Data transport in the optical domain is believed today to be the most promising technology, or more likely the only viable one, to unchain the capacity of high speed data links without crashing into unsustainable energy consumption. Cooling is the main cause of power dissipation in data centers, so that a further increase in the speed of data transmission is hardly attainable without a dramatic energy-per-bit reduction. Since copper wires are close to their transmission limit, in this game electric connections are likely to be offside; as a valuable alternative, photonics offers a unique opportunity to become the forefront technology for high speed interconnects in the near future. In order for communication between data servers, modules, boards, or even processors to transfer into the optical domain, the integration of photonic transmission systems on conventional electronic platforms becomes a primary issue.
This motivation has strongly boosted the search for CMOS-compatible optical technologies that could coexist with electronics within the same silicon chip. Optical waveguides have been the first building blocks to be realized on a silicon platform, and today they combine very low loss and strong light confinement, enabling an impressive miniaturization of photonic integrated circuits. Little by little, silicon-based platforms have been enriched with a wide variety of passive and active devices, including for instance high-speed modulators and detectors.
In this scenario, light generation has been always considered the missing piece of the jigsaw, since the indirect bandgap of silicon, similarly to other group IV materials, naturally provides poor radiation efficiency. A successful approach to make group IV lasers directly on silicon exploits the engineering of strain and doping concentration, and has been recently employed to realize the first electrically pumped Ge laser monolithically integrated into a CMOS process. Direct gap III-V semiconductor lasers are typically much more efficient, but all efforts to monolithically integrate them on a silicon platform have had very limited success. The main technological challenges here are the mismatches of both lattice constant and thermal expansion coefficient, and the high growth temperature of III-V bulk materials, which cannot be tolerated by CMOS chips.
Before the work by C. Chang-Hasnain and coworkers, heterogeneous integration was considered the only chance to place III-V lasers onto a silicon substrate, this approach suffering yet from a relatively low yield due to the need for an atomic level of surface flatness. Now we know that there is a way to grow III-V lasers monolithically onto silicon, without causing any functional damage to the underlying electronic circuitry. The secret in the pocket of the Chang-Hasnain’s group is low temperature (410 °C) growth of indium gallium arsenide (InGaAs) nanopillar lasers using metal-organic chemical vapor deposition (MOCVD). And what a better demonstration than growing a laser directly on top of a metal-oxide-semiconductor field effect transistor (MOSFET)? Not only did they do it, but they also demonstrated that III-V nanopillars can be grown both on the crystalline (100)-silicon of the source/drain region and on the polycrystalline silicon of the gate region with the same structure, shape, growth rate and crystalline phase.
The nanopillar laser consists of a 630-nm-diameter InGaAs active region surrounded by a 110-nm-thick GaAs passivation layer. Room temperature emission at a wavelength around 980 nm and with an average output power of 0.5 μW was achieved by using an optical pulsed pump at 765 nm, with a threshold power of 600 μW. Remarkably, MOSFET functionality was found to be the same before and after the nanopillar laser growth in more than 98% of the fabricated devices, demonstrating that group III-V materials can be effectively and reliably grown with no adverse effect on silicon transistors.
Laser sources on the head of silicon transistors… is there anything else that we can ask for? At the top of their dreams’ list, most people would probably place electrical pumping, which would allow nanopillar lasers to make a dramatic qualitative leap towards dense integration into sophisticated CMOS architectures. Further, shifting the laser emission at wavelengths above 1 micron would allow the use of silicon itself as a waveguiding material for on-chip light transport. And probably many other technological breakthroughs will be required in the future. The way towards full integration of photonics and electronics onto the same chip seems still quite long, but another fundamental milestone along this path has been set.
You must log in to add comments.