Ion-dynamic Capacitance Enables Multimode Transistors and Multimode Neural Networks

Abstract Transistors attractive for neuromorphic computing should offer tunable plasticity, ultrahigh apparent mobility, steep subthreshold swing for low power consumption, or memristive conductance for in-memory computation. However, current transistor structures only offer one or two of the above characteristics. Here, we derive a concise model to describe the complex transient ion-dynamic capacitance in transistors and reveal that the aforementioned characteristics could all be achieved at different operational modes on demand in a single transistor. We experimentally verified such multimode transistors with tens of nanometer thin films. The multimode transistor enables an unprecedented neural network that can switch as needed between conventional artificial neural network (ANN), reservoir neural network (RNN) and spike neural network (SNN) for image recognition in near-sensor computing. The study unambiguously unveils ion-dynamic capacitance can be the universal origin for various attractive transistors, and significantly expands the programmability of transistors to realize new analogue computing concepts.