After its invention in 1960, the laser caused a paradigm shift in long-haul telecommunications, delivering unparalleled communications bandwidth at high power. However, next-generation integrated on-chip optical data communications require different specifications, favoring efficient nanoscale transmitters operating at low-power and high direct-electrical modulation speed. This dissertation argues that a novel device called the optical antenna-enhanced light-emitting diode (antenna-LED) can meet these requirements, as an alternative to conventional semiconductor lasers.
A detailed comparison between LEDs and lasers for on-chip optical links is provided in the first part of the dissertation. We demonstrate novel methods to quantify the stimulated emission carrier lifetime (?_{st}) and spontaneous emission carrier lifetime (?_{sp}) in lasers and LEDs respectively, ultimately finding ?_{st} = 6ps for a laser at saturation and ?_{sp} = 1ns for a heavily-doped LED. While exploring the limits of ?_{sp}, we reject the standard BNP model of the LED recombination rate. We go on to show that optical antennas can enhance the rate of spontaneous emission from LEDs by several orders of magnitude. The resulting antenna-LED can reach the needed carrier lifetime due to (enhanced) spontaneous emission of 6ps, rivaling the corresponding lifetime in lasers. This allows us to quantify the direct-electrical modulation rate of antenna-LEDs versus lasers. In doing so, we reject the standard small-signal modulation approximation in favor of the large-signal digital modulation that would be required in low-power on-chip interconnects. We find that antenna-LEDs and lasers are both limited by their respective carrier lifetimes in this modulation format – indicating that antenna-LEDs can be as fast as lasers. Finally, we show that antenna-LEDs are capable of achieving practical internal quantum efficiency, provided that surface treatment processes for III-V semiconductors are improved. Putting it all together, our analysis demonstrates that an antenna-LED with 6ps carrier lifetime, 10^4cm/s surface recombination velocity, and 50% overall efficiency can reach a direct-electrical modulation speed exceeding 50Gbit/s while emitting 500 photons/bit. We argue that this is sufficient signal for next-generation receivers.
Tangentially, we demonstrate novel ways that the antenna-LED radiation efficiency and waveguide coupling efficiency can be maximized. For example, by utilizing dielectric nanostructures in the antenna gap, a known tradeoff between antenna enhancement and antenna efficiency can be overcome. Furthermore, we present the design and simulation of an optical antenna-LED with 94% coupling efficiency to a single-mode waveguide, potentially enabling efficient integrated optical interconnects.
In the final part of the dissertation, we discuss inverse electromagnetic design – computational tools and methods that can be used to efficiently optimize electromagnetic devices consisting of arbitrary numbers of geometric parameters. We provide two inverse design tutorials: (1) inverse design via the adjoint method for electromagnetic devices that satisfy Maxwell’s equations, and (2) a novel semi-analytical transfer-matrix method for the design of 1D interference filters. We then apply these techniques to conventionally difficult electromagnetic design problems. Using (1) we design 65nm-CMOS-compatible perfectly vertical grating couplers with an industry competitive simulated insertion loss of -0.52dB. Using (2) we demonstrate distributed Bragg refflectors (DBRs) with >99% reflectivity over an extremely broad spectrum. We conclude with a brief look at emerging inverse design methods that are aided by neural networks.
