CAREER: Unifying Volatile and Non-Volatile In-Materia Computing with Ionically Gated Transistors for Energy-Efficient Edge AI

Artificial intelligence is increasingly essential for small, power-limited devices that process information in real time, such as wearable health monitors, autonomous drones, and environmental sensor networks. However, most current computer hardware is inefficient for these "edge" applications because it relies on moving massive amounts of data between separate memory and processing units. This constant data movement creates a significant energy bottleneck and slows down operations. This project addresses this challenge by developing a new class of electronic hardware where memory and processing are combined within the same physical material. The research focuses on ionically gated transistors, which are electronic switches whose behavior is controlled by the movement of ions, or charged atoms, within a solid material. By mimicking the way biological brains process information, this "in-materia" approach enables hardware that can both respond dynamically to fast-changing signals and store learned information permanently in place. Beyond the technical innovations, the project provides a public benefit by strengthening the United States' semiconductor workforce. It integrates research with education by training graduate and undergraduate students in advanced microchip fabrication and modeling. Additionally, the project engages K–12 students and the local community through hands-on demonstrations that illustrate how new materials can enable the next generation of energy-efficient computing. Technically, the project establishes the fundamental science and engineering of a dual-mode ionically gated transistor that functions as a Co-located Adaptive Synapse (CAS). The CAS is a single device capable of operating in two distinct regimes controlled by the magnitude of the applied gate voltage. At lower voltages, the device operates as an electric double-layer transistor (EDLT), where ions accumulate at the interface of the channel to produce a volatile, fading-memory response. This regime is optimized for transient signal processing. At higher voltages, the device transitions into an electrochemical random-access memory (ECRAM) element, where ions physically enter the crystal lattice of the two-dimensional (2D) channel material to create a persistent, non-volatile change in electrical conductance. This regime is suitable for long-term analog weight storage. The research investigates how 2D channel materials, electrolyte compositions, and ionic transport kinetics govern the transition between these two regimes. The work proceeds through three integrated thrusts: (1) the fabrication and characterization of dual-mode devices with tunable response times and stable analog conductance states; (2) the development of a physics-based compact modeling framework that incorporates experimentally measured volatile and non-volatile behaviors into circuit-level simulations; and (3) the demonstration of compact adaptive computing architectures that utilize the same device array for both transient signal processing and in-place weight storage. The intellectual significance of this work lies in establishing the design principles for coupled ion-electron dynamics in low-dimensional materials and demonstrating a unified hardware primitive that reduces data movement, lowers energy consumption, and enables adaptive artificial intelligence at the hardware level. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2544262 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Ke Xu | Institution: Rochester Institute of Tech, ROCHESTER, NY | Award Amount: $514,145 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2544262 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2544262.html