Researchers from ITMO University, Ioffe Institute of Physics and Technology and Harbin University of Engineering have developed a stable perovskite nanomemristor that can withstand more than 1,500 rewrite cycles and does not degrade after months. In addition, the researchers achieved record-low values of size and energy consumption - 70-80 nanowatts for a 130-160 nanometer perovskite single crystal.
(a) A schematic illustration of a perovskite nanomemristor based on a single crystal on indium tin oxide (ITO) contacted with boron-doped diamond (BDD). (b) Scanning top view of the perovskite nanocrystals array. (c) Electron diffraction image of an individual CsPbBr3 nanocrystal. (d) Raman spectra obtained from the single-crystals array. Image from: Opto-Electronic Advances
This achievement could pave the way for ultracompact memristors and enable faster and more energy-efficient ultra-compact processors for neuromorphic computing.
Practical implementation of memristors became possible only in 2008 with the development of technologies for producing ultrapure semiconductor materials. The memristor changes its resistance depending on the magnitude and direction of the current flowing through it, and it can be used to store information or process it energy-efficiently, because they consume less energy than conventional silicon transistors. These properties can be used to create energy-efficient information storage and processing systems, including neuromorphic computing, signal processing in AI systems, machine vision, acoustic-speech systems, and biointerfaces.
Perovskites are promising materials for creating new microelectronics, including memristors. However, perovskite memristors remained unsuitable for real-world applications due to their instability. The problem lies in the structure of the materials themselves: polycrystalline perovskite films used by many scientists have boundaries between the crystallites. Not only moisture and oxygen penetrate through these boundaries, causing chemical degradation of perovskite, but also uncontrolled migration of metal ions from the electrodes occurs. This leads to an irreplaceable switching of the memristor state at the same voltage.
For the first time, the team proposed a perovskite memristor that can withstand more than 1,500 rewrite cycles and does not degrade after several months of indoor operation. The reliability and durability of the element is guaranteed by the use of ohmic inert contacts and monocrystalline nanocubes made of caesium bromide of lead (CsPbBr3) as a semiconductor. It is one of the most chemically resistant perovskites based on lead halide. The single crystal structure of the nanocube also ensures the stabilization of the electrochemical properties of the elements from cycle to cycle. The perovskite monocube itself is located between indium tin oxide and boron-doped diamond. These are chemically inert electrodes, which also ensure the stability of the memristor switching.
At the same time, the researchers managed to achieve one of the lowest energy consumption values and the size of a perovskite memristor - 70-80 nanowatts for a single crystal measuring 130-160 nanometers. Other similar perovskite devices spend from 200 to 35 thousand nanowatts. These parameters make the development very compact and energy efficient. The memristor also switches in less than one millisecond, and the difference in current amplitude is about 4-5 orders of magnitude, which makes the device convenient and fast for signal processing.
"We modeled the behavior of charges inside a nanocrystal and saw that it is their accumulation at the interface that leads to the effect of memristivity. In fact, this accumulation of charges creates a dipole moment that changes the energy barrier in the perovskite-contact region. It is this phenomenon that allows the memristor to switch the resistance," explained Alexandra Furasov, author of the study and a senior researcher at the New ITMO Physics and Technology Institute.
The development could enable faster and more energy-efficient ultra-compact neuromorphic processors for artificial intelligence and machine learning tasks. In addition, such single memristors can be easily assembled into scalable circuits (crossbars) for the practical implementation of logic and switching of individual semiconductors. In the future, the scientists plan to conduct experiments on an array of monocrystalline perovskite nanocubes and test other materials as semiconductors.
