**5. Applications**

Perovskite materials have the potential to change properties with light; because of this property, this material is used in the application of optoelectronic as well as in data storage devices. By combining both to prepare optoelectronic based logic gate devices, and this electrical signal is used as write, and the optical signal is used as erase in resistive switching devices and by changing the light intensity increase/ decrease set and reset voltage [70, 71]. Wang and colleagues programmed the Au/ MAPbI3-xClx structure devices to set/reset by photo/electrical bias [72]. For multilayer storage RRAMs, the set voltage falls as light intensity rises. The gadget can carry out logical operations and coincidental event detection tasks by using optical and electrical pulses. In their experiment, Chai and colleagues discovered that light might lower the device's set voltage, and this finding might be used to develop logic circuits [73].

Memristors can potentially be useful in ultrahigh storage density computing technologies. An immediate application for these devices is the resistive randomaccess memory (ReRAM). To meet the growing demands of next-generation data storage devices, ReRAMs must exhibit characteristics such as small write voltage (few hundred millivolts), short write time (<10 ns), small read voltage so that there is no change of internal resistance, high OFF-to-ON resistance ratio (>10), high endurance (∼103 ), high retention (∼10 years), and small device size (<10 nm), in addition to low-cost fabrication and flexibility [74].

Memristor can be used as a programmable logic gate with the building of crossbar architecture. When comparing CMOS-based devices with memristor devices are far more variable. The major problem in many logic gate architectures using memristor is the endurance and device-to-device variation. Also, memristor-based gates become less reliable but the capacity to accept changes in weight values has a high level of device variability tolerance.

Memristors are further used in artificial neural networks. Here, two of them are discussed, CNN (convolution neural network) and SNN (spiking neural network). **CNN (convolutional neural network)**: Memristor arrays allow for the concurrent and as well as on computation of vector–matrix multiplication operations, which significantly speeds up inference and training for convolutional neural networks (CNNs) and related deep neural networks (DNNs). Through Kirchhoff's current law and Ohm's law, the memristor crossbar arrays are employed to store the weights and carry out simultaneous multiply-accumulate operations [75]. In this system, the outputs are represented as the accumulated currents on the columns, while the inputs are represented as voltage pulses applied to the rows. ADCs or a sense amplifier may then read out the output activations after being quantized. First single layer perceptron is used to demonstrate simple pattern recognition, which serve as the foundation for memristor-based artificial neural network development [76, 77]. Later these work on multi-layer and CNN based architecture used for image recognition from well-known data sets as MNIST and CIFAR 10 [78]. **SNN (**spiking neural network**):** Memristors' internal dynamics and processes bear strong resemblances to biological processes, opening the door to the development of bio-faithful neuromorphic systems without the need for intricate circuitry and sophisticated algorithms. Information processing can instead be carried out locally by device dynamics. SNN is one of the applications of the internal dynamics based on memristor devices, this system maximizes the efficiency of complex functions. SNNs have the potential to mimic the biological brain closely due to the owing spike-driven communication. This system fires a signal when the potential from input reach to threshold and the fired signal transfer to the neighbor neuron. The STDP learning rule for SNN training comes from neuroscience. For train to SNN, two types of learning are used: supervised learning and unsupervised learning. The unsupervised STDP learning is very energy efficient because of the local learning nature and ability to learn in only few spikes. Where accuracy is considered the most important factor, supervised learning is generally utilized for training neural networks. The inability to differentiate spiking events is the main obstacle to backpropagation in SNN [79].

Memristor can also be useful as artificial vision sensor, touch sensor, and pain sensor in the arena of bionic electronics. Vision is the crucial sense system through which *Hybrid Perovskite-Based Memristor Devices DOI: http://dx.doi.org/10.5772/intechopen.109206*

#### **Figure 5.**

*Illustrations of the memristor-based bionic systems (a) memristor as a touch sensor, (b) artificial retina, and (c) nociceptor.*

most of the information are collected by humans. To replicate artificial vision sensor, memristor is used as visual neuromorphic device. This involves two types of impulse signals, optical and electrical, as presynaptic and postsynaptic impulses as input and output signals, respectively. This kind of device is also called photonic memristor and that used as photodetector with photo stimuli as retain of human eyes that collect process visual picture and transport [80].

Another application is pressure or touch sensor, where a small electrical signal obtains as a result of pressure, which is promptly transferred to a sensory receptor as an input response for further processing. This system is only initiated when the pressure touch is converted into electrical signal and therefore postsynaptic current. As shown in **Figure 5(a)**, changing pressure on device by finger folding fire signal in form of electrical current. It can measure the quantity, frequency and speed, and time duration of that touch pressure [81].

In reaction to harmful stimuli, the body feels the unpleasant emotion of pain. When external stimuli activate pain receptors in the stomach or body, the central nervous system receives, interprets, and sends the pain information. The ability of AI systems to perceive pain and become sensitized to it is essential for significantly increasing the efficacy of hardware devices since it enables them to have different sensitivities to external stimuli for various purposes [82].
