For many years, engineers successfully integrated optics and electronics by using a standard silicon CMOS process for small-bandwidth structures like photodetectors. Advances in performance and integration densities continue to energize this niche, such as with the venerable silicon photodiode in terms of functionality. Photodiode advances may fall under the radar a bit, but more visible progress can be seen with ICs like detectors, sensors, LEDs, lasers, and other devices that operate at relatively low bandwidths. They serve barcode scanners, printers, disk players, remotecontrol devices, and other consumer applications. Other markets include medical, industrial, and construction applications. Advanced Photonics, Cal Sensors, Opto Sensors, Silicon Sensors, and UDT Sensors all use silicon and other materials to produce photodetectors and sensors. Some large firms like Agere, GE, Infineon, Perkin-Elmer, and Siemens also make commercially available solid-state photodetectors and sensors.
Products from Texas Advanced Optoelectronics Solutions (TAOS) exemplify this trend. The company’s high-performance optodetectors convert light to voltage, color, digital, and frequency signals (see “Beyond Simple Photodiodes And Phototransistors” at www.electronicdesign.com, Drill Deeper 18789).
The Position Sensitive Photomultiplier Array (PSMArray) family of tiny solid-state light sensors from SensL represent the first commercially available CMOS large-array, detector-based silicon photomultiplier products. According to the company, they surpass the performance of traditional photomultiplier tubes (PMTs) and avalanche photodiodes (APDs).
The PSMArray is an arrayed version of SensL’s novel silicon photomultiplier (SPM) pixels tiled together using flipchip on glass techniques. It operates from 400 to 850 nm.
A MATERIALS CHALLENGE
For higher-bandwidth devices that serve communications and computing functions, though, the greater challenge involves various compound semiconductor materials. Nonetheless, quite a few devices are spread throughout the market, spearheading record-breaking performance levels. Also, many other products are in development or on the cusp of introduction. Companies cite progress in producing higher-performance detectors, waveguides, modulators, lasers, switches, filters, couplers, multiplexers, amplifiers, and other optoelectronic functions. Device development mostly focuses on the use of silicon (Si), germanium (Ge), and indium phosphide (InP). The holy grail for high-bandwidth optoelectronics is to integrate these materials on a standard CMOS silicon process, monolithically or in hybrid form.
Such integration would reduce manufacturing, packaging, and testing costs. It also would increase reliability and improve performance to levels practical with present and future communications and computational needs. Whether it’s possible is unclear, since large-volume applications for such devices don’t exist as of yet.
ON THE MARKET
A recent report by the Optoelectronics Industry Development Association (OIDA) predicts robust growth for optoelectronic components. The group expects steady expansion of the market, both for optoelectronic components and the technologies they will enable, with the overall industry doubling to $1.2 trillion from 2007 to 2017 (Fig. 1).
One of the first complex silicon optoelectronics products to reach commercialization, a highly accurate, multichannel model 2200 dynamic-gain equalizer module from Silicon Light Machines, arrived in 2002. Based on MEMS and CMOS processing, it consists of the patented Grating Light Valve (GLV) circuit, a light engine, and an optical circulator.
An array of parallel aluminum-covered silicon-nitride (SiN2) microribbons is suspended above an air gap. The dynamic ribbons are then deflected up and down in response to an applied electrostatic voltage, acting as variable optical attenuators in response to a light signal. The module is being used in optical wave-division multiplexing (WDM) communications, high-resolution displays, and high-end high-definition video applications.
Manufacturing more complex and integrated optoelectronic devices on a standard process like CMOS, however, is much more challenging than making photodetectors. Silicon’s bandgap is too large, compared to other materials, to effectively operate at the infrared (IR) wavelengths used for optical communications and needed for optical modulator circuits.
The CMOS integration challenge hasn’t stopped researchers from forging ahead, though. Impressive developments have been achieved in the lab for all-silicon devices with wide bandwidths. Last year, Intel researchers announced the world’s fastest optical silicon modulator, writing 30 Gbits of data into a light beam per second.
Three years ago, Intel demonstrated a 10-Gbit/s all-silicon modulator. At one end of the modulator, light enters silicon diodes that split the light beam into two beams that pass through the diodes. Applying a voltage to the diodes shifts the light beams’ phase or position. This shifting encodes data into a binary form of ones and zeros.
The 30-Gbit/s data rate is close to the 40-Gbit/s rate used in modern communications systems. “By slightly altering the chemical makeup of the diodes, we can expect 40-Gbit/s data rates to be ready for commercialization by 2010,” says Intel research fellow Mario Paniccia, director of Intel’s Silicon Photonics Research Lab. “If you take 25 of these silicon lasers and direct them into an array of 25 modulators, you can have a terabit of information all on a piece of silicon the size of a fingernail,” adds Paniccia.
Researchers at IBM developed a photonic silicon modulator that’s 100 to 1000 times faster than previous comparable modulators, and they switch at 10 Gbits/s. The ultra-low-power RF power modulator uses Mach-Zendor modulation (MZM) interferometry and consists of 200-µm long p+-i-n+ active diode regions on an effective chip area of about 0.12 µm2.
Continue to page 2
More successful results for optoelectronic devices are being achieved by combining silicon with germanium and InP. Intel has fabricated germanium-on-silicon photodetectors that feature a 29.4-GHz bandwidth and 93% quantum efficiency (Fig. 2). The product of these two parameters, 27.3 GHz, translates into a meaningful measure of a photodetector’s merit.
According to Intel, these figures are the highest reported for a photodetector operating at a wavelength of 1550 nm. Yet the device still needs more gain to operate at the 40-GHz rates required by modern communications systems. Presently, the gain is only 16 dB. But Intel’s researchers believe that by packaging the photodetector with an impedance-matched, high-speed, transimpedance amplifier, they can reach the desired gain.
Similar germanium-on-silicon research is under way at the Massachusetts Institute of Technology. To get around the 4% lattice mismatch between germanium and silicon, researchers are using a hybrid approach.
A thin buffer of germanium is deposited on silicon using low-temperature chemical- vapor deposition (CVD) processing. Next, a thicker germanium layer is grown on top of this buffer using higher processing temperatures. Then the entire structure is annealed to reduce threading dislocations. So far, researchers have achieved an efficiency of 90% and responsivity of 1.08 A/W at a wavelength of 1550 nm.
Luxtera offers a promising solution in integrating pure germanium with silicon. The company, which spun off from the California Institute of Technology (Caltech), successfully applied small deposits of germanium onto a silicon wafer during fabrication, resulting in waveguide photodetectors with significantly greater performance when compared with other available photodetectors (Fig. 3).
This method also offers higher integration levels and is less costly to test and manufacture. Thousands of germanium photodiode particles can be added to a single silicon chip. All germanium growth and processing is performed before any electrical contacts are made.
“Our germanium photodetector capability allows us to meet future communications needs for applications like DVDs, highbandwidth low-cost video-conferencing, and high-definition multimedia interface (HDMI) systems and HDTVs,” says Marck Tlaka, Luxtera’s vice president of marketing.
Luxtera uses its CMOS photonics platform to manufacture the Blazar active optical cable (Fig. 4). Designed for highproductivity computing cluster and enterprise applications, it is the first cable to connect storage sites and switches at 40 Gbits/s bidirectionally (four channels at 10 Gbits/s each). Thin, flexible, lightweight, and rugged, this cable can span any distance from 1 to 300 m as well as two attached transceivers.
INDIUM PHOSPHIDE TO THE RESCUE
InP may be the right partner for silicon when it comes to integrating high-performance photonics IC devices, which can also be made relatively inexpensively on a CMOS process. Intel researchers recently built a hybrid device that uses InP for light generation and amplification. They bonded it to a silicon waveguide that forms the laser’s cavity and determines the laser’s performance (Fig. 5). The two materials are fused together via a 25-atom thick layer.
This work was based on Intel’s research into building silicon and germanium on silicon photonics devices. It was also facilitated by the University of California at Santa Barbara’s development of photonic ICs that are capable of 160-Gbit/s data transmissions, using advanced waferbonding techniques.
The critical issue is that the bonding does not require alignment of the InP material to the silicon waveguide chip. Alternative methods required such an alignment and have been very costly and impractical for high-volume production. This new development brings together the light-emitting capabilities of InP with the light-routing capabilities of silicon.
According to Intel, dozens of future integrated terabit hybrid silicon laser ICs can be built, each emitting light at a different wavelength. They can be coupled into dozens of silicon modulators, all multiplexed into a single fiber.
The benefits of InP haven’t been lost on the Center for Integrated Photonics. Last year, this U.K. group announced what it calls the first commercially available semiconductor erbium-doped optical amplifier (SOA) to offer breakthrough performance for an all-optical, 100-Gbit/s telecommunications network (Fig. 6).
Based on 1550-nm, InP multiple quantum-well devices, the SOA features a typical saturated gain recovery time of 10 ps and 20-dB gain with a 0.2-dB polarization-saturated gain. This nonlinear optoelectronics device uses very tiny lasers (a volume of about 0.1 mm3) with high gain of about 30 dB. Less than 100 fJ is needed to generate the desired nonlinear effects. The SOA chip measures 2000 by 500 by 150 µm.
“We’ve had a successful 40-Gbit/s SOA for more than two years and employ array versions to produce highly integrated 2R regenerators,” says David Smith, the Center’s chief technology officer. “The SOA gives the development community a platform to support 100-Gbit/s all-optical architectures.” (An optical regenerator performing the reshaping and re-amplifying functions is called a 2R regenerator.)
WHEN WILL IT HAPPEN?
We may be on the brink of an integrated optoelectronics reality, but some major hurdles still must be overcome in computer communications and telecommunications. Testing and packaging remain key challenges, though optical microchip substrates and motherboards have been suggested to make optical telecommunications more viable (Fig. 7).
Optoelectronic substrates will be needed to replace printed-circuit boards with copper interconnects. Jack Fisher, a consultant at Interconnect Technology Analysis Inc. and chairman of the organic interconnect chapter of the 2007 International Electronics Manufacturing Initiative (iNEMI) roadmap, believes that several enablers are needed to make optoelectronics a more mainstream technology.
These technologies include laminated and embedded waveguide interconnects for high-speed optical backplanes and chip-to-chip communications, improved vertical-cavity surface-emitting lasers (VCSELs), and outsourcing of manufacturing. Such developments will lead to wider dissemination of closely held package, assembly process, and test knowledge.
Fisher points out that despite steady progress in optoelectronics devices, continued improvements in less expensive copper technology have kept pace with circuit bandwidth needs. The crossover point where copper and optoelectronics interconnects would compete has yet to be determined, taking into consideration cost and performance issues, as verified by work performed at the Fraunhofer Institute for Reliability and Microintegration (IZM) in Germany.
Copper may not be able to keep pace with future gigabit and terabit communications demands, so optical communications may ultimately become the alternative. Optical communications methods are sometimes more economical over distances of 10 to 100 m. In the future, optical methods are likely to be used for shorter distances as the demand for higher data-rate transmissions continues to grow and optical communications costs decrease.
Intel’s Paniccia believes that silicon photonics modulators and lasers could be readily available within a couple of years. He anticipates optical data-communications rates beyond hundreds of gigabits per second and into the terahertz range. “Our goal is to build an integrated silicon photonic chip that can transmit and receive data at 1 Tbit/s, optically. This will radically change how future computing is done,” he says.