In this paper, we propose a U-net architecture that integrates a residual skip connections and recurrent feedback with EfficientNet as a pretrained encoder. Residual connections help feature propagation in deep neural networks and significantly improve performance against networks with a similar number of parameters while recurrent connections ameliorate gradient learning. We also propose a second model that utilizes densely connected layers aiding deeper neural networks. EfficientNet is a family of powerful pretrained encoders that streamline neural network design. The proposed networks are evaluated against state-of-the-art deep learning based segmentation techniques to demonstrate their superior performance.
A microelectromechanical system (MEMS) device might function perfectly well in the controlled environment in which it has been created. However, the device can be a real viable product only after it has been fabricated with proven performance in a package. As such, the assembly yield of a MEMS package is often a challenging target to meet. The design and fabrication of a free-floating membrane on a flexible substrate to enable easy and cost-effective packaging of MEMS devices is examined. Since standard MEMS fabrication processes are designed for rigid substrates, several process modifications were required to handle flexible substrates. The adaptation of each fabrication process has been documented. Furthermore, detailed information regarding the selection of compatible materials, as well as incompatibilities that were encountered, has been presented to aid future researchers in developing processes for flexible substrates.
A theoretical model for describing the bias-dependent transient behavior of dark current in multilayer amorphous
selenium (a-Se) detectors has been developed by solving the trapping rate equations and Poisson's equation in the a-Se
layer. The transient dark currents in these detectors are measured and the proposed dark model is compared with the
measured data. The model shows a very good agreement with the experimental results. It has been found that the dark
current is mainly controlled by the Schottky emission of holes from the metal/a-Se contact. The space charge build-up
due to the hole injection and trapping in the blocking layer reduces the internal field at the metal/a-Se interface of
positive side and thus the dark current eventually is limited by the space charge. It has been found that the electric fields
at the metal contacts reduce to 20-30% of the applied field (applied voltage/thickness). The comparison of the model
with the experimental data estimates some important properties (e.g., trap center concentrations, space charges, and
effective barrier heights) of the blocking layers of the multilayer detectors. The dependence of the X-ray sensitivity of
multilayer a-Se X-ray imaging detectors on repeated X-ray exposures is studied by considering accumulated trapped
charges and their effects (trap filling, recombination, electric field profile, electric field dependent electron-hole pair
creation), the carrier transport in the blocking layers, X-ray induced metastable deep trap center generations, and the
effects of dark current. We simultaneously solve the continuity equations for both holes and electrons, trapping rate
equations, and the Poisson's equation across the photoconductor for a step X-ray exposure by the Backward Euler finite
difference method. The theoretical model shows a very good agreement with the experimental relative sensitivity versus
cumulative X-ray exposure characteristics. The electric field distribution across the multilayer detector and the dark
current density under repeated exposures are also estimated.