Multilevel Reconfigurable Differential Capacitance in HfZrO2 Ferroelectric Devices: Enabling Machine Learning-Based Classification

Abstract

Reconfiguring differential capacitance (DC = dC/dV) holds significant promise for the advancement of energy-efficient and multifunctional electronic components. Typical passive components like resistors or conventional capacitors lack the ability to exhibit cumulative charge behavior, making them unsuitable for advanced computing and data processing tasks. In this study, we introduce a nanodomain HfZrO₂ ferroelectric device capable of modulating DC values from positive to negative, a feature enabled by its intrinsic and randomly oriented ferroelectric polarization. The device demonstrates dynamic multilevel hysteresis loop openings in capacitance-voltage characteristics, confirmed through advanced characterization techniques, including vector piezoresponse force microscopy and transmission electron microscopy. Using an Op-Amp integrator circuit, we derived cumulative charge (SumQ) from capacitance data with unprecedented control and predictability, achieving high linearity (>99%) and distinct charge levels, a feat unattainable in typical resistors or capacitors. Furthermore, the SumQ data was successfully employed in machine learning (ML) models to classify human presence and absence based on WiFi received signal strength indicator signals. This application underscores the potential of our device in enabling advanced ML-driven electronic systems and security applications. These results establish our device as an innovation, bridging the gap between physical electronic behavior and computational applications, and paving the way for next-generation high-performance electronic systems.