Volatile and Non‐Volatile Ferroelectric HZO for Low‐Power Reservoir Computing

ABSTRACT We present a capacitive reservoir computing system based on hafnium‐zirconium oxide (HZO) capacitors that integrates intrinsically volatile and non‐volatile functionalities within a single material platform. Through compositional control, fabricated Hf 0 . 8 Zr 0 . 2 O 2 provides stable ferroelectric (FE) behavior for non‐volatile weight storage, whereas Hf 0 . 2 Zr 0 . 8 O 2 exhibits volatile antiferroelectric‐like (VFE) dynamics characterized by fading memory and nonlinear responses. Using polarization‐voltage, capacitance‐voltage, and dynamic transient measurements from these devices, we formulate and calibrate a modified Landau‐Khalatnikov compact model that accurately reproduces both FE and VFE switching behaviors. Implemented and validated in Cadence Spectre, the model enables realistic large‐scale circuit simulations. Leveraging this framework, a hybrid reservoir network combining an FE HZO capacitor array with transistor‐coupled VFE HZO nodes was developed. Co‐simulation between Python‐based system evaluation and Cadence circuit modeling achieves 93.5% accuracy on the Voice‐MNIST classification task with ultralow energy consumption (∼184 fJ per cell and ∼1.07 pJ per node per inference). These results establish a unified HZO‐based device‐circuit‐system framework for scalable, CMOS‐compatible, and energy‐efficient temporal neuromorphic computing.