While defects are undesirable for the reliability of electronic devices, particularly in scaled microelectronics, they have proven beneficial in numerous quantum and energy-harvesting applications. However, their potential for new computational paradigms, such as neuromorphic and brain-inspired computing, remains largely untapped. In this study, we harness defects in aggressively scaled field-effect transistors based on two-dimensional semiconductors to accelerate a stochastic inference engine that offers remarkable noise resilience. We use atomistic imaging, density functional theory calculations, device modeling, and low-temperature transport experiments to offer comprehensive insight into point defects in WSe2 FETs and their impact on random telegraph noise. We then use random telegraph noise to construct a stochastic encoder and demonstrate enhanced inference accuracy for noise-inflicted medical-MNIST images compared to a deterministic encoder, utilizing a pre-trained spiking neural network. Our investigation underscores the importance of leveraging intrinsic point defects in 2D materials as opportunities for neuromorphic computing.
Silicon nitride (Si3N4) is an extensively used material in the automotive, aerospace, and semiconductor industries. However, its widespread use is in contrast to the scarce availability of reliable interatomic potentials that can be employed to study various aspects of this material on an atomistic scale, particularly its amorphous phase. In this work, we developed a machine learning interatomic potential, using an efficient active learning technique, combined with the Gaussian approximation potential (GAP) method. Our strategy is based on using an inexpensive empirical potential to generate an initial dataset of atomic configurations, for which energies and forces were recalculated with density functional theory (DFT); thereafter, a GAP was trained on these data and an iterative re-training algorithm was used to improve it by learning on-the-fly. When compared to DFT, our potential yielded a mean absolute error of 8 meV/atom in energy calculations for a variety of liquid and amorphous structures and a speed-up of molecular dynamics simulations by 3–4 orders of magnitude, while achieving a first-rate agreement with experimental results. Our potential is publicly available in an open-access repository.
Silicon nitride films are widely used as the charge storage layer of charge trap flash (CTF) devices due to their high charge trap densities. The nature of the charge trapping sites in these materials responsible for the memory effect in CTF devices is still unclear. Most prominently, the Si dangling bond or K-center has been identified as an amphoteric trap center. Nevertheless, experiments have shown that these dangling bonds only make up a small portion of the total density of electrical active defects, motivating the search for other charge trapping sites. Here, we use a machine-learned force field to create model structures of amorphous SiN by simulating a melt-and-quench procedure with a molecular dynamics algorithm. Subsequently, we employ density functional theory in conjunction with a hybrid functional to investigate the structural properties and electronic states of our model structures. We show that electrons and holes can localize near over- and under-coordinated atoms, thereby introducing defect states in the band gap after structural relaxation. We analyze these trapping sites within a nonradiative multi-phonon model by calculating relaxation energies and thermodynamic charge transition levels. The resulting defect parameters are used to model the potential energy curves of the defect systems in different charge states and to extract the classical energy barrier for charge transfer. The high energy barriers for charge emission compared to the vanishing barriers for charge capture at the defect sites show that intrinsic electron traps can contribute to the memory effect in charge trap flash devices.