Dr Chi Chen’s swansong work in our group, “A Universal Graph Deep Learning Potential for the Periodic Table” is now published in Nature Computational Science! Interatomic potentials (IAPs), which describe the potential energy surface of atoms, are a fundamental input for atomistic simulations. However, existing IAPs are either fitted to narrow chemistries or too inaccurate for general applications. In this work, we combine graph neural networks with traditional 3-body interactions to develop a flexible, yet accurate architecture for machine learning of materials properties. Using the massive database of structural relaxations performed by the Materials Project over the past ten years, we train a universal IAP for 89 elements of the periodic table with broad applications in structural relaxation, dynamic simulations and property prediction of materials across diverse chemical spaces. Using the new capabilities of the M3GNet universal IAP, we are proud to launch matterverse.ai, a ML database of yet-to-be-synthesized materials. Matterverse.ai currently contains about 31 million hypothetical crystal structures, of which about 1.8 million materials were identified to be potentially stable. The database also provides ML properties using state-of-the-art multi-fidelity MEGNet models, such as experimental, HSE and PBE band gaps, bulk and shear moduli, etc. Check out the article here. […]
Congrats to Hideyuki Komatsu, our visiting scientist from Nissan, on his first author work “Interfacial Stability of Layered LiNixMnyCo1−x−yO2 Cathodes with Sulfide Solid Electrolytes in All-Solid-State Rechargeable Lithium-Ion Batteries from First-Principles Calculations” published in the Journal of Physical Chemistry C! In this work, we explore the relationship between the composition of layered LiNixMnyCo1−x−yO2 (NMC) cathodes and interfacial stability in all-solid-state lithium-ion batteries. A key insight is that the broader commercial trend towards high Ni content to reduce cost leads to significantly more reactive interfaces with the Li6PS5Cl argyrodite solid electrolyte. This suggests that current efforts to reduce the Co content in cathodes may compromise potential applications in all-solid-state architectures. Nevertheless, we find that common SEI phases such as Li2CO3, surface phases such as NiO, and oxide buffer layers such as LiNbO3 can provide effective protection between NMC and LPSCl. Check out the work here.
Congratulations to Manas Likhit Holekevi Chandrappa, Ji Qi, Chi Chen and Swastika Banerjee on the publication of “Thermodynamics and Kinetics of the Cathode−Electrolyte Interface in All-Solid-State Li−S Batteries” in the Journal of the American Chemical Society! Lithium−sulfur batteries (LSBs) use cheap and abundant sulfur in place of expensive metal-based cathodes. Using a solid electrolyte in place of traditional liquid electrolytes mitigates polysulfide shuttling, a key impediment to LSB commercialization. In this work, we present a comprehensive study of the thermodynamics and kinetics of the cathode−electrolyte interface in all-solid-state LSBs. Using DFT calculations, we show that among the major solid electrolyte chemistries (oxides, sulfides, nitrides, and halides), sulfides are the most stable solid electrolytes against the S cathode, as well as the most promising buffer layers if the use of other SE chemistries is desired. Finally, MD simulations with an accurate machine learning interatomic potential revealed that the most stable Li3PS4(100)/S interfaces form 2D channels with lower activation barriers for Li diffusion. These results provide critical new insights into the cathode−electrolyte interface design for next-generation all-solid-state LSBs. We gratefully acknowledge Nissan Motor Inc and Nissan North America for their generous support for this work! This work has been published open access […]
Congratulations to Jasleen on her first co-author paper on “An Optoelectronic Heterostructure for Neuromorphic Computing: CdS/V3O5” in Applied Physics Letters! Nonvolatile resistive switching is one of the key phenomena for emerging applications in optoelectronics and neuromorphic computing. However, the stochastic nature of the ion migration can be an impediment for the device robustness and controllability, with uncontrolled variations of high and low resistance states or threshold voltages. In this work, we report an optically induced resistive switching based on a CdS/V3O5 heterostructure. V3O5 is known to have a second order insulator to metal transition around 415 K, with an electrically induced threshold switching at room temperature. Upon illumination, the direct transfer of the photoinduced carriers from the CdS into V3O5 produces a nonvolatile resistive switching at room temperature. Jasleen’s contribution is in using DFT calculations to understand the defects present in V3O5 and the effects of electron doping. We show that electrons (generated by CdS under illumination) injected in V3O5 are trapped in a deep state, slowing the “low” temperature relaxation rate. For the LT phase (T < 340 K), the photoexcited electrons trapped into the oxygen vacancy are unable to overcome the barrier, and therefore, no relaxation is observed. […]
Weike’s swansong project in our group on “A Universal Machine Learning Model for Elemental Grain Boundary Energies” has just been published in Scripta Materialia! The grain boundary (GB) energy has a profound influence on the grain growth and properties of polycrystalline metals. Here, we show that the energy of a GB, normalized by the bulk cohesive energy, can be described purely by four geometric features to within a mean absolute error of 0.13 Jm-2 . More importantly, this universal GB energy model can be extrapolated to the energies of high Σ GBs without loss in accuracy. These results highlight the importance of capturing fundamental scaling physics and domain knowledge in the design of interpretable, extrapolatable machine learning models for materials science. Check out this work here.
Efficient near-infrared (NIR) LEDs are used in many applications, including medical diagnostics, food detection, security monitoring, and machine vision. In this collaborative work with the group of Prof Rong-jun Xie at Xiamen University published in Matter, Mahdi Amachraa developed descriptors of the Eu(II)-host interactions to predict the 5d-to-4f energy gap with a RMSE of 7.0 nm. By incorporating this predictor into a high-throughput screening of 223 nitride materials in the Inorganic Crystal Structure Database, we identified and experimentally validated (Sr,Ba)3Li4Si2N6:Eu(II) with NIR emissions of 800- 830 nm and high quantum efficiencies (QEs) of 30%-40%. This NIR emitter has 3x more power than prevailing NIR emitters. We demonstrate that the ultralong emission wavelength and high QE stem from a coordinated energy transfer and an optimized electronic delocalization around Eu(II). Check out this work here.
Our collaborative work with the group of Prof Claire Xiong on “Electrochemically induced amorphous-to-rock-salt phase transformation in niobium oxide electrode for Li-ion batteries” has been published in Nature Materials! In this work, we report a nanostructured rock-salt Nb2O5 electrode formed through an amorphous-to-crystalline transformation during repeated electrochemical cycling with Li+. This electrode can reversibly cycle three lithiums per Nb2O5, corresponding to a capacity of 269 mAh/g at 20 mA/g, and retains a capacity of 191 mAh/g at a high rate of 1 A/g. The main contribution from Yunxing Zuo of the Materials Virtual Lab is using DFT computations to show that the cubic rock-salt framework promotes the percolation of low-energy migration paths. We also develop a computable metric to identify other transition metal oxides with a likelihood of rock-salt formation. Our work suggests that inducing crystallization of amorphous nanomaterials through electrochemical cycling is a promising avenue for creating unconventional high-performance metal oxide electrode materials. Check out the publication here.
Mahdi’s collaborative work with the group of Prof Rong-jun Xie on “MxLa1−xSiO2−yNz (M = Ca/Sr/Ba): Elucidating and Tuning the Structure and Eu2+ Local Environments to Develop Full-Visible Spectrum Phosphors” has just been published in Chemistry of Materials! The local environments of rare-earth activators have profound effects on the luminescent properties of phosphors. Here, we elucidate the crystal structure of the LaSiO2N phosphor host using a combination of density functional theory calculations and synchrotron Xray diffraction. We determine that LaSiO2N crystallizes in the monoclinic C2/c instead of the hexagonal P6̅c2 space group. To improve the luminescence performance, divalent cations M (M = Ca/Sr/Ba) were introduced into LaSiO2N to eliminate Eu3+. A family of apatite M1+xLa4−xSi3O13−x/2:Eu2+ (x ∼ 1.5, M = Ca/Sr/Ba) phosphors was further developed with unprecedented ultra-broadband (290 nm) emission spectra and excellent thermal stability. Detailed local environment investigations reveal that the formation of oxygen vacancies within and beyond the first shell environment of Eu2+ is responsible for the redshift and broadening of the emission spectra via geometrical alteration of the Eu2+ local environment. This work provides new insights for understanding and optimizing the luminescence of rare-earth phosphors. Check out this work here.
Graph networks are an extremely powerful deep learning tool for predicting materials properties. However, a critical weakness is their reliance on large quantities of training data. In this work published in npj Computational Materials, Dr Chi Chen shows that pre-trained MEGNet formation energy models can be effectively used as “encoders” for crystals in what we call the AtomSets framework. The compositional and structural descriptors extracted from graph network deep learning models, combined with standard artificial neural network models, can achieve lower errors than the graph network models at small data limits and other non-deep-learning models at large data limits. AtomSets also transfer better in a simulated materials discovery process where the targeted materials have property values out of the training data limits, require minimal domain knowledge inputs and are free from feature engineering. Check out this work here.