Halide perovskites (HPs) have been ubiquitously utilized in optoelectronics and photovoltaics in the recent past owing to their tunable bandgap, long charge diffusion lengths and high color purity. However, their application in resistive random access switching memories (RRAMs) has been somewhat limited by material and electrical instability associated with HP thin-film devices, resulting in below-par figures-of-merits (FOMs) such as data retention time, electrical endurance and switching speed. To collectively address the shortcomings of HP thin-film based RRAMs, we developed a switching matrix which replaces the thin-film anatomy with vertically aligned, three-dimensionally integrated, ultra-high density monocrystalline HP nanowires and quantum wires rooted in a porous alumina membrane (PAM), clubbed between metal contacts acting as the top and bottom electrodes. The PAM provides excellent passivation, providing the required electrical, and material stability to the environmentally delicate HPs by drastically reducing surface diffusion pathways, thereby thwarting moisture-induced attacks.Our approach resulted in record long retention times of up to 28.3 years, measured device endurance of 5 million cycles, and switching speed of 100 ps, the best FOMs ever reported for HP RRAMs. In comparison, the HP thin-film RRAMs exhibited maximum endurance of 10 thousand cycles, best retention of 105 s and best switching speed of 10 ns. Furthermore, we developed a 14 nm lateral size HP quantum wire RRAM cell and a cross-bar device architecture with a unique metal-semiconductor-insulator-metal (MSIM) based sneaky path alleviation scheme that demonstrated the scalability potential of our devices. In addition to data storage, we pioneered the development of HP based neural networks using nanowires and quantum wires in PAM, with low power and high accuracy processing capabilities. Two different modes - electro-chemical metallization (ECM) mode with the silver top electrode and non-electro-chemical metallization (non-ECM) mode with indium tin oxide as top electrode, were used to operate the devices intended for neuromorphic computing application. The obtained multi-level states regulated the modulation of conductance values in the artificial neural networks. They were utilized to perform various image processing tasks such as outlining, sharpening, and embossing in the ECM mode. In the non-ECM mode, we replicated the cognitive learning model of Gestalt Closure Principle by utilizing the multi-level states obtained. Our approach of using HP nanowires and quantum wires embedded in a PAM offers better material reliability, retention, and jitter of the conduction states compared to their thin-film devices. The utilization of HP nanowires and quantum wires in PAM enables the creation of a physical artificial vision system by integrating a physical processor with a physical pre-processor. This integration leads to a consistent photo-synaptic behavior and stable and temporally robust conduction states. The devices demonstrate retention exceeding 105 seconds and temporal jitter less than 10%. They also exhibit photo-synaptic behavior by responding to variations in intensity, duration, and frequency of light pulses, which allows the device to enhance contrast of input visual stimuli. Moreover, we accomplished the recognition of four different geometric shapes using a 6×6 array of nanowire devices, resulting in the creation of a physical artificial vision system that positions HP as a cutting-edge and dependable active layer. This development opens up a multitude of possibilities in the field of artificial intelligence, as it emulates the human vision system.In conclusion, these remarkable advancements propel HP RRAMs to the state-of-the-art standard and highlight the potential of perovskite nanowire/quantum wire devices as a substitute technology for upcoming storage and computing modules.
Read full abstract