U.S. Air Force Research Laboratory Information Directorate

diffusive memristor, which exhibits evident advantages in scalability, circuit complexity, and power consumption. The random bits generated by the diffusive memristor true random number generator pass all 15 NIST randomness tests without any post-processing, a first for memristive-switching true random number generators. Based on nanoparticle dynamic simulation and analytical estimates, we attribute the stochasticity in delay time to the probabilistic process by which Ag particles detach from a Ag reservoir. This work paves the way for memristors in hardware security applications for the era of the Internet of Things.Memristors can switch between high and low electrical-resistance states, but the switching behaviour can be unpredictable. Here, the authors harness this unpredictability to develop a memristor-based true random number generator that uses the stochastic delay time of threshold switching.

By mimicking the highly parallel biological systems, neuromorphic hardware provides the capability of information processing within a compact and energy-efficient platform. However, traditional Von Neumann architecture and the limited signal connections have severely constrained the scalability and performance of such hardware implementations. Recently, many research efforts have been investigated in utilizing the latest discovered memristors in neuromorphic systems due to the similarity of memristors to biological synapses. In this paper, we explore the potential of a memristor crossbar array that functions as an autoassociative memory and apply it to brain-state-in-a-box (BSB) neural networks. Especially, the recall and training functions of a multianswer character recognition process based on the BSB model are studied. The robustness of the BSB circuit is analyzed and evaluated based on extensive Monte Carlo simulations, considering input defects, process variations, and electrical fluctuations. The results show that the hardware-based training scheme proposed in the paper can alleviate and even cancel out the majority of the noise issue.

A novel Ag/oxide-based threshold switching device with attractive features including ≈1010 nonlinearity is developed. High-resolution transmission electron microscopic analysis of the nanoscale crosspoint device suggests that elongation of an Ag nanoparticle under voltage bias followed by spontaneous reformation of a more spherical shape after power off is responsible for the observed threshold switching. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Experimental demonstration of resistive neural networks has been the recent focus of hardware implementation of neuromorphic computing. Capacitive neural networks, which call for novel building blocks, provide an alternative physical embodiment of neural networks featuring a lower static power and a better emulation of neural functionalities. Here, we develop neuro-transistors by integrating dynamic pseudo-memcapacitors as the gates of transistors to produce electronic analogs of the soma and axon of a neuron, with "leaky integrate-and-fire" dynamics augmented by a signal gain on the output. Paired with non-volatile pseudo-memcapacitive synapses, a Hebbian-like learning mechanism is implemented in a capacitive switching network, leading to the observed associative learning. A prototypical fully integrated capacitive neural network is built and used to classify inputs of signals.

Recent research has shown that network coding can be used in content distribution systems to improve the speed of downloads and the robustness of the systems. However, such systems are very vulnerable to attacks by malicious nodes, and we need to have a signature scheme that allows nodes to check the validity of a packet without decoding. In this paper, we propose such a signature scheme for network coding. Our scheme makes use of the linearity property of the packets in a coded system, and allows nodes to check the integrity of the packets received easily. We show that the proposed scheme is secure, and its overhead is negligible for large files.

In-memory computing represents an effective method for modeling complex physical systems that are typically challenging for conventional computing architectures but has been hindered by issues such as reading noise and writing variability that restrict scalability, accuracy, and precision in high-performance computations. We propose and demonstrate a circuit architecture and programming protocol that converts the analog computing result to digital at the last step and enables low-precision analog devices to perform high-precision computing. We use a weighted sum of multiple devices to represent one number, in which subsequently programmed devices are used to compensate for preceding programming errors. With a memristor system-on-chip, we experimentally demonstrate high-precision solutions for multiple scientific computing tasks while maintaining a substantial power efficiency advantage over conventional digital approaches.

Current electromagnetic-field (EMF) exposure limits have been based, in part, on the amount of energy absorbed by the whole body. However, it is known that energy is absorbed nonuniformly during EMF exposure. The development and widespread use of sophisticated three-dimensional anatomical models to calculate specific-absorption-rate (SAR) values in biological material has resulted in the need to understand how model parameters affect predicted SAR values. This paper demonstrate the effects of manipulating frequency, permittivity values, and voxel size on SAR values calculated by a finite-difference time-domain program in digital homogenous sphere models and heterogeneous models of rat and man. The predicted SAR values are compared to empirical data from infrared thermography and implanted temperature probes.

Information fusion utilizes a collection of data sources for uncertainty reduction, coverage extension, and situation awareness. Future information fusion solutions require systems design [1], coordination with users [2], metrics of performance [3], and methods of multilevel security [4]. A current trend that can enable all of these services is cloud computing. Cloud computing as defined by NIST is: Cloud computing is a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction. [5] Cloud computing provides capabilities (on-demand self service, broad network access, resource pooling, rapid elasticity, and measured service) over different types of clouds (private, community, public, and hybrid).

Maintaining data provenance in cloud in a tamper-resistant manner that cannot be breached by malicious parties is a necessity from the current security standpoint. Blockchain technology has emerged as a secure solution to store and share information by offering an immutable distributed ledger service. Its effectiveness hinges on the infrastructure supporting the distributed ledger and consensus protocol that governs the validity of entries in the Blockchain. Hence, Blockchain can be a potential candidate to implement data provenance; however, traditional cryptocurrency-based consensus models become a bottleneck in the cloud environment. Therefore, in this paper, we propose a Blockchain based data provenance architecture (BlockCloud) that incorporates a proof-of-stake (PoS)-based consensus protocol (CloudPoS) for securely recording the data operations occurring in cloud environment. The critical operational phases of the protocol are discussed in depth, which leverages the cloud users' cyber infrastructure resources. A cloud-based testbed environment is created using a local cluster of physical machines managed by Xen hypervisor. Resource elasticity is enabled using Kubernetes setup that interacts with the dockerized containers, which emulate as peers in the Blockchain network. We then evaluate the effectiveness of the protocol in a simulated environment and conduct performance tests of the proposed consensus.

Artificial intelligence (AI), owing to recent breakthroughs in deep learning, has revolutionized applications and services in almost all technology domains including aerospace. AI and deep learning rely on huge amounts of training data that are mostly generated at the network edge by Internet of Things (IoT) devices and sensors. Bringing the sensed data from the edge of a distributed network to a centralized cloud is often infeasible because of the massive data volume, limited network bandwidth, and real-time application constraints. Consequently, there is a desire to push AI frontiers to the network edge toward utilizing the enormous amount of data generated by IoT devices near the data source. The merger of edge computing and AI has engendered a new discipline, that is, AI at the edge or edge intelligence. To help AI make sense of gigantic data at the network edge, data fusion is of paramount significance and goes hand in hand with AI. This article focuses on data fusion and AI at the edge. In this article, we propose a framework for data fusion and AI processing at the edge. We then provide a comparative discussion of different data fusion and AI models and architectures. We discuss multiple levels of fusion and different types of AI, and how different types of AI align with different levels of fusion. We then highlight the benefits of combining data fusion with AI at the edge. The methods of AI and data fusion at the edge detailed in this article are applicable to many application domains including aerospace systems. We evaluate the effectiveness of combined data fusion and AI at the edge using convolutional neural network models and multiple hardware platforms suitable for edge computing. Experimental results reveal that combining AI with data fusion can impart a speedup of 9.8× while reducing energy consumption up to 88.5% over AI without data fusion. Furthermore, results demonstrate that data fusion either maintains or improves the accuracy of AI in most cases. For our experiments, data fusion imparts a maximum improvement of 15.8% in accuracy to AI.

1S1R (1 selector and 1 memristor) is a laterally scalable and vertically stackable scheme that can lead to the ultimate memristor density for either memory or neural network applications. In such a scheme, the memristor device needs to be truly electroforming‐free and operated at both low currents and low voltages in order to be compatible with a two‐terminal selector. In this work, a new type of memristor with a preconditioned tunneling conductive path is developed to achieve the required performance characteristics, including truly electroforming‐free, low current below 30 µA (potentially <1 µA), and simultaneously low voltage ≈±0.7 V in switching operations. Such memristors are further integrated with two types of recently developed selectors to demonstrate the feasibility of 1S1R integration.

Obtaining measured synthetic aperture radar (SAR) data for training automatic target recognition (ATR) models can be too expensive (in terms of time and money) and complex of a process in many situations. In response, researchers have developed methods for creating synthetic SAR data for targets using electro-magnetic prediction software, which is then used to enrich an existing measured training dataset. However, this approach relies on the availability of some amount of measured data. In this work, we focus on the case of having 100% synthetic training data, while testing on only measured data. We use the SAMPLE dataset public released by AFRL, and find significant challenges to learning generalizable representations from the synthetic data due to distributional differences between the two modalities and extremely limited training sample quantities. Using deep learning-based ATR models, we propose data augmentation, model construction, loss function choices, and ensembling techniques to enhance the representation learned from the synthetic data, and ultimately achieved over 95% accuracy on the SAMPLE dataset. We then analyze the functionality of our ATR models using saliency and feature-space investigations and find them to learn a more cohesive representation of the measured and synthetic data. Finally, we evaluate the out-of-library detection performance of our synthetic-only models and find that they are nearly 10% more effective than baseline methods at identifying measured test samples that do not belong to the training class set. Overall, our techniques and their compositions significantly enhance the feasibility of using ATR models trained exclusively on synthetic data.

Artificial intelligence (AI) hardware is positioned to unlock revolutionary computational abilities by leveraging vast distributed networks of advanced semiconductor chips. However, a barrier for AI scaling is the disproportionately high energy and chip area required to transmit data between the chips. Here we present a solution to this long-standing overhead through dense three-dimensional (3D) integration of photonics and electronics. With 80 photonic transmitters and receivers occupying a combined chip footprint of only 0.3 mm2, our platform achieves an order-of-magnitude-greater number of 3D-integrated channels than prior demonstrations. This enables both high-bandwidth (800 Gb s−1) and highly efficient, dense (5.3 Tb s−1 mm−2) 3D channels. The transceiver energy efficiency is showcased by a state-of-the-art 50 fJ and 70 fJ per communicated bit from the transmitter and receiver front ends, respectively, operating at 10 Gb s−1per channel. Furthermore, the design is compatible with commercial complementary metal–oxide–semiconductor foundries fabrication on 300-mm-sized wafers, providing a route to mass production. Such ultra-energy-efficient, high-bandwidth data communication links promise to eliminate the bandwidth bottleneck between spatially distinct compute nodes and support the scaling of future AI computing hardware. Dense three-dimensional integration of photonics and electronics results in a high-speed (800 Gb s−1) data interface for semiconductor chips that features 80 communication channels and consumes only tens of femtojoules per transmitted bit.

Silicon photonics holds significant promise in revolutionizing optical interconnects in data centers and high performance computers to enable scaling into the Pb/s package escape bandwidth regime while consuming orders of magnitude less energy per bit than current solutions. In this work, we review recent progress in silicon photonic interconnects leveraging chip-scale Kerr frequency comb sources and provide a comprehensive overview of massively scalable silicon photonic systems capable of capitalizing on the large number of wavelengths provided by such combs. We first consider the high-level architectural constraints and then proceed to detail the corresponding fundamental device designs supported by both simulated and experimental results. Furthermore, the majority of experimentally measured devices were fabricated in a commercial 300 mm foundry, showing a clear path to volume manufacturing. Finally, we present various system-level experiments which illustrate successful proof-of-principle operation, including flip-chip integration with a co-designed CMOS application-specific integrated circuit (ASIC) to realize a complete Kerr comb-driven electronic-photonic engine. These results provide a viable and appealing path towards future co-packaged silicon photonic interconnects with aggregate per-fiber bandwidth above 1 Tb/s, energy consumption below 1 pJ/bit, and areal bandwidth density greater than 5 Tb/s/mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> .

In recent years, using a network of autonomous and cooperative unmanned aerial vehicles (UAVs) without command and communication from the ground station has become more imperative, particularly in search-and-rescue operations, disaster management, and other applications where human intervention is limited. In such scenarios, UAVs can make more efficient decisions if they acquire more information about the mobility, sensing and actuation capabilities of their neighbor nodes. In this study, we develop an unsupervised online learning algorithm for joint mobility prediction and object profiling of UAVs, to facilitate control and communication protocols. The proposed method not only predicts the future locations of the surrounding flying objects, but also classifies them into different groups with similar levels of maneuverability (e.g., rotatory and fixed-wing UAVs) without prior knowledge regarding these classes. This method is flexible in admitting new object types with unknown mobility profiles, and is thereby applicable to emerging flying ad-hoc networks (FANETs) with heterogeneous nodes.

In this paper we propose a distributed congestion control algorithm for tree based communications in wireless sensor networks, that seeks to adaptively assign a fair and efficient transmission rate to each node. In our algorithm, each node monitors its aggregate output and input traffic rates. Based on the difference of the two, a node then decides either to increase or decrease the bandwidth allocable to a flow originating from itself and to those being routed through it. Since the application requirements in sensor network follows no common trait, our design abstracts the notion of fairness, allowing for the development of a generic utility controlling module. Such separation of the utility and fairness controlling modules enables each one to use a separate control law, thereby portraying a more flexible design. The working of our congestion control is independent of the underlying routing algorithm and is designed to adapt to changes in the underlying routing topology. We evaluate the performance of the algorithm via extensive simulations using an event-driven packet level simulator. The results suggest that the proposed protocol acquires a significantly high goodput of around 95% of the actual transmission rate, converges quickly to the optimal rate, and attains the desired fairness.

This work presents ChainFS, a middleware system that secures cloud storage services using a minimally trusted Blockchain. ChainFS hardens the cloud-storage security against forking attacks. The ChainFS middleware exposes a file-system interface to end users. Internally, ChainFS stores data files in the cloud and exports minimal and necessary functionalities to the Blockchain for key distribution and file operation logging. We implement the ChainFS system on Ethereum and S3FS and closely integrate it with FUSE clients and Amazon S3 cloud storage. We measure the system performance and demonstrate low overhead.

Cybersecurity is among the highest priorities in industries, academia and governments. Cyber-threats information sharing among different organizations has the potential to maximize vulnerabilities discovery at a minimum cost. Cyber-threats information sharing has several advantages. First, it diminishes the chance that an attacker exploits the same vulnerability to launch multiple attacks in different organizations. Second, it reduces the likelihood an attacker can compromise an organization and collect data that will help him launch an attack on other organizations. Cyberspace has numerous interconnections and critical infrastructure owners are dependent on each other's service. This well-known problem of cyber interdependency is aggravated in a public cloud computing platform. The collaborative effort of organizations in developing a countermeasure for a cyber-breach reduces each firm's cost of investment in cyber defense. Despite its multiple advantages, there are costs and risks associated with cyber-threats information sharing. When a firm shares its vulnerabilities with others there is a risk that these vulnerabilities are leaked to the public (or to attackers) resulting in loss of reputation, market share and revenue. Therefore, in this strategic environment the firms committed to share cyber-threats information might not truthfully share information due to their own self-interests. Moreover, some firms acting selfishly may rationally limit their cybersecurity investment and rely on information shared by others to protect themselves. This can result in under investment in cybersecurity if all participants adopt the same strategy. This paper will use game theory to investigate when multiple self-interested firms can invest in vulnerability discovery and share their cyber-threat information. We will apply our algorithm to a public cloud computing platform as one of the fastest growing segments of the cyberspace.

The value memristor devices offer to the neuromorphic computing hardware design community rests on the ability to provide effective device models that can enable large scale integrated computing architecture application simulations. Therefore, it is imperative to develop practical, functional device models of minimum mathematical complexity for fast, reliable, and accurate computing architecture technology design and simulation. To this end, various device models have been proposed in the literature seeking to characterize the physical electronic and time domain behavioral properties of memristor devices. In this work, we analyze some promising and practical non-quasi-static linear and non-linear memristor device models for neuromorphic circuit design and computing architecture simulation.

Training deep learning-based synthetic aperture radar automatic target recognition (SAR-ATR) systems for use in an “open-world” operating environment has, thus far proven difficult. Most SAR-ATR systems are designed to achieve maximum accuracy for a limited set of classes, yet ignore the implications of encountering novel target classes during deployment. Even worse, the standard deep learning training objectives fundamentally inherit a closed-world assumption, and provide no guidance for how to handle out-of-distribution (OOD) data. In this work, we develop a novel training procedure called adversarial outlier exposure (AdvOE) to codesign the ATR system for accuracy and OOD detection. Our method introduces a large, diverse, and unlabeled auxiliary training dataset containing samples from the OOD set. The AdvOE objective encourages a deep neural network to learn robust features of the in-distribution training data, while also promoting maximum entropy predictions for adversarially perturbed versions of the OOD data. We experiment with the recent SAMPLE dataset, and find our method nearly doubles the OOD detection performance over the baseline in key settings, and excels when using only synthetic training data. As compared to several other advanced ATR training techniques, AdvOE also affords significant improvements in both classification and detection statistics. Finally, we conduct extensive experiments that measure the effect of OOD set granularity on detection rates; discuss the implications of using different detection algorithms; and develop a novel analysis technique to validate our findings and interpret the OOD detection problem from a new perspective.