Reinforcement Learning-guided Long-timescale Simulation of Hydrogen Transport in Metals

8/20/2025 | Ju Li (MIT)

This paper explores new methods for simulating how atoms diffuse in complex metal alloys over longer timescales. By using reinforcement learning, two techniques were developed to enhance simulation capabilities: one for tracking transition kinetics and another for sampling low-energy states. The researchers tested these methods by studying hydrogen movement in pure metals and a specific alloy, showing that they could reveal unexpected movement patterns. Overall, these advancements improve the accuracy of simulations compared to traditional approaches.

Thousands of Conductance Levels in Memristors Integrated on CMOS

8/20/2025 | Ju Li (MIT)

Researchers have made significant advances with memristors, a type of electronic component that can remember previous charge flows. They achieved 2,048 distinct conductance levels in memristors, greatly surpassing previous records. This enhancement helps improve machine learning and AI efficiency. By studying the causes of conductance fluctuations, they developed a method to enhance precision in memristor operations. These findings pave the way for better memristor technology in commercial applications, particularly in edge computing for AI.

Designing Contact Independent High-performance Low-Cost Flexible Electronics

8/19/2025 | Oana Jurchescu (Wake Forest University)

Researchers used simulations to find a way to create organic transistors with high mobility, independent of contact work function. This led to the design of affordable, high-performance transistors made entirely from solution-deposited materials, suitable for flexible surfaces. By testing over 2000 virtual designs, they minimized the need for physical prototypes, ultimately achieving transistors with mobility higher than 5 cm²/V/s, marking a significant advancement in all-solution-processed devices.

Computational Fluid Dynamic Modeling of Methane-Hydrogen Mixture in Pipelines

8/18/2025 | T. A. Venkatesh (Stony Brook University)

A recent study focused on the blending of hydrogen with natural gas as a step towards carbon neutrality. Using computer modeling, researchers examined how these gas mixtures behave in pipelines. Findings showed that transporting methane-hydrogen blends requires more energy depending on factors like hydrogen volume, pipe size, and surface roughness. The study also revealed that the gas mixture forms a specific flow pattern, with denser methane flowing along the walls and lighter hydrogen in the center.

Designing Optical Materials Emit Chiral Light using Small-molecule Ionic Isolthatation Lattices (SMILES)

8/13/2025 | Amar Flood (Indiana University)

Researchers have created a new way to design optical materials using small-molecule, ionic isolation lattices (SMILES). They demonstrated that these materials can emit circularly polarized light (CPL) using a special dye called "helicene." The resulting SMILES crystals and nanoparticles show strong light-emitting capabilities, matching or exceeding previous results. This breakthrough can impact fields like 3D displays and bioimaging by translating optical properties from solutions to solid materials. Future developments will focus on improving dyes and crystal structures.

Unique Conductivity Behavior in Water-in-Salt Electrolytes Driven by Ion Clusters

8/12/2025 | Y Z (University of Michigan) and Tao Li (Northern Illinois University)

A new framework has been developed to better understand how ions move in watery electrolyte solutions, which is important for energy storage and biological applications. This approach shifts focus from concentration measures to volume fraction, revealing a consistent peak conductivity at a 37% volume fraction. Research using small-angle X-ray scattering and molecular dynamics shows that tiny ion clusters are key to this behavior. These findings could lead to better designs for electrolytes and enhance related scientific studies.

Stabilizing Graphite Anode in Electrolytes with Nanoscale Anion Networking for High-Rate Lithium Storage

8/12/2025 | Tao Li (Northern Illinois University)

A recent study introduced a new type of electrolyte designed to better support graphite anodes in lithium-ion batteries. By using a concentrated mixture of long-chain lithium salts, the researchers created a nanoscale network that reduces harmful interactions between graphite and solvent molecules. This helps to prevent graphite layer breakdown during charging, even with less stable solvents. These findings could lead to more effective battery designs by overcoming current limitations in electrolyte choices.

Enhanced Synaptic Memory Window and Linear Planar In2Se3 Ferroelectric Junctions

8/4/2025 | Keji Lai (University of Texas – Austin)

Researchers have made progress in improving neuromorphic computing systems inspired by the brain, addressing limitations of traditional computing. They focused on synaptic memristors made from 2D ferroelectric junctions, which show a voltage memory window of 16 V and an impressive on/off current ratio, leading to very low power consumption of 10^-5 W. Their new design achieved nearly 90% accuracy in on-chip training, providing a promising alternative to traditional computing methods while enhancing performance.

Design Rules to Optimize the Intermolecular Long-range Packing of Organic Semiconductor Crystals

8/1/2025 | Oana Jurchescu (Wake Forest University) John Anthony (University of Kentucky)

Researchers have investigated how the structure and interactions within organic semiconductors (OSCs), particularly benzodithiophene (BDT) trimers, can be enhanced through fluorine substitutions. Using density functional theory (DFT), they found that these substitutions improve charge mobility by facilitating hydrogen bonding, which helps the molecule adopt a more stable, planar shape. The study highlights the importance of intermolecular interactions in optimizing OSC properties and offers insights for developing better materials for electronics and other applications.

Predicting the Excited-state Properties of Crystalline Organic Semiconductors using Machine Learning

7/9/2025 | Noa Marom (Carnegie Mellon U.)

A study explored how machine learning (ML) can help predict key properties of organic semiconductors, which are important for electronic devices. Researchers used the SISSO ML algorithm to develop models predicting various excitation energies in polycyclic aromatic hydrocarbon crystals. By training models on a dataset of advanced calculations, they achieved predictions close to established reference values. This approach can effectively narrow down material choices for further, more detailed evaluation, saving time and resources in material discovery.

Research Highlights