NSAM- ML Consortium Research Highlights

Scientists and students of the NSAM-ML Consortium have been focusing on a wide range of research topics of interest to NNSA and high-tech industry. The first is on discovering new concepts of new nanosensors useful for the NNSA mission, specifically for monitoring the health of nuclear hardware and other critical mechanical parts found in aerospace industry, military equipment and many other expensive infrastructure. Such sensors require advancement in materials science that allows the discovery of appropriate materials and manufacturing methods; these are required, for instance, to preserve the integrity of nanoscale properties of such materials and to allow proper fabrication and high-throughput processing. NSAM-ML researchers have been focusing on the discovery of new nanomaterials and on understanding their behavior and ways to fabricate them. Manufacturing nanomaterials require a new paradigm, which is emerging and will be one of the labels of the consortium.

The second area of research is on the discovery and investigation of super-hard and radiation-resistant materials, the example of High Entropy Materials is exemplary for our consortium. High-temperature materials for Nuclear Reactors and for Plasma Reactors are being devised.

The third area of research, in which the consortium has produced several excellent papers, covers the production of hydrogen and conversion of carbon dioxide to alternative fuels or high-value materials, both using solar energy. The consortium prides itself on working on solutions for the dual problem of energy sources and pollution while allowing hydrocarbons to be used as high-value materials. The ongoing research in this area is promising significant improvements. The solutions for the energy-pollution dual problem offered to American citizens and the world are critical for the sustainability of our lifestyle.

The fourth area of research is quantum computing and nanomaterials usable for quantum devices. The consortium is contributing to this highly innovative research drive and has published a large number of papers in the field, which puts it in a unique position for serving the future quantum computing technologies and manufacturing materials and devices that serve such a great purpose. Quantum computing will provide future generations with computing power, energy conservation and an economic boom. Scientists envision that one quantum computer will replace hundreds and maybe thousands of nowadays supercomputers; thus, such technology will get rid of the immense amount of consumed energy used to provide computing power to millions and to maintain produced data, which are currently hosted in humongous infrastructure. 

To accelerate the research in the four areas, machine learning is being utilized at different levels. For instance, combined with quantum mechanics calculations, it helps in discovering new high-entropy materials. Also combined with large-scale molecular dynamics, it is utilized for building a database of carbon nanotube (CNT) mechanical properties; we refer to it as the NCCU CNT Database. The new database will help materials engineers and scientists discover new functional nanocomposite materials containing CNTs.

The Consortium scientists have been doing research on the physics of new nanosensors of different types for various applications. One worth mentioning here is a nanosensor that can detect and resolve micron and submicron strain that occurs in mechanical parts due to fatigue, as they are operated, and/or due to the effect of a harsh environment. Such a sensor is ideal for predicting in real time the mechanical failure of expensive and/or critical equipment, by detecting defects at their early stage of formation. Such undertaking is significant for preventing the catastrophic failure of mechanical structures and expensive machinery.

The concept presented on the left shows variants of a 3D printed miniature sensor capable of detecting micron-sized strain components (shear and normal deformations); strain detected at the substrate surface is indicative of the initiation of defects that may lead to failure. The substrate could be a section of a critical part of a nuclear contraption, for instance, the core of a nuclear reactor. Also, in civil applications, the health of a lightweight structure, for instance, the wing of a plane made of composite materials, can be monitored using this new type of sensor.

1. Quantum Dot Arrays

The Consortium also studies the tiny current that tunnels between QDs and between QDs and nanoelectrodes. These elements constitute the essence of a new generation of nanosensors and quantum devices.  The researchers established quantum-based theories on the localization and delocalization of the electrons in the various elements of such devices. They unraveled the conditions for the QDA to be in particular quantum states and the electrons to take certain quantified energies, as well as conditions where quantum chaos occurs. This knowledge enables designing unique nanosensors that can detect the most elemental signal that humans can think of. With our scientific and technology approaches, the scientists are set to access and handle in unique manners quantum information that enables building not only the most sensitive sensors but also many other quantum devices for future quantum computers, or quantum communication networks.

Nanomaterial modeling using high-performance computing helps scientists of the Consortium expedite learning about the physical properties of new materials they develop, their chemistry and the phase transforms that occur during processing, as well as the material implications on the device physics, the material-device codesign, and the fabrication process development and improvement. Furthermore, machine learning (ML) is utilized to push the boundaries of materials and device discovery, development and manufacturing. With fast-moving systematic research and today’s industry dynamics, the Consortium adopted ML as a key method for accelerating the technology and process development, starting with material screening, and going through materials, device and manufacturing optimization, as well as overall performance enhancement.

2.  Vacancy and Vacancy Linear Array Formation in Graphene for Qubit Arrays

Nitrogen-Vacancy (NV) in graphene is proposed by Sahtout and Karoui to be used for making qubits. The generation of ordered vacancy (V) arrays during sustained tensile strain of graphene can be done in a controlled fashion if the V generation dynamics are clarified. Karoui et al. reported on the dynamics of the formation of single V and V arrays in pristine graphene sheet under tensile load. To that end, they performed Molecular Dynamics (MD) for uniaxial loads using LAMMPS and AIREBO potential. First, we investigated AIREBO potential capability to accurately capture the interactions of carbon atoms in graphene. In order to obtain accurate graphene properties, an atomic system of several million atoms has been used. The carbon-based interactions in AIREBO are parameterized with cut-off function that truncates the potential energy between certain inner and outer cut-off interatomic distances. Improper cut-off values resulted in under or over-coordination between atoms, which led to non-physical interactions and properties. Runs on Solo supercomputer with AIREBO “as is”, produced stress-strain curves of graphene similar to those of brittle materials, and mechanical properties were very close to experimental data.

For the first time, we could observe during the plastic regime, the complete dynamics of single V, V array and crack formation in pristine graphene. Note that published work on defect propagation in graphene started with defected graphene sheets [2]. MD simulation performed on Solo with Intel compiler, which allowed minimizing the propagation of round-off errors, consistently showed the formation of single V then V array parallel to the pulling direction. Under the tensile load V cluster in vacancy lines (VL) defect. Each VL is bordered on one side with a single-atom Klein edge defect and on the opposite side with a zigzag edge. The first vacancy (V) appears at 4.2% strain as a result of a stress of 35 GPa in the pulling direction (i.e., // to the zigzag). The V line appears at 5.4% strain, at a stress of 44 GPa with an average interval of 7 C-cycles (equal to 16A spacing). During subsequent loading, a mirror defect image parallel to the original VL emerges and the graphene patch between two VLs keeps sliding through successive “zipping” and “unzipping” of the VLs. Subsequent load increase generates several vacancy line pairs.

3.  Large Scale Fabrication of SiGe Nanodots

Sensors operating at the quantum frontier are built using quantum dot arrays (QDA) grown on high-quality, highly insulating silicon on insulator (SOI) substrates. The quantum dot (QD) is a nanometer-sized object that has unusual behaviors, such as the confinement of electrons in the dot's nanosized volume. The QD itself could be a molecule, cluster of atoms, or the top of a silicon-germanium (SiGe) peak, as illustrated in the atomic force microscopy (AFM) images.

SiGe nanodots form out of nanometers thick SiGe layers and get enhanced by rapid thermal annealing. The formation of these structures is optimized through studying heat and mass transfer at the surface and by appropriately engineering the local stresses and strains.

The Consortium researchers are using rapid thermal chemical vapor deposition to manufacture pyramidal nanodots over large silicon wafers with a uniform size distribution. The AFM image on the right shows an example of such regular pyramidal nanodots, along the Raman spectrum averaged over the scanned area. Quantum dots are formed in the apex of the nanoscale pyramids, where germanium was shown to segregate and strain the material.

1.  Intermediate Band (IB) Induced by Nitrogen Complexes in Silicon

At NCCU, Dr. Karoui’s team has been investigating the modification of silicon to improve its light absorption over the entire solar spectrum and beyond. The work also aims to provide control of the infrared (IR) absorption over a wide spectral range to enable a wide range of new applications, such as future tunable IR sensors, high-resolution IR cameras, and optoelectronic-microelectronic integration for high-speed and high-density microprocessors.

Nitrogen hyperdoping has been proposed to generate an intermediate band (IB) in silicon to enable two-photon absorption for low-energy photons. The enhanced IR absorption adds photogenerated carriers to the carriers generated through band-to-band transitions. Density Functional Theory (DFT) is utilized to calculate electron energy band structure as a function of N concentration, as well as the effect of O and N-O clusters on the energy band structure. A large Si supercell of 800 Si atoms reduces the effects of O atoms, while its size causes the atomic system to exactly match the N and O concentrations found experimentally in hyperdoped Cz silicon. Such large cell led to extensive computation. The DFT calculations have shown that high concentration of N and O impurities in silicon bonded to vacancies (V), or located in interstitial sites as well as clusters of such species significantly transform Si energy band structure. The added oxygen atoms do not fundamentally change the energy band structure. This result has been verified for a wide range of N-related complexes detected by high-resolution FTIR mapping along the depth of the specimen.

2.  Rutile TiO₂ Nanorods Structure, Electronic, and Mechanical Properties

Rutile TiO₂ nanorods crystal structure, electronic and mechanical properties, and elastic and plastic deformation under tensile load were investigated by mean of density functional theory (DFT) and molecular dynamics (MD) simulations. The purpose of this study is to use these nanorods to control the conductivity of polymer nanocomposites for strain piezoelectric sensors.

While the nanorod fillers are intended to control the conductivity of the nanocomposite, they affect the material mechanical properties and its microstructure. Similar phenomenon is being studied in PVDF polymer mixed with a ceramic micro- and nano-powders, by Dr. F. Akram, an NSAM-ML postdoctoral scientist.

The electronic properties were assessed using CASTEP-DFT within the local density approximation (LDA) functional and improved LDA+U functional. The DFT+U method combines the high efficiency of DFT with an explicit treatment of electronic correlation with a Hubbard-like model for a subset of states in the system. For the latter the effect of the U parameter value (0 < U < 10 eV) on the bandgap is analyzed. Rutile TiO₂ tensile test is performed using the large-scale atomic/molecular massively parallel simulator (LAMMPS) and TiO MEAM potential.

Rutile TiO₂ has a tetragonal structure with space group D14-4h P42/mnm with Patterson symmetry P4/mmm. For the primitive unit cell (see the above figure) DFT calculated cell parameters are a=4.594 Å, b=4.594 Å and c=2.959 Å, which are in good agreement with reported values. The calculated diffraction peaks at 2𝛳, 27.45, 36.15, 39.05, 41.3, 44.05 are in agreement with measured 27.4, 36.1, 39.2, 41.2 , and 44.0 XRD peaks, (which correspond respectively to (110), (101), (200), (111) and (210) crystal planes of rutile TiO₂. CASTEP LDA gave an indirect fundamental bandgap of 1.953 eV (Γ-R) and a direct bandgap of 1.954 eV (Γ- Γ) for TiO₂ rutile well below the experimental range 3.0-3.1 eV expected for rutile TiO₂. Rutile TiO₂, the direct bandgap Γ- Γ is very close to the indirect bandgap Γ-R, indicating a quasi-direct bandgap character. CASTEP LDA+U Hubbard-like correction term (U) substantially improved the accuracy of the calculated bandgap.

We found that a value of U=8 eV for rutile TiO₂ gave the closer bandgap value to the experimental one, 2.771 eV for the indirect fundamental bandgap and 2.779 eV for the direct bandgap (see above figure). The significant mixing of O-2p states and Ti-3d states in both the σ and σ* bands in the density of state (DOS) is direct evidence of a very strong hybridization between O-2p and Ti 3d states, see the DOS in the above figure. As shown in the figure below, the stress–strain curves TiO2 rutile nanorods at various strain rates exhibit unusual behavior. At low strain rate we observe three deformation stages before fracture: elastic deformation, plastic deformation and strain hardening in the plateau region.

At high strain rate, 1E10 /s, the onset of plastic deformation occurs before the 2% strain is reached, while for lower strain rates it happens around 6% strain. We also observed strain hardening for a short period of time between 2% and 6% strain followed by a phase transform of the crystal. A Young’s modulus of about 277 GPa has been deduced in the range of 160-380 GPa reported values for TiO₂ rutile. TiO₂ rutile fracture as brittle materials and at low strain rates, fracture will happen much later (between 27% and 32% strain) while at high strain rate it will occur earlier (between 21% and 24%). The stress-strain curves at various strain rates show that the flow stress and work-hardening of rutile TiO₂ is dependent on the strain rate.

3.  Hybrid Piezo-Ferroelectric Nanomaterials

The research on sensors manufactured by 3D printing is connected to the Consortium project on piezoelectric nanocomposite materials. These are suitable for making the strain sensor, as well as developing new processes to fabricate a complex structure in the sensor core. An advanced manufacturing approach consists of 3D printing the nanosensor directly on mechanical parts that need to be monitored.

To apply advanced manufacturing principles, the invention of new fabrication methods (e.g., digital, high throughput, low cost, …) yet efficient in producing new, highly performant materials is required. For instance, a research team from the Consortium has invented a new class of materials for flexible devices, and new fabrication methods. The materials are useful for miniature sensors and the fabrication method is suitable for large sensor arrays. The material is a nanocomposite of special ceramic nanoparticles blended with ferroelectric polymers. The team has been working on improving the material towards stronger piezoelectric properties.

Likewise, the Consortium is working on developing new artificial piezoelectric materials made of organized nano-objects. The organization of nano-objects is a challenging research problem, but it is essential to solve it if we want to obtain extremely amplified material propriety. Enabling such organized nanomaterials will result in a strong group response of the filler, the active component blended with polymers.

4. High performance ferroelectric and piezoelectric of lead-free ceramics

Karoui’s team has strong interest in studying BiFeO33-BaTiO₃ ceramics for advanced manufacturing, as it is promising a wide range of future electronic applications. Nb-modified lead-free xBi_1.03FeO₃-(1-x)BaTi1-xNbxO₃ ceramics are synthesized by solid-state reactions combined with quenching. The ferroelectric and piezoelectric properties process are investigated. Nb modification of the material induced a phase change from the rhombohedral and tetragonal mixed-phase to the pseudocubic phase. This chemical modification transforms the long-range ferroelectric state to a short-range relaxor ferroelectric state. As a result, an excellent piezoelectric performance (high d33 and d33* are observed) with high a Curie temperature (TC »473 °C) and low piezoelectric strain hysteresis, 20% were achieved. The high piezoelectric response is suggested to be related to multiple effects such as crystal structure of morphotropic phase boundary and phase fraction, an optimum grain size (»5 mm), high lattice distortion (cT/aT = 1.01), and soft-ferroelectric effect induced by the donor-doping. Furthermore, high d33* at low electric field in the range 25 kV/cm and the temperature-insensitive piezoelectric response suggest highly promising lead-free BF-BT ceramics.

5. The NCCU Carbon Nanotube Database, for Machine Learning

The lack of a complete Carbon nanotubes (CNTs) database of physical properties of CNTs limits the extent of subsequent research on CNTs, slows the engineering of new and composite materials, and leads to wasted research findings and engineering resources. The primary goal of this student team project is to generate, using molecular dynamics (MD); a database of CNT properties. We aim to provide complete CNT data sets that enable Machine Learning (ML) work. CNT atom coordinates with varying chirality and lengths are generated using VMD software, these are used as input data for a simulated tensile test that allows determining the elastic modulus, the yield strength, the ultimate tensile strength, and the fracture strength of the CNT. The data is generated using the Large-scale Atomic/Molecular Massively Parallel (LAMMPS) software run on the supercomputer Bridges-2, thanks to Dr. Karoui’s XSEDE User Grant. LAMMPS output files of the tensile tests are analyzed using in-house developed computer tools and Ovito software.

The developed database will be used to train future ML models, which will be used to design new CNT based composite materials. One of the studies carried out by the students established the independence of MD results of CNT length which supports the hypothesis of the local character of CNT mechanical properties. More studies have been devised to constitute an excellent platform for student education in advanced physics and materials science, training on High Performance computing (HPC) and supercomputers, and research on nanomaterials for future applications.

1. Electrocatalytic production of hydrogen

Burning fossil fuels for energy application is the leading cause of global warming and their inadequate availability is forcing us to look for clean energy sources. Hydrogen is a potential candidate to replace fossil fuels as a green energy source for future generations. Dr. Kumar’s team at ECSU focuses on electrocatalytic production of hydrogen by cathodic half-reaction on a suitable metal catalyst surface. Platinum (Pt) is the best catalyst to produce hydrogen via electrochemical hydrogen evolution reaction (HER). High cost and scarcity of Pt requires developing an efficient catalytic system with reduced Pt loading and maintaining performance alike bulk Pt metal is in the center of progressive hydrogen fuel economy. Our group is working to develop novel electrocatalysts for efficient HER. Our recent study on the electrochemical performance of sputter deposited molybdenum (Mo) thin film with low loading of dispersed Pt for HER displayed promising results comparable to bulk Pt as shown figure 1. Our two-compartment three electrode electrochemical cell combined with in situ differential electrochemical mass spectroscopy (DEMS) tool enable us to quantify products accurately. Excellent stability and high catalytic activity of our Pt dispersed Mo thin film makes it a promising catalyst for HER and other electrochemical applications.

Figure (a) compares HER polarization curves of Pt dispersed Mo thin film and other electrocatalysts. Inset displays overpotential to achieve current density of 10 mA cm-2; (b) Linear sweep voltammetry curve (black) of Pt dispersed Mo thin film electrocatalyst and corresponding DEMS result (blue) indicating hydrogen production.

2. Electrochemical reduction of carbon dioxide

Carbon dioxide emission into the atmosphere can be solved by solar energy driven electrochemical conversion. In this context, Dr. Kumar’s team is converting CO₂ into value-added fuels and chemicals resulting in environment friendly energy production. In order to use this technology at large scale, catalysts with high energetic efficiency are required. The team has been optimizing electrocatalysts capable of efficiently converting CO₂ into fuels and chemicals with high selectivity. Scientists focus on developing novel nanostructured electrocatalysts which can perform efficient electrochemical conversion of CO₂. In situ DEMS and gas chromatography (GC) techniques are utilized to quantify the reaction products. Silver micropillar structure synthesized utilizing DC magnetron sputtering technique combined with photolithography process are produced, these nanostructures showed outstanding CO₂ conversion performance as displayed in the voltammetry curves of bulk silver and silver micropillar catalysts for CO₂ conversion. Inset displays optical microscope image of silver micropillar structure.

3. Electrochemical nanosensors

Electrochemical nanosensors that detect biological and chemical species represent a revolutionary new generation of medical devices. For instance, continuous blood glucose monitoring is critical to detecting and treating patients in the early stages of diabetes. Detection of volatile organic compounds (VOCs) generated in the human body could provide reliable and valuable indication of human health. If VOCs are accurately detected and identified, by examining the exhaled breath early stage detection of lung cancer, for example, can be easily achieved by examining the exhaled breath.  In that respect, the ECSU has been developing sensors with high selectivity, excellent reproducibility, reusability, and long-term stability.

The below figure provides in (a) an amperometric response of the Cu-Ni thin film electrode to successive addition of glucose in 0.1M NaOH at an applied potential of 0.65 V, (b) amperometric response of the Cu-Ni thin film electrode to interferences ascorbic acid, dopamine and uric acid; (c) relative sensing pattern for Mo3P/MWCNT sensor, and (d) relative amplitude versus concentration of analyte gas. The relative sensing pattern of VOCs detected by using our Mo3P/MWCNT sensor is reproducible.

This data is some of our recent results on non-enzymatic glucose biosensors prepared with heteroatomic Cu-Ni thin films using RF magnetron sputtering technique has displayed excellent sensitivity and selectivity.

Another type of sensor under investigation is built from carbon nanotubes (CNTs) and is being spearheaded by the ECSU research group. These tiny cylindrical carbon objects are visible with electron microscopes, they have shown formidable properties. The research will lead to employing these CNTs to make sensors that can detect a few molecules of gas in the atmosphere.

1. Fabrication of High Entropy Materials

On another front, the Consortium focusing on the safety of nuclear infrastructure has been developing refractory high-entropy alloys, such as the alloy C_0.1Cr₃Mo_11.9Nb₂₀Re₁₅Ta₃₀W₂₀.

Superalloys are highly demanded in many sectors, including but not limited to nuclear, aerospace, defense, and various industries. These materials can withstand extremely harsh conditions, such as heat and radiation. An example of the microstructure of the as-cast alloy is shown below. The microstructure and the local composition of these materials are the subjects of extensive research aiming at maximizing the material resistance to thermo-mechanical stresses and radiations. The superalloy materials are analyzed with state-of-the-art analytical tools, like in the following micrographs obtained using electron dispersive spectroscopy; in an electron microscope. The analysis is done in selected areas. The false-colored maps show the distribution of the chemical elements constituting the alloy, which turned out to be microstructure-dependent. The maps are highly correlated, except for chromium (Cr), which appears evenly distributed and does not follow the alloy microstructure.

The analysis is done in selected areas. The false colored maps show the distribution of the chemical elements constituting the alloy, which turned out to be microstructure dependent. The maps are highly correlated, except for chromium (Cr), which appears evenly distributed and does not follow the alloy microstructure.

2. SUBR Database for High Entropy Materials

One of the research goals of Dr. Yang’s team, at SUBR, is to develop a dataset and machine learning (ML) tools to predict performance of high entropy alloys (HEA) under normal and extreme conditions (high pressure, high temperature and radiation).

The database is being built using DFT and phase diagram calculations combined with experimental data, some of which were developed by the SUBR research team. The dataset features are checked before training machine learning (ML) models. Neural network and random forest models are being developed for the ML.

1. Biovia Materials Studio, by Dassault Systems.
2. J. Hubbard, Proc. R. Soc. London, Electron Correlations in Narrow Energy Bands, Proc. R. Soc. Ser. A 276, 238 (1963).
3. W.J. Joost, S. Ankem, M.M. Kuklja, A modified embedded atom method potential for the titanium-oxygen, System Modeling and Simulation in Materials Science and Engineering, 23, pp. 015006 (2015).
4. Y.M Kim, B.J. Lee, M.I. Baskes, Modified embedded-atom method interatomic potentials for Ti and Zr, Physical Review B, 74, pp. 014101 (2006).
5. M.I Baskes, Modified embedded atom method calculations of interfaces, Tech. Rep. SAND-96-8484C, CONF-9603153-1, Sandia National Laboratories, Livermore, USA.
6. M. E. Arroyo-de Dompablo, A. Morales-García, and M. Taravillo, DFT+U calculations of crystal lattice, electronic structure, and phase stability under pressure of TiO2 polymorphs, The Journal Of Chemical Physics, 135, 054503 (2011).
7. Zhaoyue Liu, Xintong Zhang, Shunsuke Nishimoto, Ming Jin, Donald A. Tryk, Taketoshi Murakami, and Akira Fujishima, Anatase TiO2 Nanoparticles on Rutile TiO2 Nanorods: A Heterogeneous Nanostructure via Layer-by-Layer Assembly, Langmuir, 23, 10916-10919 (2007).
8. Chinedu E. Ekuma and Diola Bagayoko, Ab-initio Electronic and Structural Properties of Rutile Titanium Dioxide, Jpn. J. Appl. Phys. 50 101103 (2011).
9. Isaak, D., Carnes, J., Anderson, O. et al. Elasticity of TiO2 rutile to 1800K. Phys Chem Min 26, 31–43 (1998).
10. J. B. Wachtman, Jr., W. E. Tefft, and D. G. Lam Jr., Elastic Constants of Rutile (TiO2), Journal of Research of the National Bureau of Standards-A. Physics and Chemistry, Vol. 66A, No. 6, 465 (1962).
11. Sergey I. Lukyanov, Andrei V. Bandura and Robert A. Evarestov, Young's modulus and Poisson's ratio for TiO2-based nanotubes and nanowires: modeling of temperature dependence, RSC Adv., 6, 16037 (2016).