Groovy Nitrides Could Light Up Silicon
The silicon industry dwarfs our own but it does have a few weaknesses. These include silicon s inability to emit light with any great efficiency, which rules out the construction of lasers and optoelectronic ICs built from only this material. But such circuits are of great importance as they could offer a solution to overcoming the data-transfer bottlenecks in ICs and help to cut fiber-to-the-home deployment costs.
The benefits of equipping silicon ICs with a laser source have already motivated a tremendous amount of research, including attempts to build GaAs- and InP-based designs on silicon substrates. However, producing monolithic chips from these combinations of materials is incredibly challenging because the significant differences in lattice constant lead to a high density of misfit dislocations that act as non-radiative recombination centers. This has prevented the fabrication of any optoelectronic devices using these material systems that can deliver long-term stability, and it has driven researchers to look for possible alternative solutions.
Two approaches that have generated considerable interest are erbium-doped silicon and low-dimensional silicon systems, which include porous silicon, silicon nanocrystals and nanopillars, and specific silicon-interface structures. Remarkable progress has been made in the basic understanding of the light-emission process of these systems, including optical gain in silicon nanocrystals, but nobody has been able to build a laser.
Greater success has come through the study of nonlinear optical processes, such as stimulated Raman scattering in a silicon-based waveguide structure. In 2005, Intel demonstrated that it is possible to produce a continuous-wave silicon Raman laser in a specifically designed reverse-biased p-i-n diode embedded in a silicon waveguide. However, this technique still needs an efficient laser source to pump and to initiate the nonlinear process. This obstacle may have encouraged Intel to develop other approaches because it recently announced a technology that involves the bonding of small InP laser chips onto silicon substrates.
At the University of Marburg, Germany, we have been developing a new process that promises to unite III-Vs and silicon monolithically via the novel material system GaNAsP. This dilute nitride can be grown epitaxially and lattice-matched to silicon by using the nitrogen content to tune the lattice constant of the compound to the underlying substrate.
We have laid the groundwork for this technology by growing a variety of dilute nitride laser structures on an Aixtron AIX 200-GFR tool designed for research and development. Film growth was performed using a reduced reactor pressure of 0.05 atm, hydrogen as the carrier gas, and the metal-organic precursors triethylgallium, tertiarybutylarsine, tertiarybutylphosphine, 1,1-dimethylhydrazine and triethylboron. We chose these because they efficiently decompose at 575 °C – a relatively low growth temperature that is needed to produce dilute nitrides with significant amounts of nitrogen. (Arsine, phosphine and ammonia – the more common group-V sources for MOCVD growth – hardly decompose at this temperature and would lead to a very inefficient growth process.) After growth, our epiwafers are annealed in the reactor for an hour at 750 °C. This improves the optoelectronic properties of the active region without compromising structural integrity.
Our GaNAsP/GaP multiple quantum-well heterostructure s (MQWH) high crystal quality is revealed by X-ray diffraction spectra and by transmission electron microscopy images (figure 1). These don t provide any evidence for either dislocation formation or any related inhomogeneous strain relaxation in the MQWH. We have also studied our active region s direct-bandgap characteristics by using photoluminescence-based techniques, which show excellent agreement with our theoretical model (figure 2).
To allow us to assess the capability of our GaNAsP/GaP-MQWH for various optical devices, we have also grown a range of structures on GaP substrates that integrate this active region in AlGaP/GaP-separate confinement heterostructures. GaP substrates can easily be cleaved to form simple cavities and, because they have a similar lattice constant to silicon, the process is compatible. Silicon substrates, in comparison, are harder to cleave and laser facets have to be formed by more complex dry chemical etching processes.
We began our device development by studying the optical gain in simple structures that can cover the 850–950 nm wavelength range used for data communication applications. Measurements of the modal gain curve revealed that our structure s spectral width and peak modal gain were similar to those produced by standard III-V material systems.
These promising results have encouraged us to fabricate broad-area devices that can deliver room-temperature laser emission when driven in pulsed mode. The typical threshold current density for these pulsed devices is 40 kA/cm2, which is very high. However, these are early-stage results and the threshold current will come down as we optimize the growth and annealing conditions for our structure, and improve the optical and electrical confinement of our waveguide.
We believe that the proof of concept provided by these simple laser structures, which are built on GaP substrates, underlines the enormous potential for applications integrating III/V-based optoelectronics with silicon-based microelectronics. We are now starting to work towards this goal by developing a technology for the full monolithic integration of our GaNAsP material system with silicon microelectronics. Such a process must be as compatible as possible with existing standard CMOS-process technology so that it delivers the full benefits of integration. This means that it must be designed with the most common form of silicon used by this industry – (100) substrates.
CMOS processing demands the exactly orientated form of this substrate, which has typically been overlooked by many research groups developing heteroepitaxial processes for the growth of III-Vs on silicon. Instead, these groups have used off-axis wafers that improve material quality and feature a series of atomic steps. However, with our proprietary process, we can deliver the benefits of an off-orientated substrate from an exactly orientated form of this platform (see box "Double stepping on a flat surface").
The high-quality buffer layers produced by this approach have been used as the basis for growing our GaNAsP-based laser structures, which have excellent crystal quality according to high-resolution X-ray diffraction measurements (figure 3).
We plan to deposit this laser structure in recess stripes of silicon substrates, which are formed by selective-area growth (figure 4). Such an approach guarantees the planarity of the wafer for subsequent process steps and is compatible with a 750 °C post-growth anneal that improves the structure s optoelectronic properties. This anneal would take place before CMOS processes are carried out that cannot stand such high temperatures.
Once the laser has been deposited, its sides will be passivated with SiOx and SiNy, and high-reflection and anti-reflection coatings applied to the device s end facets, the latter formed by dry-etching processes. Heavily doped p-type and n-type silicon layers can then be used as contacts, thanks to the use of lattice-matched material for the laser. This avoids all of the problems associated with the use of non-CMOS-compatible metallization schemes, and it also "hides" the III-V material, thus allowing CMOS processing without any threat of cross-contamination.
Our progress has enabled us to secure €78.5 million ($12.3 million) to continue to develop this technology, through funding from Germany s federal ministry of education and research High Technology Initiative program. This three-year venture, called MonoLaSi, will run until 2010 and has two main aims: to fabricate prototype monolithic continuous-wave ridge waveguide lasers on silicon for data communication applications, and to develop a bespoke growth tool.
The partners involved in this project come from industry, publicly funded research institutions and academia. They are led by NAsP III/V, our spin-off from the University of Marburg. NAsP III/V will carry out the MOCVD deposition of the monolithic-ridge waveguide laser structures on silicon wafers with recessed stripes. Aixtron will develop a growth tool capable of depositing III-Vs and silicon on 300 mm wafers, Dockweiler Chemicals will provide specifically purified group-V precursors and Osram Opto Semiconductors will offer expertise in qualifying the technology and laser manufacture.
The Fraunhofer Institute for Applied Solid-State Physics, Freiburg, will perform processing, develop chip technology and characterize the lasers, which includes lifetime testing. Among the players from academia, the Ruhr-University Bochum will carry out experimental characterization of the gain and the laser properties of our dilute nitrides, and Philipps-University Marburg will be analyzing the MOCVD growth process and performing first-principle calculations of the gain and lasing characteristics.
We believe that our monolithic III/Vs-on-silicon laser will be a key component for delivering future on-chip, chip-to-chip and/or backplane optical interconnects. However, this is not the only sector that will benefit from such a device. This technology promises to deliver low-cost devices for fiber-to-the-home connectivity and terrestrial high-efficiency III/V-multijunction solar-cell stacks on germanium and silicon substrates, as well as the development of novel n-channel III/V-layers for CMOS-transistor devices. Our consortium will provide research and development results and technology processes for all of these applications, as well as the production tool for the manufacture of all of these types of device.