Entire Stack Research Articles

Simple and Versatile Platforms for Manipulating Light with Matter: Strong Light-Matter Coupling in Fully Solution-Processed Optical Microcavities.

Planar microcavities with strong light-matter coupling, monolithically processed fully from solution, consisting of two polymer-based distributed Bragg reflectors (DBRs) comprising alternating layers of a high-refractive-index titanium oxide hydrate/poly(vinyl alcohol) hybrid material and a low-refractive-index fluorinated polymer are presented. The DBRs enclose a perylene diimide derivative (b-PDI-1) film positioned at the antinode of the optical mode. Strong light-matter coupling is achieved in these structures at the target excitation of the b-PDI-1. Indeed, the energy-dispersion relation (energy vs in-plane wavevector or output angle) in reflectance and the group delay of transmitted light in the microcavities show a clear anti-crossing-an energy gap between two distinct exciton-polariton dispersion branches. The agreement between classical electrodynamic simulations of the microcavity response and the experimental data demonstrates that the entire microcavity stack can be controllably produced as designed. Promisingly, the refractive index of the inorganic/organic hybrid layers used in the microcavity DBRs can be precisely manipulated between values of 1.50 to 2.10. Hence, microcavities with a wide spectral range of optical modes might be designed and produced with straightforward coating methodologies, enabling fine-tuning of the energy and lifetime of the microcavities' optical modes to harness strong light-matter coupling in a wide variety of solution processable active materials.

MODELING AND IMPLEMENTATION OF DOMAIN EVENTS IN THE DOMAIN-DRIVEN ARCHITECTURE ON THE .NET CORE PLATFORM

A software architecture centered on a domain model can provide significant advantages over other types of architectures in the long-term development and maintenance of systems with complex domain logic. At the same time, domain-driven design and approaches to software implementation of systems are relatively new concepts and continue to develop in application to various platforms, technologies and programming languages, and are of considerable interest to designers and developers. The presented work examines existing approaches to modeling and implementing domain events on the .NET Core platform in domain-driven architecture, which is one of the newest patterns. There are two approaches to implementing domain event behavior: immediate and delayed event propagation. These two approaches are analyzed and their features are described in detail. The implementation of instant propagation of domain events within a command execution transaction in the CQRS architecture is described. This implementation allows you to get rid of external dependencies, achieve purity of domain entities, as it eliminates the need to inject services, repositories in entities, and also prevents memory leaks and is safe for multithreaded use. Schematically depicts the abstract process of an external command entering a domain model, which causes a change in the state of an aggregate and the propagation of side effects with domain events. This process takes into account the capabilities of the Entity Framework object-relational mapping framework to retrieve context objects that have been changed during process. The entire stack of objects involved in this activity is located in the shared process memory, and the interaction occurs in synchronous mode. For the conceptual detection of events and aggregates, the event storming technique is used, the features of which are discussed in the article.

Lesion detection in digital breast tomosynthesis: human reader experiments indicate no benefit from the integration of information from multiple planes.

In digital breast tomosynthesis (DBT), radiologists need to review a stack of 20 to 80 tomosynthesis images, depending upon breast size. This causes a significant increase in reading time. However, it is currently unknown whether there is a perceptual benefit to viewing a mass in the 3D tomosynthesis volume. To answer this question, this study investigated whether adjacent lesion-containing planes provide additional information that aids lesion detection for DBT-like and breast CT-like (bCT) images. Human reader detection performance was determined for low-contrast targets shown in a single tomosynthesis image at the center of the target (2D) or shown in the entire tomosynthesis image stack (3D). Using simulations, targets embedded in simulated breast backgrounds, and images were generated using a DBT-like (50deg angular range) and a bCT-like (180deg angular range) imaging geometry. Experiments were conducted with spherical and capsule-shaped targets. Eleven readers reviewed 1600 images in two-alternative forced-choice experiments. The area under the receiver operating characteristic curve (AUC) and reading time were computed for the 2D and 3D reading modes for the DBT and bCT imaging geometries and for both target shapes. Spherical lesion detection was higher in 2D mode than in 3D, for both DBT- and bCT-like images (DBT: , , ; bCT: , , ), but equivalent for capsule-shaped signals (DBT: , , ; bCT: , , ). Average reading time was up to 134% higher for 3D viewing (). For the detection of low-contrast lesions, there is no inherent visual perception benefit to reviewing the entire DBT or bCT stack. The findings of this study could have implications for the development of 2D synthetic mammograms: a single synthesized 2D image designed to include all lesions present in the volume might allow readers to maintain detection performance at a significantly reduced reading time.

Combining seeded region growing and k-nearest neighbours for the segmentation of routinely acquired spatio-temporal image data

PurposeThe acquisition conditions of medical imaging are often precisely defined, leading to a high homogeneity among different data sets. Nonetheless, outliers or artefacts still appear and need to be reliably detected to ensure a reliable diagnosis. Thus, the algorithms need to handle small sample sizes especially, when working with domain specific imaging modalities.MethodsIn this work, we suggest a pipeline for the detection and segmentation of light pollution in near-infrared fluorescence optical imaging (NIR-FOI), based on a small sample size. NIR-FOI produces spatio-temporal data with two spatial and one temporal dimension. To calculate a two-dimensional light pollution map for the entire image stack, we combine region growing and k-nearest neighbours (kNN), which classifies pixels into fore- and background by its entire temporal component. Thus, decision-making on reduced data is omitted.ResultsWe achieved a F_1 score of 0.99 for classifying a data set as light polluted or pollution-free. Additionally, we reached a total F_1 score of 0.90 for detecting regions of interest within the polluted data sets. Finally, an average Dice’s coefficient measuring the segmentation performance over all polluted data sets of 0.80 was accomplished.ConclusionsA Dice’s coefficient of 0.80 for the area segmentation does not seem perfect. However, there are two main factors, besides true prediction errors, lowering the score: Segmentation mistakes on small areas lead to a rapid decrease in the score and labelling errors due to complex data. However, in combination with the light-polluted data set and pollution area detection, these results can be considered successful and play a key role in our general goal: Exploiting NIR-FOI for the early detection of arthritis within hand joints.

Advancements, Versatility, and Flexibility of Dual-Layer Material System for Advanced Packaging Applications

Integrated circuits (IC) have seen a major shift in development within the past 10-20 years with traditional lithography methods showing a drastic increase in development time for more advanced nodes, as well as an exponential increase in cost to achieve the same performance gains as before. The increase in cost and the decline in development in lithography have resulted in looking at advanced packaging technologies to achieve the same performance gains by changing the way IC design is approached. The look towards advanced packaging technologies for the future in terms of increasing performance for a substantially lower cost has resulted in considering the IC as a system of components working intertwined with each other rather than as individual components. This shift in mindset has resulted in technologies such as system in package (SiP), package on package (PoP), and fan-out wafer-level packaging (FOWLP). One advanced packaging technology that plays a critical role enabling these aforementioned technologies is temporary bond and debond (TB/DB). TB/DB’s crucial role in advanced packaging is due to the enablement of backside processing such as wafer thinning, bumping of the wafer, die stacking, and chemical vapor deposition/physical vapor deposition (CVD/PVD)–type processes by the use of a support carrier wafer. The support carrier wafer also allows for the use of highly warp-prone materials such as epoxy mold compound (EMC), which are critical in FOWLP applications, by reducing the overall warpage of the entire wafer stack. To utilize the support carrier wafer, a robust material solution is required to allow for the wafers to be bonded together and subsequently released after backside processing by one of the primary separation methods of thermal slide, mechanical, or laser debond. Brewer Science has designed and developed a dual-layer temporary bonding system. This system consists of two materials, a thermoplastic layer that is typically coated on the device, and a thermoset layer which is typically coated on the carrier. The materials developed for the dual-layer system demonstrate very good performance in very high temperature applications, EMC wafer handling, and device thinning to sub-20 µm. In this paper, we will summarize their capabilities and introduce how adhesion between the two temporary layers can be adjusted by material design. We will also cover a new feature of the thermoset layer that can be patterned to allow for the use of a patterned bonding material for TB/DB-type applications.

ShieLD: Shielding Cross-Zone Communication Within Limited-Resourced IoT Devices Running Vulnerable Software Stack

Securing IoT devices is gaining attention as the security risks associated with these devices increase rapidly. TrustZone-M, a Trusted Execution Environment (TEE) for Cortex-M processors, ensures stronger security within an IoT device by allowing isolated execution of security-critical operations, without trusting the entire software stack. However, TrustZone-M does not guarantee secure cross-world communication between applications in the Normal and Secure worlds. The cryptographic protection of the communication channel is an obvious solution; however, within a low-power IoT device, it incurs high overhead if applied to each cross-world message exchange. We present ShieLD, a framework that enables a secure communication channel between the two TrustZone-M worlds by leveraging the Memory Protection Unit (MPU). ShieLD guarantees confidentiality, integrity and authentication services without requiring any cryptographic operations. We implement and evaluate ShieLD using a Musca-A test chip board with Cortex-M33 that supports TrustZone-M. Our empirical evaluation shows, among other gains, the cross-zone communication protected with ShieLD is <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">5 times</i> faster than the conventional crypto-based communication.

(Invited) Automated Production of High Performance PEMFC Stacks and Components According to Automotive Requirements

EKPO Fuel Cell GmbH was founded in 2021 by two parent companies, ElringKlinger and Plastic Omnium, whom both are strong and established automotive suppliers with mass manufacturing capabilities and strategic approaches to hydrogen and the fuel cell businesses. The aim of EKPO is the development and production of proton exchange membrane fuel cell stack components and stack modules. Drawing from the experiences of the pilot automated production line since 2016, a fully optimized automated production, with improved production technologies and products, was developed by EKPO in 2021 to achieve an output capability of 10,000 stacks per year. The line was designed with a modular and scalable approach that allows flexibility as well as increased capacity by parallelization. The implementation is successful of producing EKPO’s various stack platforms while maintaining a high level of quality and production standard for the automotive industry.Since EKPO offers a wide range of products, the automated line can cover stack power ranges from 10 kW up to > 200 kW. As a prerequisite for excellent technical performance, in addition to function quality measures were set as key targets from the start of the development. EKPO has thus established consistent and stringent quality process flows and processes for all production steps. These measures include the early involvement of measurability and manufacturability assessment on all process and analysis steps on the different levels (i.e., components and stack). The feedback from each workflow can then continuously improve products, processes, and documentation regarding product specific codes standards and regulations (CSR).The accumulative knowledge from two decades of activity of fuel cell component and stack development enabled EKPO to establish a production range starting from cell and stack components (i.e., bipolar plates, sealings, and plastic end plate modules) to fuel cell stacks and modules produced by fully automated MEA and stack assembly and end-of-line testing. The automated production covers all processes from bipolar plate manufacturing and assembly, CCM and GDL roll good cutting, MEA-assembly, stacking, and the final compression and characterization of the fuel cell stack. The quality matrixes include inspecting performance key figures as well as quality and geometries of the stack and its components. The incoming materials are typically supplied as roll goods, (i.e., GDL, CCM, and metal coil) or as raw material (i.e., plastic granules). As the incoming materials and finished components are very sensitive to environmental conditions and particle contamination, the entire MEA processing and stack assembly process takes place in a cleanroom with controlled climate for maximum process stability and reliability.With a key approach of “design to manufacture” and deep understanding of stack technology and component requirements, EKPO is able to deliver two best in class stacks, namely the NM5 evo and NM12 single and twin platforms.

Modeling of a Hybrid Fuel Cell Powertrain with Power Split Logic for Onboard Energy Management Using a Longitudinal Dynamics Simulation Tool

This work aims to develop a mathematical model for the simulation of a fuel cell (FC) hybrid powertrain. The work starts from modeling a single cell to obtain information on the entire FC stack. The model obtained was integrated into a simulation tool presented in the literature that simulates the longitudinal dynamics of auxiliary power unit hybrid electric vehicles and fully electric vehicles. Therefore, the integrated model allows the simulation of hybrid vehicles equipped with FC and a battery pack that acts as a peak power source. The tool simulates the mechanical and electrical behavior of the vehicle, introducing an investigation of the power flows relating to the FC and batteries. An appropriate power split logic has been implemented, allowing the correct management of the power distribution between the FC and the batteries. The importance of analyzing FC vehicles’ behavior arises from the recent necessity to find alternative propulsion systems, overcoming the range problems associated with fully electric vehicles. The innovation lies in the versatility and modularity of the model, which is open to modifications and features a low computational burden, making it suitable for testing new solutions by performing first design and sizing calculations.

Application-Oriented Analysis of Ageing Processes Using the Example of Li-C Half Cells

Mechanical modifications or harmful side reactions are undesirable ageing effects that can occur during cycling. These phenomena have a negative impact on cell performance and consequently limit cycling stability. The focus here is on investigations of ageing processes such as volume changes and electrolyte decomposition during cycling.Thickness measurements as well as gas analytical studies of Li-C half cells with a carbonate-based electrolyte are presented. Graphite and Hard Carbon, respectively, were used as carbon materials. The electrolyte applied consists of 1M LiPF6 in EC:DMC (1:1, wt). All investigations were carried out using specially developed multifunctional test setups and accordingly modified test cells.Application-oriented analysis such as gas analysis, thickness measurements or both combined are suitable methods to define operating parameter, to benchmark and to identify decomposition reactions.In-operando electrochemical dilatometry was conducted to check whether an irreversible increase in thickness occurred. On the corresponding cell configurations, the dilatation was recorded over the entire electrode stack as well as of the separate working electrode only.Furthermore, in-operando MS could be used to detect gaseous substances produced by (electro)chemical processes as a function of the state of charge. With the GC-MS, a post-mortem analysis could be performed to identify the individual substances qualitatively.Results of the different half cells are shown and can be compared with each other. A detailed relationship can be demonstrated between the change in thickness and the potential curve. Moreover, any emerging Li-plating is clearly detectable. Among others, carbon dioxide, ethane as well as ethene were identified as degradation products of the carbonate-based electrolyte and are formed during the charging process. In addition, possible correlations between gas evolution and thickness changes can be shown.These analytical studies make an important contribution to get a more detailed insight into ageing processes that take place. The possible combination of in-operando dilatometry with in-operando mass spectrometry opens up new interesting perspectives to analyse cell performance. Finally, the shown results lead to a better understanding and help to develop appropriate countermeasures in order to reduce the negative effects and thus to ensure a higher cycling stability.This work is supported by the Fraunhofer and Max Planck cooperation program (Germany) in the project “ClusterBatt” with Fraunhofer Institute for Material and Beam Technology IWS, Dresden and Max Planck Institute of Colloids and Interfaces MPIKG, Potsdam. Figure 1

How Optical and Electrical Properties of ITO Coated Willow Glass Affect Photonic Curing Outcome for Upscaling Perovskite Solar Cell Manufacturing

Indium tin oxide (ITO) coated Willow glass is an excellent substrate for roll-to-roll manufacturing of perovskite solar cells (PSCs) but can have large variability in its optical and electrical properties. Photonic curing uses intense light pulses instead of heat to process materials and can facilitate faster processing speeds in roll-to-roll manufacturing to upscale the production of PSCs. Since the entire film stack can absorb light and contributes to the photonic curing outcome, the substrate materials’ properties play an integral role. Here, we present the effect of ITO transmittance on the photonic curing of nickel nitrate sol-gel precursors into nickel oxide and consequently the performance of PSCs fabricated with only photonic curing and no thermal annealing. Unexpectedly, ITO samples processed by photonic curing show improved optical and electrical properties.

A Heuristic-Based Task Scheduling Policy for QoS Improvement in Cloud

Cloud computing is a big step in the parallel and distributed computing that offers pervasive access to the entire stack of computing resources located in the data center via the Internet in a virtualized manner. QoS is an important research direction in the cloud and is a collection of constraints that meets the service level agreement (SLA) between the users and service providers. These constraints are waiting time, completion time, response time, makespan, resource utilization, effective utilization of bandwidth, and load balancing. This work presents a distribution plan for transferring task loads to different virtual machines (VMs) with an efficient load balancing mechanism that is best suited in heterogeneous environments and improve the QoS parameter of users and cloud vendors simultaneously. In this paper, the expected uniform load of tasks that can be mapped to a particular VMs is calculated and after that, the optimal average completion time (OACT) of expected uniform tasks load to each VM is calculated.

Subwavelength Bayer RGB color routers with perfect optical efficiency

Abstract We introduce subwavelength color routers with perfect optical efficiency in a red-green-green-blue (RGGB) Bayer layout for solid state image sensors. This is the first demonstration of a subwavelength device concept that shows the full potential of color routing, i.e., perfect routing without loss of photons, with a broadband, polarization-independent, and angular robust response. As an example, we show a color router for 320 nm wide image sensor pixels, which are two times smaller than the smallest state-of-the-art pixels, that features perfect optical efficiency, i.e., no crosstalk between color channels, no reflections, and no leakage into non-photodetector regions, even though the pixel photodetectors are 2–3 times smaller than the wavelength of the incident light. Our color router outperforms all other color separation approaches and can replace the entire optical stack in solid state image sensors.

Potential of Convolutional Neural Networks for Forest Mapping Using Sentinel-1 Interferometric Short Time Series

The monitoring of land cover and land use patterns is pivotal in the joint effort to fight deforestation in the Amazon and study its relation to climate change effects with respect to anthropogenic activities. Most of the region, typically monitored with optical sensors, is hidden by a persistent cloud cover for most of the wet season. The necessity for a consistent and reliable deforestation warning system based on cloud-independent radar data is therefore of particular interest. In this paper, we investigated the potential of combining deep learning with Sentinel-1 (S-1) Interferometric Synthetic Aperture Radar (InSAR) short time series (STS), covering only 24 d of acquisitions, to map endangered areas in the Amazon Basin. To this end, we implemented a U-Net-like convolutional neural network (CNN) for multi-layer semantic segmentation, trained from scratch with different sets of input features to evaluate the viability of the proposed approach for different operating conditions. As input features, we relied on both multi-temporal backscatter and interferometric coherences at different temporal baselines. We provide a detailed performance benchmark of the different configurations, also considering the current state-of-the-art approaches based on S-1 STS and shallow learners. Our findings showed that, by exploiting the powerful learning capabilities of CNNs, we outperformed the STS-based approaches published in the literature and significantly reduced the computational load. Indeed, when considering the entire stack of Sentinel-1 data at a 6 d revisit time, we were able to maintain the overall accuracy and F1-score well above 90% and reduce the computational time by more than 50% with respect to state-of-the-art approaches, by avoiding the generation of handcrafted feature maps. Moreover, we achieved satisfactory results even when only S-1 InSAR acquisitions with a revisit time of 12 d or more were used, setting the groundwork for an effective and fast monitoring of tropical forests at a global scale.

Human-Scale Mobile Manipulation Using RoMan

We present the design, integration, and evaluation of a full-stack robotic system called RoMan, which can conduct autonomous field operations involving physical interaction with its environment. RoMan offers autonomous behaviors that can be triggered from succinct, high-level human input such as “open this box and retrieve the bag inside.” The robot’s behaviors are driven by a set of planners and controllers grounded in perceptual reconstructions of the environment. These behaviors are articulated by a behavior tree that translates high-level operator input into programs of increasing sensorimotor expressiveness, ultimately driving the lowest-level controllers. The software system is implemented in ROS as a set of independent processes connected by synchronous and asynchronous communication, and distributed across two on-board planning/control computers. The behavior stack drives a novel platform consisting of a pair of custom, 500 Nm/axis manipulators mounted on a rotatable torso aboard a tracked platform. The robot’s head is equipped with forward-looking depth cameras, and the arms carry wrist-mounted force-torque sensors and a mix of three- and four-finger grippers. We discuss design and implementation trade-offs affecting the entire hardware-software stack and high-level manipulation behaviors. We also demonstrate the applicability of the system for solving two manipulation tasks: 1) removing heavy debris from a roadway, where 64% of end-to-end autonomous runs required at most one human intervention; and 2) retrieving an item from a closed container, with a fully autonomous success rate of 56%. Finally, we indicate lessons learned and suggest outstanding research problems.

Autonomous Vehicles and Machines 2022 Conference Overview and Papers Program

Advancements in sensing, computing, image processing, and computer vision technologies are enabling unprecedented growth and interest in autonomous vehicles and intelligent machines, from self-driving cars to unmanned drones, to personal service robots. These new capabilities have the potential to fundamentally change the way people live, work, commute, and connect with each other, and will undoubtedly provoke entirely new applications and commercial opportunities for generations to come. The main focus of AVM is perception. This begins with sensing. While imaging continues to be an essential emphasis in all EI conferences, AVM also embraces other sensing modalities important to autonomous navigation, including radar, LiDAR, and time of flight. Realization of autonomous systems also includes purpose-built processors, e.g., ISPs, vision processors, DNN accelerators, as well core image processing and computer vision algorithms, system design and architecture, simulation, and image/video quality. AVM topics are at the intersection of these multi-disciplinary areas. AVM is the Perception Conference that bridges the imaging and vision communities, connecting the dots for the entire software and hardware stack for perception, helping people design globally optimized algorithms, processors, and systems for intelligent &#x201C;eyes&#x201D; for vehicles and machines.

AUTOMATIC EFFUSION MEASUREMENT WITH LIMITED LABELED DATA

Accurate quantification of effusion volume is becoming more important as OA is increasingly recognized to have inflammatory components including joint effusion. Effective clinical management requires accurate quantification of features related to inflammation; particularly synovitis and joint effusion which represents a target for therapy. Rather than relying on highly variable subjective assessment of joint effusions, with the development of deep learning techniques we have witnessed increased automation of radiological assessments. However, most of these techniques use supervised learning and are trained on large annotated datasets. Acquiring “ground truth” labels for large datasets is expensive and time-consuming. This becomes even more complicated and challenging when it comes to detecting medical image features like effusion. Detecting effusions accurately requires a high level of expertise. Because pockets of joint fluid can be in unexpected locations, human experts frequently need to step back to look at the entire image stack multiple times to prevent inattentional error. Therefore, proposing a method for learning from limited labeled data is crucial for medical image analysis. To perform automated measurement of knee effusion volume using minimal labeled data. In this method, we aim to develop an alternative deep learning training strategy based on self-supervised pretraining on unlabeled knee MRI scans. This study requires two sets of data: 1) a large unlabeled dataset, and 2) a small labeled dataset. For the large unlabeled dataset, we have used 4 different OAI MRI sequence data (64k slices have been selected randomly). For the labeled dataset we have used low-resolution sagittal Turbo Spin Echo (TSE) MR sequences from the OAI dataset. Effusion regions were extracted by a trained musculoskeletal radiologist (DC) for 23 participants (a total of 31 scans) using an interactive software developed in-house. We have used 5 scans for the test and 26 scans (a total of 700 slices) for training. In this method, IMaskRCNN deep learning architecture with the ResNet-101 backbone have been used. The training process has been performed in two-phase, self-supervised pretraining, and fine-tuning. The self-supervised pretraining using the fill-blank method was performed on unlabeled images to learn visual representations in musculoskeletal MRIs. After pretraining, we fined-tuned last layers of the network (backbone weights have been frozen) using the training division of the small TSE dataset. The dice similarity coefficient, Intersection over union (IoU) was used to compare the result of the segmentation between the predicted output and the ground truth. Qualitative results of the proposed method is presented in the figure. As shown the network was able to provide a good prediction of effusion location and extracting it. The segmentation model using the proposed pretraining method was successful in finding the location of effusion for 90% of the slices with IoU = 60%. The dice score for the detected effusion is 0.70 indicates a high agreement between the automatic segmentation and the labels provided by the expert. Self-supervised pretraining on the OAI dataset, can be a solution to the scarcity of pixel-wise ground truth (GT) for effusion quantification. Alberta Innovates, AHS Chair in Diagnostic Imaging, Medical Imaging Consultants, CIHR. CORRESPONDENCE ADDRESS: banafshe.felfeliyan@ucalgary.ca

ALPINE: Analog In-Memory Acceleration with Tight Processor Integration for Deep Learning

Analog in-memory computing (AIMC) cores offers significant performance and energy benefits for neural network inference with respect to digital logic (e.g., CPUs). AIMCs accelerate matrix-vector multiplications, which dominate these applications' run-time. However, AIMC-centric platforms lack the flexibility of general-purpose systems, as they often have hard-coded data flows and can only support a limited set of processing functions. With the goal of bridging this gap in flexibility, we present a novel system architecture that tightly integrates analog in-memory computing accelerators into multi-core CPUs in general-purpose systems. We developed a powerful gem5-based full system-level simulation framework into the gem5-X simulator, ALPINE, which enables an in-depth characterization of the proposed architecture. ALPINE allows the simulation of the entire computer architecture stack from major hardware components to their interactions with the Linux OS. Within ALPINE, we have defined a custom ISA extension and a software library to facilitate the deployment of inference models. We showcase and analyze a variety of mappings of different neural network types, and demonstrate up to 20.5x/20.8x performance/energy gains with respect to a SIMD-enabled ARM CPU implementation for convolutional neural networks, multi-layer perceptrons, and recurrent neural networks.

Probabilistically Guaranteeing End-to-end Latencies in Autonomous Vehicle Computing Systems

Good responsiveness of autonomous vehicle computing systems is crucial to safety and performance of the vehicles. For example, an autonomous vehicle (AV) may cause an accident if the end-to-end latency from sensing a pedestrian to emergency stop is too high. However, the AV software stacks are too complex to probabilistically analye the end-to-end latency on a multi-core system. They consist of a graph of tasks with different periods, and have a large variability in the task execution times, which may lead to the maximum core utilization Umax >= 1.0 on some cores. This paper proposes a novel stochastic analysis of the end-to-end latency over the AV stacks that allows Umax to exceed 1.0 on each core. The proposed analysis models the entire stack as a graph of task graphs under a multi-core partitioned scheduling and provides a probabilistic guarantee that the analyzed latency distribution upper-bounds the one observed from a real system under the assumption of independent task execution times. Using the Autoware stack with inter-task dependent execution times, it is shown that our analysis, combined with a task grouping to mitigate the inter-task correlations, can give a latency distribution for each task path that almost upper-bounds the observed one.

Defect localization during laser microwelding of battery connectors using long exposure imaging and few-shot learning

In the production of battery systems for electromobility, even a single weak electrical connection of the battery cells can lead to a failure of the entire stack. During laser welding of dissimilar materials such as aluminum and copper, real-time information about the actual part quality is essential for cost-efficient production and high product quality. To address these needs, long exposure optical imaging is used to generate detailed images during the process based on local near-infrared process emission. The high-resolution images are subsequently processed by using a novel data analysis approach. Existing image classification methods based on supervised learning usually require thousands of labeled images for training. To overcome these shortcomings, a few-shot learning approach based on Siamese Neural Networks is investigated, which require only a few samples to detect critical welding defects. The approach is evaluated on industry-relevant experiments and demonstrates promising results in terms of high defect recognition accuracies.

Automated detection and reacquisition of motion-degraded images in fetal HASTE imaging at 3 T.

Fetal brain Magnetic Resonance Imaging suffers from unpredictable and unconstrained fetal motion that causes severe image artifacts even with half-Fourier single-shot fast spin echo (HASTE) readouts. This work presents the implementation of a closed-loop pipeline that automatically detects and reacquires HASTE images that were degraded by fetal motion without any human interaction. A convolutional neural network that performs automatic image quality assessment (IQA) was run on an external GPU-equipped computer that was connected to the internal network of the MRI scanner. The modified HASTE pulse sequence sent each image to the external computer, where the IQA convolutional neural network evaluated it, and then the IQA score was sent back to the sequence. At the end of the HASTE stack, the IQA scores from all the slices were sorted, and only slices with the lowest scores (corresponding to the slices with worst image quality) were reacquired. The closed-loop HASTE acquisition framework was tested on 10 pregnant mothers, for a total of 73 acquisitions of our modified HASTE sequence. The IQA convolutional neural network, which was successfully employed by our modified sequence in real time, achieved an accuracy of 85.2% and area under the receiver operator characteristic of 0.899. The proposed acquisition/reconstruction pipeline was shown to successfully identify and automatically reacquire only the motion degraded fetal brain HASTE slices in the prescribed stack. This minimizes the overall time spent on HASTE acquisitions by avoiding the need to repeat the entire stack if only few slices in the stack are motion-degraded.