Projects

This project focuses on developing compact and energy-efficient intelligent sensing systems through device–hardware–algorithm co-design. The central idea is to use programmable material and device responses as part of the computing pipeline, so that sensing, feature extraction, data compression, and inference are no longer treated as separate stages. Representative material platforms include, but are not limited to, electrochromic (EC) materials and dual-gated two-dimensional (2D) semiconductors, which provide electrically tunable optical, electronic, and optoelectronic responses. By integrating these tunable devices with compact neural networks, analog computing hardware, and task-aware learning algorithms, this research demonstrates how device physics can be engineered to reduce data movement, simplify post-sensor processing, and improve system-level efficiency. The long-term goal is to build adaptive, scalable, and low-power sensing hardware for intelligent vision, scientific imaging, autonomous systems, robotics, and edge AI.

Advanced optical and spectroscopic characterization plays a critical role in understanding the structure and properties of nanomaterials, yet conventional data analysis often relies heavily on expert intuition, labor-intensive model fitting, and manual interpretation, limiting the extraction of complex physical information and real-time measurements. This project explores deep learning as a general framework for automated and intelligent optical characterization of nanomaterials. Using two-dimensional materials as a model system, neural networks are developed to identify material type and thickness directly from optical microscopy images with high accuracy and real-time processing capability, while transfer learning extends the approach to additional material systems and characterization tasks. Building on this concept, recurrent neural networks are further used to extract complex refractive indices directly from reflectance spectra, enabling in-operando tracking of material properties under external stimuli such as temperature, electric fields, magnetic fields, and mechanical deformation. These approaches demonstrate how AI can transform complex imaging and spectroscopic data into quantitative materials information, enabling automated characterization, real-time monitoring, and accelerated understanding and discovery of emerging nanomaterials.

Manipulating materials with atomic-scale precision is essential for the development of next-generation material design toolbox. The family of 2D materials provides an ideal platform to realize atomic-level material architectures. We developed a novel atomic-scale material design tool that selectively breaks and forms chemical bonds of 2D materials at room temperature, called atomic-layer substitution (ALS), through which we can substitute the top layer chalcogen atoms within the 3-atom-thick transition-metal dichalcogenides using arbitrary patterns. Flipping the layer via transfer allows us to perform the same procedure on the other side, yielding programmable in-plane multi-heterostructures with different out-of-plane crystal symmetry and electric polarization. This electric dipole can be selectively patterned to be zero (MoS2, MoSe2), positive (MoSSe), and negative (MoSeS) on such a 2D material “canvas”, which would enable many novel nanostructures and devices with intriguing electrical and optoelectronic properties. 

Atomically thin two-dimensional (2D) semiconductors have great potential for realizing ultimately scaled high-performance electronic devices. However, because of metal-induced gap states (MIGS), energy barriers at the metal-semiconductor interface, which fundamentally lead to high contact resistances and poor current-delivery capabilities, have restrained the advancement of 2D semiconductor transistors to date. In this project, we study a novel ohmic contact technology between semimetallic bismuth and semiconducting monolayer transition metal dichalcogenides (TMDs) where MIGS is sufficiently suppressed and degenerate states in the TMD are spontaneously formed in contact with bismuth. Through this approach, we achieve zero Schottky barrier height, a record-low contact resistance (RC ) of 123 Ω μm, and a record-high on-state current density (ION) of 1135 µA/µm on monolayer MoS2. We also demonstrate that excellent ohmic contacts can be formed on various monolayer semiconductors, including MoS2, WS2, and WSe2. Our reported RC values are a significant improvement for 2D semiconductors, and approaching the quantum limit. This technology unveils the full potential of high-performance monolayer transistors that are on par with the state-of-the-art 3D semiconductors, enabling further device down-scaling and extending Moore’s Law.