Germanium as the unifying material for silicon photonics
In recent years, high-performance active devices based on germanium (Ge) have been developed for monolithic integration into silicon (Si)-based photonic systems. Most of this progress is based on the development of epitaxial Ge growth directly on Si.1, 2 The initial Ge-based devices were photodetectors using the high absorption coefficient of Ge up to wavelengths of about 1550nm. These devices have shown larger than 90% internal quantum efficiencies at 1550nm and bandwidth above 30GHz.3, 4 Today, monolithically integrated Ge photodetectors are ubiquitous and can be found in active optical cables or fully integrated with transimpedance amplifiers for telecoms applications.
The next novel Ge-based device was an electro-absorption modulator that could be based either on the capacitive electro-optical quantum-confined Stark effect5 or the Franz-Keldysh effect in Ge.6 The best performance for a Franz-Keldysh-based Ge modulator is 30GHz bandwidth7 without resorting to resonant enhancement as required in plasma dispersion-based Si microring modulators. Both Ge detectors and modulators are integrated with a Si waveguide, resulting in very low capacitance (of the order of a few femtofarads). If directly integrated with the analog circuit (driver or transimpedance amplifier), power consumption is reduced to a few tens of femtojoules per bit. That should be compared with consumption of picojoules per bit in conventional components based on wire bonding.
We recently demonstrated a Ge-based device, an electrically pumped laser that can be fully integrated into a Si CMOS process.8, 9 The lasing in Ge is enabled by strain and n-type doping to allow dominant direct bandgap recombination in an indirect semiconductor.
The Ge photodetector is the most advanced Ge-based photonic device and can already be found integrated with Si CMOS. Integration of Ge modulators will most likely follow soon. The Ge laser is the least advanced device, and more development is necessary to increase its reliability and enable full CMOS integration.
Integrating all Ge devices into one photonic circuit requires designing a process flow for fabricating the system. The biggest challenge is to determine the material composition and doping. Strained Ge detectors (0.25%) work best for wavelengths below 1560nm due to the direct band edge onset at that range. Ge modulators based on the Franz-Keldysh effect will perform best in the same wavelength range when adding 0.7% Si in Ge.6 The direct band edge of the materials would be at about 1480nm. A detector with this particular composition will not be very efficient at absorbing light at 1550nm. While Ge will do so within 5μm, the absorption length for 0.7% Si in Ge is more than five times longer than for 100% Ge. In a waveguide configuration, the path length of the detector's light can be extended without a significant performance penalty, mainly because the capacitance is still very low due the device's small size. The benefit of a detector waveguide configuration has already been demonstrated.3
We grew Ge detectors and modulators in silicon oxide trenches with a width of 500nm and lengths between 5 and 30μm using ultra-high vacuum chemical vapor deposition. We added 0.7% Si to the Ge during growth to tune the modulator response to 1550nm. The Ge devices were fabricated after the silicide formation but before the back-end-of-line processing.10 We found that limiting the operation range to about 15nm enables use of the same GeSi composition for both modulators and detectors with a waveguide configuration. Fabrication is thus achieved in a single Ge process sequence, limiting the thermal budget and simplifying process flow. Figure 1 shows the wavelength performance range we obtained for modulators and detectors using different material compositions.
Ge detectors and modulators rely on undoped Ge, but Ge lasers need high n-type doping for gain. As long as high n-type Ge can only be grown epitaxially (i.e., in layers), simultaneous Ge growth for detectors, modulators, and lasers is not possible. However, the thermal budget for Ge lasers is significantly lower than that for Ge detectors and modulators due to the high diffusivity of n-type dopants in Ge, requiring low process temperatures to prevent out-diffusion.11 Consequently, it is possible to fabricate Ge lasers after the Ge detectors and modulators without decreasing their performance. Figure 1 also shows the measured gain spectrum for n-type Ge with a doping level of 4×1019cm−3. The spectrum spreads over nearly 200nm and covers the performance range of all GeSi compositions for detectors and modulators.
In summary, using Ge as a material for the most essential active photonic components enables development of an integrated, CMOS-compatible process flow without adding new materials or processes that reduce the yield and reliability of the final photonic and electronic systems. We have shown how to fully integrate source, modulation, detection, and electrical networks in a CMOS-compatible process in which all packaging costs (typically laser, electrical components, and wiring) are eliminated. The reduced cost, together with energy-efficient design, shows that photonic interconnection is a possible solution to poor scaling of on-chip wires and I/O bandwidth density for future technology nodes and promises manufacturability of very large volume applications.
Our initial feasibility demonstration is a photonic link between a Ge laser and a Ge photodetector. Moving forward, we will add active and passive photonic devices to show a fully monolithically integrated link with increasing complexity.
Jurgen Michel is a senior research scientist in MIT's Microphotonics Center. He was formerly at AT&T Bell Laboratories. He earned his diploma at the University of Cologne and his doctorate and habilitation at the University of Paderborn. He has co-authored more than 200 refereed scientific papers.
Marco Romagnoli is a visiting scientist at MIT, developing an optically integrated multiprocessor chip, and director of the Boston operations of PhotonIC Corp. He has more than 25 years of experience in the research field, especially in the area of photonic technologies. He holds a physics degree from the University of Rome (La Sapienza). He is the author of more than 160 journal papers and conference contributions, and is an inventor on more than 40 patents. He is on the technical committee of major photonics conferences and has served as expert evaluator for the European Commission's 6th Framework Programme. He won the Philip Morris prize for optical innovation in 1994.
