Materials and devices as solutions to computational problems in machine learning

Conduction Modulation of Solution‐Processed 2D Materials

“…The above significant nonlinear switching coupled with the memory effect can be potentially exploited to enable non-volatile memory electronics toward the development of neuromorphic hardware and systems for high-efficiency artificial neural network computing. [30] However, P(VDF-TrFE) can suffer from rapid ferroelectric fatigue, [26] that can lead to the converging of the memory, as we demonstrate with the two-terminal vertical devices. In the endeavor to develop non-volatile memory electronics, the memory (i.e., the switching hysteresis) needs to be enhanced.…”

Optoelectronic synapses based on a triple cation perovskite and Al/MoO3 interface for neuromorphic information processing

“…The reservoir part of RC has two requirements: it must be made up of individual nonlinear units and must be able to store information, which maps the inputs into a higher-dimensional computing space and then conducts pattern analysis in a readout section. 59 Thus, the key step to implement the RC is how to build a dynamic “reservoir,” which can map complex timing signals into a new space, and reduce the difficulty of subsequent calculations. Recent studies have shown that the combination of optical synaptic devices can realize RC image-processing systems, streamline network structures, and reduce energy consumption.…”

Nanoscale memristor devices: materials, fabrication, and artificial intelligence

“…A key advantage of using memristors in these computing paradigms is that only a few devices are required to fulfill the computing requirement, compared with potentially thousands using digital logic, promising a substantial reduction in materials and therefore great potential for higher-density circuits. 278 The past decade has witnessed great development of various types of artificial neural networks created by integrating memristors arrays, for convolutional neural networks, 195 spike neural networks, 279,280 recurrent neural networks, 281–283 deep neural networks, 193,284 and echo state neural networks. 285 These different neural networks have been used for a variety of applications such as pattern and face classification, 276,286,287 sparse coding, 288 multilayer perceptron network, 289 coincidence detection, 290 language learning, 291 reservoir computing, 256,292,293 reinforcement learning, 294 etc.…”