image: (a) Schematic diagram of the heteroepitaxial laser integration on Si platform. (b) Schematic diagram of the pattern formation process before III-V deposition. (c) As-patterned 300 mm Si wafer. view more
Credit: by Chen Shang, Kaiyin Feng, Eamonn T. Hughes, Andrew Clark, Mukul Debnath, Rosalyn Koscica, Gerald Leak, Joshua Herman, David Harame, Peter Ludewig, Yating Wan, and John E. Bowers
Internet data traffic has seen a compound annual growth rate of 27% in the last few years and exceeded one zettabyte in 2017. Electrical connections between the high-bandwidth chips such as compute and switch processors and between EPGAs suffer from bandwidth density and power consumption limitations. To fulfill this soaring demand for higher data traffic, optical interconnects with Si photonics offers orders of magnitude improvements in bandwidth densities and greatly improved long distance efficiencies in terms of energy/bit. However, despite the fact that numerous successes in high performance passive components have been demonstrated at 300 mm scale, integrating III-V lasers is a vital step toward widespread adoption due to the indirect bandgap nature of Si and Ge. Monolithic integration is the most economical approach and offers the highest integration density. Yet, merging the dissimilar III-V and Si via direct epitaxial growth generates inevitable crystalline defects. Previously, with novel defect management epi layer designs and leveraging the defect insensitivity of QDs, scientists at UC Santa Barbara has previously demonstrated high performance and long lasting QD lasers grown on (001) blanket Si substrate operating at 80 °C. However, the thick buffer layers required for defect reduction prevent evanescent light coupling to the underlying Si waveguides. The solution is thus to deposit the QD laser material in the narrow oxide pocket, patterned on Si wafers, and butt-coupled to the waveguides embedded within.
In a new paper published in Light Science & Application, a team of scientists, led by Professor John E. Bowers from University of California Santa Barbara, in collaboration with SUNY Polytech and IQE, have demonstrated the first electrically pumped QD laser grown by molecular beam epitaxy (MBE) in narrow oxide pockets patterned on CMOS compatible Si substrate. Transferring the success from blanket Si substrate to pattern substrate is a non-trivial task for the template architecture induced unexpected challenges, especially for the delicate QD nucleation process. These scientists have summarized the major complications for growing and fabricating devices in the narrow pockets.
“In MBE growth, the growth surface temperature is the most important parameter that we need to measure precisely. It is normally done by looking down directly at the surface with a pyrometer in-situ and the temperature window for high quality QD nucleation is about ±2 °C. However, when the substrate is mostly covered with oxide, such temperature measurement cannot be executed accurately as the oxide surface has a very different emissivity then the III-V and much lower heat conductivity. As the growth proceeds, since MBE is a non-selective technique, poly crystal III-V would be deposited onto the oxide. These poly crystals are highly defective and thus absorptive of the ambient radiation which causes additional uncertain heating. We had to then seek temperature sensitive material properties as ex-situ gauges to infer the temperature offset. We were then able to locate the narrow temperature window for high quality QD nucleation. The deposited poly crystal and the high aspect ratio of the structure also post fabrication challenges. We have developed a wet etching technique to selectively remove only the poly crystal. The metal contacts were then successfully realized in the deep pocket.”
“The temperature offset is expected to be different depending on the template architecture and the III-V epi structure. The method we proposed is applicable to any combinations of those, provided certain layers within the III-V stack are temperature sensitive” they added.
“The current devices perform similarly, in terms of reliability, to our previous devices grown on blanket substrate with similar defect level. Growing in narrow pockets has other advantages as well. It has previously been shown that the cause of degradation of these epitaxially grown QD lasers is primarily due to the residual stresses. The narrow pocket geometry could significantly reduce that residual stress. Thus, the in-pocket lasers could potentially exceed the performance of those previously demonstrated lasers grown on blanket Si.” The scientists forecasted.
Journal
Light Science & Applications