Short-term Synaptic Plasticity in Oxide Phototransistors for Temporal Data Processing in Neuromorphic Computing

Abstract

Artificial synapses exhibiting short-term plasticity (STP) are crucial for advanced neural networks that can process the time sequences of signals. Artificial synapses made from CMOS-based neuromorphic hardware require cumbersome circuits to simulate synaptic STP adequately and the investigation of non- CMOS neuromorphic hardware is paramount. Two-terminal memristors have been investigated but they suffer from cross-bar interference problems. Three-terminal thin-film transistors have emerged as a promising alternative to emulate synapses, especially when using phototransistors where multi-input sensing, data memory, and processing functions can be realized in one unit for in-senor computing. We propose oxide-based thin-film phototransistors (TFPTs) to emulate the dynamical STP of synapses using a single device encompassing conductance relaxation after the stimulus, noted as fading memory . The TFPTs will respond to electrical and-or optical pulse streams, facilitating twofold temporal data processing. The separation of the programming terminal (gate or light) and reading terminal (drain) will enable the simultaneous recording of the conductance switching and the transient current response. The fabrication of TFPTs on plastic substrates will enable wearable and mobile applications. Four main tasks will be performed by a collaborative Taiwanese (TW) and US team. During the project, thin-film transistors and TFPTs will be fabricated using physical vapor deposition, solution processing, and photonic curing. Different dielectrics and light-absorbing layers will be investigated. Transient characteristics and photoresponse of the TFPTs will be characterized. TFPT device structures will be simulated to accelerate experimental progress. A comprehensive investigation of the fundamental limits on energy consumption of the oxide TFPTs using an analytical model and machine learning will be developed. With this comprehensive approach of integrating experimental work with simulations, multi-input metal oxide devices will be attained to better emulate the dynamic STP of synapses for energy-efficient spatiotemporal information processing in neuromorphic computing.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 05, 2025
Source ID
FA23862414041

Entities

People

  • Julia W P Hsu

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Texas at Dallas

Tags

Readers

  • Distributed Systems and Data Platform Development
  • Integrated Circuit Design and Technology.
  • Neuroscience

Technology Areas

  • AI & ML