Tube Lasers Prepare To Light Up Silicon Circuits
Silicon has a major weakness. It’s a lousy light emitter, and this means that it cannot be used to build monolithic ICs incorporating lasers or LEDs. So to address this Achilles heel, many researchers are trying to find ways to unify compound semiconductor light sources with silicon. If a cost-effective technology for high-volume manufacturing could be found, then many would welcome it. The electrical interconnects that are currently being used for chip-to-chip data transfer have very little headroom left, and switching to an optical approach would alleviate a looming bottleneck. In addition, silicon ICs with opto-electronic functionality could spur developments in biological sensing applications. Efforts at developing III-V-on-silicon light sources go back a long way and progress has been hampered by several major differences between the two types of material. In fact, it might actually make more sense to keep them slightly apart, rather than putting these unhappy bedfellows together.
One researcher holding precisely that view is Zetian Mi from McGill University, Montreal. He is leading a team that’s developing a free-standing tube laser that sits a few hundred nanometers above the silicon platform. This GaAs-based novel architecture promises to provide the circuit with a light source operating at incredibly fast modulation speeds, while consuming very little power.
Mi wants to keep the III-Vs and silicon apart due to their incompatibility – these materials have significant differences in polarity, lattice constant and thermal coefficients of expansion. “When you grow GaAs on silicon, in some regions the gallium atoms attach to silicon first, and in other regions arsenic attaches to silicon first. So as you grow more and more layers there will be a boundary, called the anti-phase domain boundary."
This interface, which stems from differences in polarity, creates a high density of dislocations in the material, leading to a leakage current path that causes local heating. These problems are compounded by defects that stem from strain in the epilayers caused by lattice mismatch. Some researchers, including those from Intel, sidestep these problems by growing lasers on a native platform, and then bonding their devices to silicon. But this approach cannot eliminate stresses that are caused by thermal expansion coefficient differences. What’s more, this type of laser cannot be scaled to the sub-micron sizes needed for realizing ultra-low power consumption alongside modulation speeds of hundreds of GHz.
Mi’s approach, which promises to deliver on all these fronts, begins by taking a GaAs substrate and depositing a strained structure with an active region onto it. The epitaxial layers are patterned into a U-shape, before a sacrificial layer is removed. The strain in the remaining structure causes it to roll into a tube, which can then be transferred to silicon with the aid of a solvent (see Figs. 1 and 2).
Mi did not invent micro-tube technology – this accolade goes to Victor Prinz from the Institute of Semiconductor Physics (Siberian Branch) at the Russian Academy of Sciences. He developed the technology for producing a micro-tube structure in the late 1990s and reported his results in 2000.
Since then trailblazing has transferred to Germany. Efforts by Oliver Schmidt and co-workers at the Max Planck Institute for Solid State Research in Stuttgart, along with Tobias Kipp’s team at the University of Hamburg have led to an improved understanding of the formation of these structures, and observations of the optical emission from GaAs-based devices at very low temperatures and SiO2/silicon heterostructures at room temperature. Mi’s two recent, major contributions to this field are the demonstration of coherent emission - and ultimately lasing - from III-V micro-tubes; and the development of a process to transfer these structures onto a silicon substrate.
The McGill academic employs MBE for the growth of his epiwafers, which comprise a 50 nm-thick AlAs sacrificial layer, a 20 nm thick In0.18Ga0.82As layer, and a 30 nm thick GaAs cap that incorporates two In0.5Ga0.5 As quantum dots layers. Some of these structures were grown in Pallab Bhattacharya’s group at the University of Michigan, but Mi has plans to move production in-house after his team has conditioned a recently purchased MBE tool that is dedicated to arsenide deposition.
Mi says that his active region contains quantum dots, rather than quantum wells, because they offer superior carrier confinement. “The resulting near-discrete density of states promises both large gain and large differential gain for laser operation, compared to quantum wells." In addition, the dots provide strong carrier localization that greatly reduces non-radiative recombination associated with surface defects.
A U-shaped GaAs-based structure is defined by photolithography and subsequent etching into the InGaAs layer. Self-rolling of these epilayers is then initiated by selective etching of the AlAs sacrificial layer with hydrofluoric acid. This eventually leads to the formation of “fully released" quantum dot micro-tube structures on a GaAs substrate.
Other groups have already developed methods for the transfer of III-V devices onto alternative substrates, but Mi says that these processes are incompatible with his micro-tubes. “The reason that we don’t use dry printing is that our structures are hollow and fragile, and you cannot press on the surface. Solution casting works very well, but you can’t position these micro-tubes where you want to."
Instead, the micro-tubes are directly transferred from the GaAs substrate to a silicon platform with the aid of a solvent. According to Mi, removal of the GaAs substrate creates freestanding micro-tubes that preferentially stay on silicon due to the gravitational force induced by the solvent in and around the tube. Solvent evaporation leaves the micro-tubes bound to silicon via a Van De Waals attraction. Mi believes that his micro-tube laser fabrication process has several strengths. “You combine the benefits of top-down and the bottom-up up processes. For example, the tube diameter is determined by layer composition, and the wall thickness by the etching process." The fabrication process is also simple and controllable, says Mi, and it can produce incredibly smooth surfaces with a roughness that’s determined by MBE growth.
From emission to lasing
Mi has made rapid progress since he started developing micro-tube lasers following an appointment at McGill’s Department of Electrical and Computer Engineering in September 2007. Optically pumped emission from his microcavities at 77K was realized in late 2008, and this year he has progressed to coherent emission at room temperature and finally lasing from these structures. Advances are the fruits of processing improvements resulting from ever greater familiarity with device fabrication.
The emission wavelengths produced by these tubes are governed by the lateral dimensions of the waveguide, and its associated whispering-gallery modes. This leads to multiple emission modes. The dominant lasing wavelengths produced by one of the most recent structures occur at 1194 nm, 1217 nm and 1241 nm, and there are also weaker, subsidiary peaks (see Fig 3). If Mi’s micro-tubes were perfect ring resonators, emission would be produced from every point in the tube. But these tubes have a starting and stopping edge that scatter light, and this dominants the emission because the rest of the structure is incredibly smooth. Light that emanates from these points rapidly diverges.
One of the promising aspects of Mi’s approach is its very high yield. This is partly thanks to the simplicity of the process, which involves just one photolithographic step. Emission from the tubes may not be in the preferred direction, but this can be remedied by simply rotating them.
If these micro-tube lasers are to kick-on and enjoy commercial success, then they will have to be driven by an electrical source. “It is difficult to achieve high efficiency, electrically injected micro and nano-scale lasers, and making an electrical contact directly onto the free-standing laser without adversely affecting device performance is particularly challenging," admits Mi. “However, we have devised an approach and expect to achieve electrically injected devices within one year." The spectral emission profile from these electrically pumped lasers is expected to feature multiple-emission lines that are already seen in optically pumped counterparts. “For some applications you want a single wavelength, but for other applications, such as wavelength division multiplexing, you need multiple, evenly spaced channels", explains Mi. “One of the objectives of our research is to develop these multiple wavelength tube lasers."
Modulation speeds for these lasers are expected to be well over 40 GHz, and Mi believes that this can be increased to the terahertz range by scaling devices so that all of their dimensions are less than a micron. But he admits that realizing this goal will be particularly tough. Another aspect of his technology that he is looking to advance is the precision with which the micro-tube is positioned on the silicon surface.
The existing substrate-to-substrate transfer process cannot guarantee that the laser will be successfully coupled to a waveguide on the silicon IC. But Mi’s team is addressing this issue by developing a process to move the tubes on the silicon surface. And if the Van de Waals forces are not strong enough to lock them in position for a commercial product, then they can be bonded in place with passivation techniques.
Mi and his team are clearly identifying and eliminating the barriers to commercializing their technology. Patent application filing is already underway through McGill University, and they have also initiated a collaboration with Reflex Photonics, a Montreal-based start up that is developing high-speed optical connections for semiconductor packaging and data transfer.
So it may be that it is a novel, ring-based approach that will hold the key to a long and happy marriage between silicon ICs and III-V lasers.