27 August 2020 in Journal of Applied Physics
As the challenges in continued scaling of the integrated circuit technology escalate every generation, there is an urgent need to find viable solutions for both the front-end-of-line (transistors) and the back-end-of-line (interconnects). For the interconnect technology, it is crucial to replace the conventional barrier and liner with much thinner alternatives so that the current driving capability of the interconnects can be maintained or even improved. Due to the inherent atomically thin body thicknesses, 2D materials have recently been proposed and explored as Cu diffusion barrier alternatives. In this Perspective article, a variety of 2D materials that have been studied, ranging from graphene, h-BN, MoS2, WSe2 to TaS2, will be reviewed. Their potentials will be evaluated based on several criteria, including fundamental material properties as well as the feasibility for technology integration. Using TaS2 as an example, we demonstrate a large set of promising properties and point out that there remain challenges in the integration aspects with a few possible solutions waiting for validation. Applications of 2D materials for other functions in Cu interconnects and for different metal types will also be introduced, including electromigration, cobalt interconnects, and radio-frequency transmission lines.
22 October 2020 in Machine Learning: Science and Technology
Data analyses based on linear methods constitute the simplest, most robust, and transparent approaches to the automatic processing of large amounts of data for building supervised or unsupervised machine learning models. Principal covariates regression (PCovR) is an underappreciated method that interpolates between principal component analysis and linear regression and can be used conveniently to reveal structure-property relations in terms of simple-to-interpret, low-dimensional maps. Here we provide a pedagogic overview of these data analysis schemes, including the use of the kernel trick to introduce an element of non-linearity while maintaining most of the convenience and the simplicity of linear approaches. We then introduce a kernelized version of PCovR and a sparsified extension, and demonstrate the performance of this approach in revealing and predicting structure-property relations in chemistry and materials science, showing a variety of examples including elemental carbon, porous silicate frameworks, organic molecules, amino acid conformers, and molecular materials.
14 July 2021 in Machine Learning: Science and Technology
Selecting the most relevant features and samples out of a large set of candidates is a task that occurs very often in the context of automated data analysis, where it improves the computational performance and often the transferability of a model. Here we focus on two popular subselection schemes applied to this end: CUR decomposition, derived from a low-rank approximation of the feature matrix, and farthest point sampling (FPS), which relies on the iterative identification of the most diverse samples and discriminating features. We modify these unsupervised approaches, incorporating a supervised component following the same spirit as the principal covariates (PCov) regression method. We show how this results in selections that perform better in supervised tasks, demonstrating with models of increasing complexity, from ridge regression to kernel ridge regression and finally feed-forward neural networks. We also present adjustments to minimise the impact of any subselection when performing unsupervised tasks. We demonstrate the significant improvements associated with PCov-CUR and PCov-FPS selections for applications to chemistry and materials science, typically reducing by a factor of two the number of features and samples required to achieve a given level of regression accuracy.
18 October 2019 in The Journal of Chemical Physics
We have analyzed structural motifs in the Deem database of hypothetical zeolites to investigate whether the structural diversity found in this database can be well-represented by classical descriptors, such as distances, angles, and ring sizes, or whether a more general representation of the atomic structure, furnished by the smooth overlap of atomic position (SOAP) method, is required to capture accurately structure–property relations. We assessed the quality of each descriptor by machine-learning the molar energy and volume for each hypothetical framework in the dataset. We have found that a SOAP representation with a cutoff length of 6 Å, which goes beyond near-neighbor tetrahedra, best describes the structural diversity in the Deem database by capturing relevant interatomic correlations. Kernel principal component analysis shows that SOAP maintains its superior performance even when reducing its dimensionality to those of the classical descriptors and that the first three kernel principal components capture the main variability in the dataset, allowing a 3D point cloud visualization of local environments in the Deem database. This “cloud atlas” of local environments was found to show good correlations with the contribution of a given motif to the density and stability of its parent framework. Local volume and energy maps constructed from the SOAP/machine learning analyses provide new images of zeolites that reveal smooth variations of local volumes and energies across a given framework and correlations between the contributions to volume and energy associated with each atom-centered environment.
1 March 2020 in AIP Advances
We report on the synthesis of cubic-phase garnet-type solid-state electrolytes based on Bi-doped Li7La3Zr2O12 (LLZO). Bi aliovalent substitution of Zr in LLZO utilizing the Pechini processing method is employed to synthesize Li7−xLa3Zr2−xBixO12 compounds. A strong dependence of the ionic conductivity on Bi content is observed, and under our synthesis and sintering conditions, a >100-fold increase over the un-doped sample is observed for x=0.75. Cubic-phase Li6La3Zr1BiO12 compounds are generated upon annealing in air in the temperature range 650 °C–900 °C. In contrast, in the absence of Bi, the cubic garnet phase of Li7La3Zr2O12 is not formed below 700 °C and a transformation to the tetragonal phase is observed at ∼900 °C for this un-doped compound. The role of Bi in lowering the formation temperature of the garnet cubic phase and in the ionic conductivity improvements is investigated in this work. We ascribe the effect of Bi-doping on ionic conductivity increments to changes in Li+-site occupancy and lattice parameters and the reduction in the formation temperature for the cubic-phase formation to rate enhancements of the solid-state reaction. To identify the site occupancy of Bi in the garnet structure, we employ synchrotron extended x-ray absorption fine structure spectroscopy. Our results indicate that Bi additions occupy the Zr-type sites exclusively, to within the accuracy of the measurements.
18 September 2023 in Open Research Europe
Easy-to-use libraries such as scikit-learn have accelerated the adoption and application of machine learning (ML) workflows and data-driven methods. While many of the algorithms implemented in these libraries originated in specific scientific fields, they have gained in popularity in part because of their generalisability across multiple domains. Over the past two decades, researchers in the chemical and materials science community have put forward general-purpose machine learning methods. The deployment of these methods into workflows of other domains, however, is often burdensome due to the entanglement with domainspecific functionalities. We present the python library scikit-matter that targets domain-agnostic implementations of methods developed in the computational chemical and materials science community, following the scikit-learn API and coding guidelines to promote usability and interoperability with existing workflows.