Temporal data learning of ferroelectric HfAlOx capacitors for reservoir computing system

Abstract

Extensive research has been directed towards HfOx-based ferroelectric capacitor in contrast to perovskite-based ferroelectric capacitors. HfOx-based ferroelectric capacitor present advantages for high-density memory applications due to their compatibility with complementary metal-oxide semiconductor technology, efficient power consumption, and rapid operational capabilities. Particularly, Al-doped HfOx exhibits superior ferroelectric properties owing to the smaller atomic radius of Al compared to Hf. This study conducts electrical analysis by varying the type of metal electrode (W, Mo, TiN, and Ni) in the metal-ferroelectric-insulator-semiconductor (MFIS) device to achieve excellent ferroelectric memory performance. The choice of W as the metal electrode, characterized by a smaller thermal expansion coefficient (4.59 × 10−6/°C) compared to the other three electrodes, results in a high remnant polarization value (18.35 µC/cm2). Additionally, W demonstrates a stable high on/off ratio at low voltages, as verified by the I–V characteristics. Nonetheless, the ferroelectric capacitor within the MFIS structure experiences a depolarization field in the opposite direction of the aligned polarization. Consequently, a minor issue arises regarding retention loss. This phenomenon will be leveraged in reverse to demonstrate encompassing paired-pulse facilitation, spike-timing-dependent plasticity, spike-rate dependent plasticity, and long-term potentiation and depression among various synaptic applications in neuromorphic computing. In conclusion, we successfully implemented a 4-bit reservoir computing system utilizing a physical reservoir. This demonstration serves as evidence that reservoir computing is well-suited for application in image recognition technology. This comprehensive approach underscores the significant potential of W/HfAlOx-based ferroelectric capacitors in advancing artificial neural networks, aligning with the innovative trajectory of memristor technology. © 2024 Elsevier B.V.