The increasing demands of artificial intelligence (AI), big data, and the Internet of Things (IoT) have strained the performance and energy efficiency of conventional computing systems. The von Neumann architecture, characterized by the physical separ...
The increasing demands of artificial intelligence (AI), big data, and the Internet of Things (IoT) have strained the performance and energy efficiency of conventional computing systems. The von Neumann architecture, characterized by the physical separation of memory and processing units, is hindered by excessive data movement, high power consumption, and limited scalability. To overcome these limitations, brain-inspired neuromorphic computing has emerged as a promising paradigm, necessitating advanced devices that integrate memory and computation with low energy overhead.
Ferroelectric field-effect transistors (Fe-FETs) are considered strong candidates for neuromorphic hardware due to their non-volatile memory capabilities and CMOS compatibility. However, traditional ferroelectric materials face challenges in miniaturization and efficient channel control. Recently, two-dimensional (2D) ferroelectric materials, such as CuInP2S6 (CIPS) and α-In2Se3, have gained attention for their stable ferroelectric behavior at the atomic scale.
This dissertation presents the design and experimental demonstration of 2D ferroelectric-based neuromorphic devices for energy efficient in-memory computing. The proposed all-ferroelectric FET, leveraging dipole coupling effects, exhibits a large memory window and analog conductance modulation via partial polarization switching, enabling synaptic functionalities. Additionally, a self-powered artificial visual system was realized, integrating light sensing, memory storage, and information processing within a single device.
A CIPS ferroionic-based threshold switching FET (TS-FET) is further introduced as a steep-switching transistor, achieving an ultra-steep subthreshold swing (~7.5 mV/dec) and an ON/OFF current ratio exceeding 107. Also, this device emulates essential neuronal behaviors such as leaky integrate-and-fire (LiF) dynamics and adjustable spiking thresholds without requiring external reset circuitry. By integrating CIPS-based synaptic and neuron devices, a spiking neural network (SNN) was constructed and validated through a face recognition task using a 2D convolutional neural network (CNN), achieving 95.83% accuracy.
In conclusion, this work lays the groundwork for next-generation in-memory computing architectures by advancing the development of all-ferroelectric FETs, self-powered artificial visual systems, and ferroionic TS-FET-based artificial neurons, demonstrating the scalability and efficiency of 2D ferroelectric materials for neuromorphic hardware.