Inverse Design and Lithographic Pattern Transfer of a Thin Near-Infrared All-Silicon Absorber

Abstract

All-silicon photodetectors are seldom used for Near-Infrared (NIR) Light Detection and Ranging (LiDAR) due to the low absorption coefficient of silicon in the NIR wavelength range. The absorptance of the absorption region is typically increased by increasing the depth of the absorption region, however, this approach adds timing jitter to a photode- tector that can severely limit the ranging resolution of the LiDAR system. This work maximized the absorptance of a silicon absorber in a fixed-depth 2.475 μm absorption re- gion from a theoretical baseline value of 2.58% to a simulated value of 8.33% for 950 nm wavelength incident light, a more than 3-fold improvement. The simulated absorptance enhancement compared to the baseline value was 3.2, whereas perfect black silicon only displays an absorptance enhancement of less than 1.5 at the 950 nm wavelength, giving the structure designed and simulated in this work a more than 2-fold absorptance enhancement improvement over perfect black silicon.

This work applied topological optimization and inverse design methodologies to gen- erate the unique all-silicon 2 μm by 2 μm meta-atom absorber with vertical sidewalls. A pattern transfer using electron-beam lithography on a planar silicon sample with 340 nm of ZEP520A resist was conducted as a proof of concept that the pattern can be transferred faithfully into a real silicon sample. At the time that this work was conducted, the author was unaware of any such studies that applied topological optimization to maximize the absorptance of a thin all-silicon NIR absorber. This work showed that topological opti- mization can be effectively utilized to automatically design photonic devices where classical device architectures and structures designed by humans have poor performance.

This work also proposes an initialization procedure for generating initial parameters for the optimization procedure. The structure optimized in this work was seeded using the initialization procedure showcasing its validity and effectiveness.