posted on 2015-04-17, 00:00authored byChad A. Stephenson
Fifty years ago, Gordon Moore formulated his well-known prediction that transistor count would double every two years. This prediction has set a benchmark for semiconductor processor manufacturers to achieve and driven them to aggressively reduce transistor dimensions with every succeeding generation. However, as transistors approach their fundamental limits, major advances must be made in other areas of processor chips to continue the increase in computing performance. Currently, one of the biggest bottlenecks is the interconnect bandwidth, or the ability to pass data quickly between chips or among the multiple cores within each processor. The telecommunications industry was revolutionized by the advent of optical communications over optical fibers, replacing copper wires. A similar shift from electrical wires to optical interconnects could increase chip-level interconnect bandwidth by one to two orders of magnitude and allow continued performance enhancements. However, dense optical interconnects require an efficient, intimately- ii integrated laser on Si, something that does not yet exist. This work highlights two routes to achieving optical interconnects on Si: tensile-strained Ge and dilute Ge1-xCx alloys. While Ge does not efficiently emit light due to its indirect bandgap, tensile- strained Ge can be made to be a direct bandgap semiconductor that emits orders of magnitude more efficiently than Si. In addition, COMSOL strain simulations of Ge waveguides with Si3N4 stress liners show that sufficient strain to change the refractive index and allow optical mode confinement within a strained Ge waveguide is achievable within conventional wafer fabrication techniques. Similarly, Ge1-xCx alloys may offer a direct bandgap according to earlier simple models, but published results have been ambiguous. This work offers definitive support of a strongly direct bandgap based on both a band anticrossing (BAC) model and highly accurate density functional theory (DFT) using hybrid functional techniques, as well as experimental growth of Ge0.998C0.002 using gas source molecular beam epitaxy that supports our computational model. The splitting of the conduction band predicted by our BAC and HSE models is confirmed by a 50 meV band splitting by photoreflectance measurements. Our models further predict the onset of a direct bandgap for carbon concentrations of just 0.8% C.