Flexible Zr-doped hafnium oxide ferroelectric memcapacitive synaptic devices for neuromorphic computing

Abstract Conventional von Neumann architecture-based computing hardware faces high power consumption challenges and the mismatch between memory and computation performance when handling increasingly large and complex datasets. The rise of artificial neural networks (ANN) has driven the development of bio-inspired memory devices, such as artificial synapses, offering new opportunities to overcome the limitations of the post-Moore era. Ferroelectric memcapacitors, which store data based on capacitive principles, provide advantages such as resistance to read disturbance and low static power consumption compared to memristors, making them highly attractive for ANN applications. In this work, we fabricated Zr-doped HfO 2 (HZO) ferroelectric memcapacitive synaptic devices on mica substrate featuring a simple metal-ferroelectric-metal structure. We investigated the impact of bending on the ferroelectric performance and, through finite element analysis, found that stress concentration might lead to an increase in grain boundary defects, resulting in excess oxygen vacancies and a slight decrease in ferroelectric polarization. Additionally, we studied the memcapacitive characteristics of the HZO ferroelectric synaptic devices and successfully obtained the potentiation and depression characteristics. The synaptic plasticity curves were also fitted by least squares to obtain asymmetries of 0.030 and 0.068 for the flat and tensile states, respectively. The devices demonstrated remarkable neuromorphic computing capabilities, reaching a digital recognition accuracy as high as 97.63%. Notably, the maximum energy consumption per device was approximately 1.96 nJ in the training process and 4.03 pJ in the inference process. This study provides new directions for further developing flexible hafnium-based ferroelectric materials and devices. Graphical Abstract