output power of GaN VCSELs. Choquette is also working on creating so-called coherent arrays of VCSELs. Today two-dimensional VCSEL arrays might come in groups of, say, 300 lasers, but they’re independent of one another. While all the lasers emit at close to the same wavelength, that can vary slightly from one device to the next due to variations in temperature and current. There may also be differences in the phase of the light from one laser to the next.
Choquette would like to be able to have all the VCSELs in an array operating at the same wavelength and phase, essentially creating one laser mode spread out over the entire array. It’s similar, he says, to the concept of beam combining in which the output from several lasers is mixed to form one, much brighter, laser beam. That might be useful when it’s necessary to produce a more powerful beam that can travel farther. “The incoherent arrays will be sort of like Fords and coherent arrays would be like Ferraris,” Choquette says. “We just haven’t gotten around to being able to afford the Ferraris quite yet in the applications, but that’ll probably come.”
Choquette has been in the VCSEL field since he earned his PhD in materials science in 1990, several years before the first VCSEL was commercialized. He invented a version of the device known as the oxide-confined VCSEL, which is widely used in data communications and sensing. He sees the field being just as vibrant as ever. “After 30 years in this, watching this research, there continues to be new applications of VCSELs,” he says. “That’s why the train keeps rolling.”
Neil Savage writes about science and technology in Lowell, Massachusetts.
