Xianyu Hu, Xinglong Wang, Kangping Songjian, Zixiong Liu, Zhongfeng Ning, Yunlei Zhong, Bowen Shen, Anquan Jiang
The von Neumann bottleneck limits energy-efficient computing. While ferroelectric hafnium zirconium oxide (HZO) memories are promising for in-memory computing, achieving high speed, endurance, and reliable multilevel control remains challenging. This work addresses these challenges through interfacial engineering with an ultrathin ZrO2 seed layer. Atomic-resolution microscopy reveals that this interlayer promotes preferential c-axis orientation of the ferroelectric orthorhombic phase, aligning the polarization axis with the applied electric field. This enables nanosecond (6 ns) switching, long-term retention (>104 s), and stable programming of 10 distinct polarization states. Based on these capabilities, we demonstrate an in-memory differentiator within a single device. Analog values encoded as discrete polarization levels enable direct first- and second-order derivative calculations, where the transient switching current represents the differential output. This atomic-scale structural control provides a materials-to-system link that may facilitate real-time, energy-efficient data processing.