Ming C. Wu (Advisor)

3D perception is an essential task for autonomous driving, and thus building the most accurate, computationally efficient, fast, and label efficient models is of great interest. In particular, label-efficient 3D detection is attractive as manual labeling of 3D LiDAR point clouds is both costly and time-consuming. Autolabeling is a machine learning paradigm in which a model is trained on a (small) set of labeled data before being used to generate predictions, known as pseudo-labels, on a large set of unlabeled data which can then be used to train an accurate downstream model with only a...

Optical interconnects have displaced electrical wires in data centers and high performance computers due to higher efficiency and bandwidth. The development of on-chip optical interconnects is the next frontier for development, with the promise of reducing on-chip energy consumption. Light emitting diodes have high efficiencies and can achieve low footprint; however, spontaneous emission is typically far too slow to be useful in an interconnect. This can be understood by modeling spontaneous emission as dipole emission with a dipole length on the order of the atomic spacing (~1.3nm) which...

We developed a new architecture suitable for building a large-scale optical switch with fast response time. We have demonstrated switches with a scale of 240x240 and speed of sub microsecond using our new architecture. The switch architecture consists of an optical crossbar network with MEMS-actuated couplers and is implemented on a silicon photonics platform. To our knowledge this is the largest monolithic switch, and the largest silicon photonic integrated circuit, reported to date. The passive matrix architecture of our switch is fundamentally more scalable than that of multistage...