Researchers target imaging systems and optical networks

The nitride community is devoting a great deal of effort to improving LED efficacies because this could spur the growth of solid-state lighting. Similarly, SiC researchers are unified by their goal – more efficient power devices that can ultimately cut carbon footprints. But if you attend an InP meeting, you ll fail to pick up on a common mantra.

The reason behind this is not down to any weakness of InP but is actually a result of its strength. That s because InP is an incredibly versatile material, and its inherent properties have positioned it as the key material for telecom-wavelength lasers, modulators and detectors, and the best III-V for making high-speed transistors for imaging systems.

Both of these themes came out in force at the 20th international conference on Indium Phosphide and Related Materials (IPRM), which was hosted in beautiful Versailles, on the outskirts of Paris.

InP-based HEMTs have been grabbing the headlines since the International Electron Devices Meeting (IEDM) in Washington last December, due to the claim for the first terahertz transistor by Richard Lai s group at Northrop Grumman Space Technology.

These HEMTs provided the building blocks for the first low-noise amplifier (LNA) operating at 340 GHz. An amplifier operating at this high frequency improves the spatial resolution of millimeter-wave imaging systems and can tighten airport security by identifying concealed weapons.

Although nobody eclipsed Northrop Grumman s terahertz claim at the IPRM meeting, it is clear that many other players are catching up fast. IAF Fraunhofer in Germany, for example, has now scaled its process down to 35 nm gate lengths.

Arnulf Leuther revealed that this has increased its HEMT s fmax to more than 700 GHz. This device s fT is 515 GHz and it produces a maximum transconductance of 2500 mS/mm. The drain current peaks at 1600 mA/mm and is half of this value when at maximum transconductance.

The IAF group s HEMT is very similar in design to Northrop Grumman s, although the transistor is grown on a GaAs substrate and features a metamorphic InGaAlAs buffer. This raises the question of why the IAF transistor has not hit the terahertz regime.

This was discussed when the floor was opened for questions, but neither Leuther nor Lai could provide any definitive reasons for the speed difference. However, it was noted that the cross-sectional area of the gate plays a key role in determining fmax.

Leuther went on to describe the performance of a 0.15 mm2 single-stage LNA with a 2 × 10 µm metamorphic HEMT that has a 35 nm gate length. A gain of 2.5 dB was produced at 320 GHz – the upper frequency limit for this measurement – and hit 7 dB at 270 GHz. The transistor actually delivers 9 dB of gain at 270 GHz, according to the researchers, with losses occurring due to passive circuit elements.

Fujitsu s Masaru Sato gave an invited talk in the same session and described efforts to address real-time imaging at 94 GHz. He identified two essential requirements: arrays of detectors rather than single receivers; and LNAs that deliver 30 dB of gain with a noise figure of less than 5.5 dB.

The Japanese company has targeted these LNA specifications with an InP HEMT technology featuring 130 nm T-shaped gates on 100 µm thick InP substrates. These deliver 1500 mS/mm with an fT of 220 GHz.

Sato told the audience that high-gain amplifiers can be plagued by instability, due to feedback that travels through the substrate. To combat this, Fujitsu researchers simulated the performances of devices employing thin-film microstrip lines and inverted microstrip lines in conjunction with a flip-chip geometry.

Calculations revealed that the former design was unsuitable because feedback at 70–120 GHz could hit –20 dB. However, the latter design could suppress feedback to less than –40 dB. The team then used the superior design in a seven-stage 2.5 mm × 1.2 mm LNA chip that provided more than 35 dB of linear gain between 90 and 110 GHz. Noise was just 4–4.5 dB.

Sato s team went on to build a passive millimeter-wave imager with this LNA, by incorporating a detector MMIC, a linearly tapered slot antenna and a 20 cm diameter polyethylene lens. This system has captured several images, including one of a person behind 10 mm thick plywood.

Although Northrop Grumman representatives didn t announce any advances on the results presented at IEDM, they did disclose more details about their record-breaking transistors. When Compound Semiconductor interviewed Lai earlier this year, he was coy regarding the details of the high-mobility InAs channel that helped to boost transistor speed.

But at IPRM, Lai s colleague, Michael Lange, revealed that the channel comprised an InAs film sandwiched between In0.53Ga0.47As layers. The team investigated what effect the thickness of the bottom InGaAs cladding layer had on HEMT performance and found that a thicker layer can produce a higher fmax.

IPRM 2008 also focused on devices for optical networks, with technical sessions devoted to lasers, integrated devices, modulators and detectors.

Ute Troppenz from the Heinrich Hertz Institute (HHI) in Berlin, Germany, kicked off the laser session with an invited talk on 40 Gbit/s directly modulated passive feedback lasers operating at 1.55 µm. She explained that these operate in a "self-pulsation" regime that allows fast and flexible locking to an external optical data stream.

The HHI edge emitter comprises a distributed feedback (DFB) laser and a passive section, and it is fabricated with the same processing steps that are used to make a conventional laser. The 250 µm long DFB section features a ridge waveguide and an InGaAsP multiple quantum well active region. An identical length is used for the passive section. This doubles the size of the laser but has minimal impact on the technological demands, according to the researchers.

Trials at 40 Gbit/s revealed that HHI s laser is capable of successful transmission over 2 km fiber links. This makes it a promising device for cutting costs in very short networks, such as Ethernet, thanks to the elimination of separate modulators.

Another directly modulated laser design – 1.55 µm buried heterostructure lasers – was the subject of a presentation from Ian Lealman from the Center for Integrated Photonics (CIP), Ipswich, UK.

Iron has traditionally been used as a dopant in these devices, but this tends to interdiffuse with zinc during MOCVD growth. CIP has been hoping to overcome this problem by developing a rubidium alternative, through a program involving SAFC Hitech and the University of Surrey.

This UK-based partnership is not the first to turn to rubidium-based precursors, but previous efforts have been restricted by low-volatility sources, which make it difficult to incorporate rubidium in the epilayers. This team has enjoyed success by switching from linear molecules to a ring-shaped cyclic organometallic, which has opened the door to atmospheric pressure MOCVD growth.

Lasers that are 0.35 mm long have been produced with the new precursors, which feature nine InGaAsP quantum wells. Testing revealed similar threshold and slope efficiencies to iron-doped alternatives. Initial reliability tests are encouraging and show no degradation in threshold current or light output after 5700 hours of operation at 85 °C.

Hirosha Yasaka from NTT Photonics Laboratory, Japan, gave an invited talk on the development of a superior waveguide structure for high-speed Mach–Zehnder (MZ) modulators. Improvements were driven by a switch from a PIN structure to a NIN design. Removal of the p-doped layer reduces absorption losses and increases operating speeds.

The modulators were grown on semi-insulating InP substrates and each featured a 0.2 µm thick InGaAlAs/InAlAs multiple quantum well core layer sandwiched by 50 nm thick undoped InGaAsP layers. InP n-type layers clad this structure, and wafer processing produced 40 Gbit/s modulators on 4.5 × 0.8 mm chips that cover 1530–1565 nm. "We ve confirmed error-free modulation for all wavelengths," remarked Yasaka.

He and his co-workers have also developed chips capable of 80 Gbit/s modulation, which employ differential quadrature phase-shift keying (DQPSK). This method is currently enjoying a great deal of popularity thanks to its high spectral efficiency and robustness to dispersion.

Modulators for DQPSK require two MZ modulators, which act as phase modulators, alongside an optical "π/2" phase shifter and an optical combiner. NTT s engineers have built a 7.5 × 1.3 mm chip incorporating all of these elements, which features an NPIN structure with a thin p-doped layer.

This additional layer takes on the role of the semi-insulating layer in a NIN structure and should enable monolithic integration with semiconductor active devices, such as laser diodes and photodetectors.

Testing the MZ modulator revealed a minimum fiber-to-fiber loss of 13 dB, which includes a 3.5 dB/facet loss associated with optical coupling between the chip and the lensed fiber. Extinction ratios exceed 18 dB and a clear opening in the eye diagrams reveals successful transmission at 80 Gbit/s.

The lasers and modulators session included a talk by Anne Rouvié from the Alcatel-Thales III-V lab on the effects of carbon doping in AlInAs avalanche photodiodes (APDs). Rouvié says that these detectors are candidates for deployment in optical telecommunication networks thanks to a signal-to-noise ratio that beats PIN diodes by more than 10 dB.

MOCVD-grown APDs frequently use zinc dopants for the growth of the thin (50 nm), highly doped charge layer. Zinc is not ideal, however, because it diffuses into the surrounding material, so Rouvié and co-workers have developed a carbon-doping process for the growth of this critical layer.

Measurements on the carbon-doped APDs reveal that they are superior to zinc-doped devices. Dark current at a multiplication factor of 10 is just 17 nA – nearly three orders of magnitude less than the zinc-doped ones – while the gain-bandwidth product is 140 GHz and responsivity at 1.55 µm is 0.9 A/W.

Integrating different types of device into a larger InP chip was the theme of Ben Yoo s invited talk. His group at the University of California, Davis, is following a similar path to that of Infinera, with the goal of reducing the need for a large number of discrete components.

Yoo believes that cost savings can come through this approach, thanks to a reduction in overall material costs. However, tough challenges are also faced, in terms of yield, reliability, isolation, and the management of high-power densities, heat dissipation and cross-talk.

Yoo s team has produced a variety of large InP chips that employ optical code division multiple access, a technology for local area networks that is more than 20 years old but is unsuitable for discrete components, he says.

These chips include a 16.8 × 11.4 mm design operating at 1.55 µm, which features mode-locked lasers, arrayed waveguide gratings, MZ modulators, and amplitude and phase modulators. They can operate over 128 channels that are spaced by 10 GHz.

The interesting talks at IPRM 2008 are testament to the fact that InP research is alive and kicking. Diversity may mean that this community is not united by a common goal, but they will continue to extol the virtues of this material at the next meeting, to be held in spring 2009 at Newport Beach, CA.   

View pdf of article