Practical needs in technology capability assessment for extremely low-energy neuromorphic computing is addressed via a novel development/analysis concept integrating atomic-level material modeling, statistical simulations of charge transport in a device material stack and verification of the modeling scheme against measurements emulating circuitry operation conditions for applications in specific neural networks (NN). This multi-scale concept - from materials to applications - directly links materials to their electrical properties, and the latter to NN algorithms. Such link enables identifying structural features controlling device characteristics and the range of operation conditions delivering performance targets for a given technology implementation. In comparison to widely employed memristor analyses primarily based on TCAD-type methodology with adjustable phenomenological parameters, the proposed approach allows to deliver feedback on favorable material compositions and cell architecture/dimensions to modify memristor fabrication process. Implementation of this technology evaluation approach to carbon nanotube (CNT) memristors enables identifying structural and operation conditions delivering optimal performance ahead of actual circuitry fabrication.

Density functional theory (DFT) with generalised gradient approximation (GGA) functionals is commonly used to predict defect properties in 2D transition metal dichalcogenides (TMDs). Since GGA functionals often underestimate band gaps of semiconductors and incorrectly describe the character of electron localisation in defects and their level positions within the band gap, it is important to assess the accuracy of these predictions. To this end, we used the non-local density functional Perdew-Burke-Ernzerhof (PBE)0-TC-LRC to calculate the properties of a wide range of intrinsic defects in monolayer WS2. The properties, such as geometry, in-gap states, charge transition levels, electronic structure and the electron/hole localisation of the lowest formation energy defects are discussed in detail. They are broadly similar to those predicted by the GGA PBE functional, but exhibit numerous quantitative differences caused by the degree of electron and hole localisation in charged states. For some anti-site defects, more significant differences are seen, with both changes in defect geometries (differences of up to 0.5 angstrom) as well as defect level positions within the band gap of WS2. This work provides an insight into the performance of functionals chosen for future DFT calculations of TMDs with respect to the desired defect properties.

A simulation package for CNT memory cells is developed, based on computational modeling of both the mesoscopic structure of carbon nanotube films and the electrical conductivity of inter-CNT contacts. The developed package enables the modeling of various electrical measurements and identification of a range of operation conditions delivering desirable device characteristics. This approach opens the path for optimization of the CNT fabric to meet performance requirements.

Root-cause of instability in carbon nanotubes memristors is analyzed employing ultra-short pulse technique in combination with atomic-level material modeling. Separating various factors affecting switching operations allowed to identify structural features and operational conditions leading to improved cell characteristics.

Predicting the degree of localization and calculating the trapping energies of polarons in insulators by density functional theory (DFT) is challenging. Hybrid functionals are often reparametrized to obtain accurate results and the a priori selection of these parameters is still an open question. Here we test the accuracy of several range-separated hybrid functionals, all reparametrized to produce an accurate band gap, by calculating the charge transition levels (CTLs) of experimentally well-studied hole polaron defect centers in MgO. We show that the functional with screened long-range exact exchange is moderately but consistently more accurate than functionals which do not include long-range exact exchange. We provide evidence that the source of the improved accuracy is the eigenvalue associated with the valence band maximum of the bulk material. We discuss the extent to which this accuracy relates to Koopmans’ compliance of the defect energy level.

Charged defects are often studied within the periodic density functional theory (DFT), but this introduces strong finite-size artifacts. In this work, we develop an electrostatic image interaction correction (IIC) method based on the direct solution of the Poisson equation for charge models constructed directly from DFT calculations. These IICs are found to be detail-insensitive, depending almost entirely on bulk dielectric properties. As these IICs are not able to fully explain the observed finite-size scaling, we explore potential alignment in detail and introduce a novel decomposition to separate out different contributions. We find that the two main sources of potential alignment are defect image interactions and changes in the number of atoms present in the supercell. This first effect is accurately predicted by the periodic part of our IIC. The second contribution is unrelated to the IIC and justifies the common observation that the magnitude of finite-size dependence can strongly vary between vacancy and interstitial defects. It can be approximately predicted using atomic radius, but is strongly sensitive to the pseudopotential employed. Combined, these developments provide a new justification for known finite-size scaling rules. Our results suggest that for cubic supercells, the Lany-Zunger IIC, combined with simplified potential alignment between neutral systems, can yield accurate corrections in spite of the simplicity of the approach.

The most direct signature of electron localisation is the tendency of an electron in a many-body system to exclude other same-spin electrons from its vicinity.By applying this concept directly to the exact many-body wavefunction, we find that localisation can vary considerably between different ground-state systems, and can also be strongly disrupted, as a function of time, when a system is driven by an applied electric field.We use this measure to assess the well-known electron localisation function (ELF), both in its approximate single-particle form (often applied within density-functional theory) and its full many-particle form.The full ELF always gives an excellent description of localisation, but the approximate ELF fails in time-dependent situations, even when the exact Kohn-Sham orbitals are employed.

Electron localization is the tendency of an electron in a many-body system to exclude other electrons from its vicinity. Using a new natural measure of localization based on the exact manyelectron wavefunction, we find that localization can vary considerably between different ground-state systems, and can also be strongly disrupted, as a function of time, when a system is driven by an applied electric field. We use our new measure to assess the well-known electron localization function (ELF), both in its approximate single-particle form (often applied within density-functional theory) and its full many-particle form. The full ELF always gives an excellent description of localization, but the approximate ELF fails in time-dependent situations, even when the exact Kohn-Sham orbitals are employed.

We introduce a new functional for simulating ground-state and time-dependent electronic systems within density-functional theory. The functional combines an expression for the exact Kohn-Sham (KS) potential in the limit of complete electron localization with a measure of the actual localization. We find accurate self-consistent charge densities, even for systems where the exact exchange-correlation potential exhibits non-local dependence on the density, such as potential steps. We compare our results to the exact KS potential for each system. The self-interaction correction is accurately described, avoiding the need for orbital-dependent potentials.