Highly Reliable Implementation of Optimized Multicomponent Oxide Systems Enabled by Machine Learning‐Based Synthetic Protocol

Abstract

current, good stability, and large-area scalability. [6,7] In particular, AOSs that can replace the conventional polycrystalline Si have been actively researched in displays due to the growing demands for highresolution and high-frame-rate displays. [8] To obtain high-mobility AOSs, various multicomponent oxide systems have been investigated including In-Zn-Sn-O [9] (IZTO), In-Ga-Zn-O, [10] In-Ga-Sn-O, [11] and Al-In-Zn-O [12] channels. In the case of multicomponent AOSs, electrical properties such as field-effect mobility, threshold voltage (V TH ), and on/off ratio are strongly dependent on the cation ratio. Therefore, to find the optimal cation ratio, a combinatorial study on various cation components has been typically carried out. [13,14] However, although the conventional combinatorial approaches can allow successful discovery of optimal cation ratios, rather exhaustive experimental procedures are usually required, especially when there are many different cationic components to examine. Moreover, the complicated contributions from extrinsic parameters on thin-film transistor (TFT) mobility value, such as film quality and interface effects, make the determination of optimized device challenging, even with extensive first-principle study on amorphous multicomponent semiconductors using huge computation cost. [15,16] In this regard, machine learning (ML)-based synthetic approach can be a promising solution to optimize the cation ratio by predicting the electrical properties of multicomponent AOSs with a minimal trial and error via reduced number of experimental data.As is well known, ML is a technique that predicts the results by learning the patterns of training data. For this reason, ML has been widely utilized in various applications such as speech recognition, natural language processing, and robotics control. [17] Recently, using the ML, there has been a lot of interest in revealing the complex correlation between composition, processing, structure parameters, and electrical properties of various material systems. For instance, the discovery of new ternary oxide compounds for metallic glasses [18] and batteries, [19] and parameterizing the point defects in fin field-effect transistors [20] and band gap in inorganic solids [21] have been reported. In the case of multicomponent AOSs, although the ML can be aThe ORCID identification number(s) for the author(s) of this article can be found under https://doi.org/10.1002/smtd.202101293.

Retina‐Inspired Color‐Cognitive Learning via Chromatically Controllable Mixed Quantum Dot Synaptic Transistor Arrays

“…For more convincing future approaches to come up with the fully reproducible biological visual system, intensive learning structure including more visual components like a rod in the biological system can be considered. Throughout accurate and reliable prediction based on intensive learning computing, [57] further optimization of the M-QD-based photonic visual system can be expected.…”

Less‐Energy Consumed Hydrogen Evolution Coupled with Electrocatalytic Removal of Ethanolamine Pollutant in Saline Water over Ni@Ni3S2/CNT Nano‐Heterostructured Electrocatalysts

“…[ 38,39 ] If the anionic precursor of diphenyl disulfide is timely injected into the reaction system once the temperature reaches 220 °C, the Ni 3 S 2 nuclei are grown by polycondensation of those released cationic and anionic species from precursors’ decomposition, thus resulting in the formation of Ni 3 S 2 nanocrystals in CNTs networks. If the hot injection is postponed for several minutes, the metal Ni nanoparticles with smaller particle size will be formed as metal core before injection, and then the rest of nickel precursor will be reacted with sulfur precursor after injection to form the Ni 3 S 2 shell on the surface of Ni core in order to minimize the surface free energies, [ 38–40 ] thus leading to the formation of Ni@Ni 3 S 2 core@shell nanocrystals dispersed in CNTs network.…”

Artificial Intelligence for Surface‐Enhanced Raman Spectroscopy

“…Numerical simulations like finite-difference time-domain, finite element method, and discrete dipole approximation are therewith developed, addressing to substrates with much more complicated configurations and obtaining multiple properties including far-field optical cross-sections and near-field electromagnetic enhancement. [79] GPR, [69] RR, [42] KNN, [42] ANN [ 42,[70][71][72][73] 2D/3D image 3D CNN, [74] U-net, [75] CNN+RNN [76] Optical properties Optimize the structure design LR, [77] RL, [78] DNN, [79][80][81][82] cGAN, [ 83,84] DeepAdjoint [85] Composition ratio Predict the carrier mobility SVR+RBF [86] Molecular descriptors Predict the energy levels Histogram gradient boosting [87] Reporter selection and design…”

“…[175,176] SVR algorithm with RBF kernel, for instance, has been exploited to accurately predict the optimal material ratio of a multi-component semi-conductive substrate for the carrier mobility. [86] Another work by Mubashir et al, applied the Histogram Gradient Booting Regressor to predict the HOMO and LUMO and select the semi-conductor toward efficient charge transfer, followed by SHapley Additive exPlanations (SHAP) to further explain the model. [87] More explorations are on the way for AI-driven design of semi-conductive substrates particularly to optimize SERS signals for certain molecules.…”