Search

Deep neural networks use skip connections to improve training convergence. However, these skip connections are costly in hardware, requiring extra buffers and increasing on- and off-chip memory utilization and bandwidth requirements. In this paper, we show that skip connections can be optimized for hardware when tackled with a hardware-software codesign approach. We argue that while a network’s skip connections are needed for the network to learn, they can later be removed or shortened to provide a more hardware efficient implementation with minimal to no accuracy loss. We introduce Tailor , a codesign tool whose hardware-aware training algorithm gradually removes or shortens a fully trained network’s skip connections to lower their hardware cost. Tailor improves resource utilization by up to 34% for BRAMs, 13% for FFs, and 16% for LUTs for on-chip, dataflow-style architectures. Tailor increases performance by 30% and reduces memory bandwidth by 45% for a 2D processing element array architecture.

As COVID-19 began to grip healthcare systems worldwide, worst-case models predicted huge demands for ventilators. The global community sprang to action, producing a large number of emergency "makeshift" ventilator designs. This brought about another problem: a gap between the quantity of new mechanical ventilators and the number of skilled physicians to operate them. New physicians could not complete training at the pace of ventilator production, which threatened to leave patients sitting untreated, next to unusable ventilators. To address this challenge, we developed a universal remote control system for makeshift ventilators that uses low-cost hardware add-on modules to connect to different ventilators, and a three-tier control architecture to interface the ventilators with telemedicine software. We demonstrate system integration with two representative ventilator designs, adding a remote control option that allows caregivers to quickly and easily monitor and control these ventilators remotely.

Neuromorphic vision sensors are an emerging technology inspired by how retina processing images. A neuromorphic vision sensor only reports when a pixel value changes rather than continuously outputting the value every frame as is done in an 'ordinary' Active Pixel Sensor (ASP). This move from a continuously sampled system to an asynchronous event driven one effectively allows for much faster sampling rates; it also fundamentally changes the sensor interface. In particular, these sensors are highly sensitive to noise, as any additional event reduces the bandwidth, and thus effectively lowers the sampling rate. In this work we introduce a novel spatiotemporal filter with O(N)O(N) memory complexity for reducing background activity noise in neuromorphic vision sensors. Our design consumes 10× less memory and has 100× reduction in error compared to previous designs. Our filter is also capable of recovering real events and can pass up to 180 percent more real events.

High-level synthesis (HLS) allows hardware designers to think algorithmically and not worry about low-level, cycle-by-cycle details. This provides the ability to quickly explore the architectural design space and tradeoffs between resource utilization and performance. Unfortunately, security evaluation is not a standard part of the HLS design flow. In this article, we aim to understand the effects of memory-based HLS optimizations on power side-channel leakage. We use Xilinx Vivado HLS to develop different cryptographic cores, implement them on a Spartan-6 FPGA, and collect power traces. We evaluate the designs with respect to resource utilization, performance, and information leakage through power consumption. We have two important observations and contributions. First, the choice of resource optimization directive results in different levels of side-channel vulnerabilities. Second, the partitioning optimization directive can greatly compromise the hardware cryptographic system through power side-channel leakage due to the deployment of memory control logic. We describe an evaluation procedure for power side-channel leakage and use it to make best-effort recommendations about how to design more secure architectures in the cryptographic domain.

Groups of coordinated underwater vehicles or sensors are powerful tools for monitoring the ocean. A requirement of many coordinated surveys is to determine a spatial reference between each node in a swarm. This work considers the self-localization of a swarm of independently moving vehicles using acoustic noise from a dominating incoherent source recorded by a single hydrophone onboard each vehicle. This method provides an inexpensive and infrastructure-free spatial reference between vehicles. Movement between the vehicles changes the swarm geometry and a self-localization estimate must be generated from data collected on short time scales. This challenges past self-localization approaches for acoustic arrays. To overcome this challenge, the proposed self-localization algorithm jointly estimates the vehicle geometry and the directionality of the ambient noise field, without prior knowledge of either estimate. To demonstrate this method, experimental results are provided when a boat is the main dominating source. The results demonstrate the ability to both estimate the direction of arrival of the boat and the relative positions of the vehicles in the swarm. The approach in this paper is not limited to moving vessels. Simulations are provided to examine three different factors that affect the proposed solution: inter-vehicle motion, vehicle geometry, and the azimuthal variance of the noise field.

Hardware Trojan detection is a very difficult challenge. However, the combination of symbolic execution and metamorphic testing is useful for detecting hardware Trojans in Verilog code. In this paper, symbolic execution and metamorphic testing were combined to detect internal conditionally triggered hardware Trojans in the register-transfer level design. First, control flow graphs of Verilog code were generated. Next, parallel symbolic execution and satisfiability modulo theories solver generated test patterns. Finally, metamorphic testing detected the hardware Trojans. The work used Trust-Hub benchmarks in experiments.

Cryptographic cores are known to leak information about their private key due to runtime variations, and there are many well-known attacks that can exploit this timing channel. In this paper, we study how information theoretic measures can quantify the amount of key leakage that can be exacted from runtime measurements. We develop and analyze 22 Rivest-Shamir-Adleman (RSA) hardware designs - each with unique performance optimizations, timing channel mitigation techniques, or discretization/randomization countermeasures. We demonstrate the effectiveness of information theoretic measures for quantifying timing leakage through correlation analysis of information theoretic measurements and attack results. Experimental results show that mutual information is a promising technique for quantifying timing leakage for RSA, advanced encryption standard, and elliptic curve cryptography ciphers, i.e., the mutual information correlates to being able to successfully guess the value of the private key. This is an important step toward a hardware security metric which allows designers to reason about security alongside traditional hardware design metrics like area, performance, and power.

While it has been recognized that the multipath structure of the underwater acoustic (UWA) channel offers the potential for compressed sensing (CS) sparsity exploitation, the rapidly time varying arrivals induced by highly dynamic surfaces unfortunately pose significant difficulties to channel estimation. From the viewpoint of underwater acoustic propagation, with the exception of the highly time varying arrivals caused by dynamic surface, generally there exist relatively stationary or slowly changing arrivals caused by direct path or bottom reflection, which imply the adoption of a discriminate estimation method to handle sparse components with different time variation scale. By modeling the time varying UWA channels as a sparse set consisting of constant and time-varying supports, in this paper, estimation of time varying UWA channel is transformed into a problem of dynamic compressed sensing sparse recovery. The combination of a Kalman filter and compressed sensing is adopted to pursue the solution of it. Numerical simulations demonstrate the superiority of the proposed algorithm. In the form of a channel-estimation-based decision-feedback equalizer, the experimental results with the field data obtained in a shallow water acoustic communication experiment indicate that the proposed dynamic compressed sensing algorithm outperforms classic algorithms as well as CS algorithms.

Distributed compressed sensing techniques are applied to enhance sparse channel estimation performance in underwater acoustic multiband systems. The core idea is to use receptions from multiple sub-bands to enhance the detection of channel tap positions. A known variant of the orthogonal matching pursuit (OMP) algorithm based on the distributed compressed sensing principle is simultaneous orthogonal matching pursuit (SOMP). However, the impulse responses across multiple sub-bands may have different arrival structures, although they often show a certain level of similarity. To address such differences at the sub-bands, a multiple selection strategy is applied to select multiple candidates at individual sub-bands at each iteration. This is different from the conventional OMP and SOMP algorithms that select only one candidate at each iteration. When the multiple selection strategy is combined with the SOMP algorithm, the proposed algorithm is referred to as JB-MSSOMP algorithm. To take advantage of channel coherence between adjacent data blocks from different sub-bands, the multiple selection strategy is further used over time. This leads to JBT-MSSOMP algorithm. Computer simulations show improved channel estimation performance of the proposed JB-MSSOMP and JBT-MSSOMP algorithms over the OMP or SOMP algorithms. Communication data from a recent acoustic experiment demonstrates improved receiver performance with the proposed channel estimators.

Imaging flow cytometry and high speed microscopy have shown immense promise for clinical diagnostics, biological research, and drug discovery. They enable high throughput screening and sorting using biological, chemical, or mechanical properties of cells. These techniques can separate mature cells from immature ones, determine the presence of cancerous cells, classify stem cells during differentiation, and screen drugs based upon how they affect cellular architecture. The process works by imaging cells at a high rate, extracting features of the cell (e.g., size, location, circularity, deformation), and using those features to classify the cell. Modern systems have a target throughput of thousands of cells per second, which requires imaging at rates of more than 60,000 frames per second. The cellular features must be calculated in less than a millisecond to enable real-time sorting. This creates challenging computing performance constraints in terms of both throughput and latency. In this paper, we present a hardware accelerated system for high throughput cellular image analysis. We carefully developed algorithms and their corresponding hardware implementations to meet the strict computational demands. Our algorithm analyzes and extracts cellular morphological features from low resolution microscopic images. Our hardware accelerated system operates at over 60,000 frames per second with 0.068 ms latency. This is almost 1400× faster in throughput than similar software based analysis and 335× better in terms of latency.

There is increasing interest in deploying swarms of underwater vehicles for marine surveys. One of the main challenges when designing these systems is coming up with an appropriate way to localize each vehicle in relation to one another. This paper considers the self-localization of a deforming swarm of subsurface floating vehicles using impulsive sources of opportunity, such as the sounds of snapping shrimp that are present in warm coastal waters. Impulsive sound sources provide high intensity, broadband signals that facilitate accurate arrival time detections across each vehicle. This makes them useful references for a self-localization solution. However, the similarity between different signals presents a significant correspondence problem, which must be solved to provide accurate estimates of the changing geometry of the swarm. A geometric solution to this correspondence problem is shown and an optimization procedure is proposed to track the geometry of a swarm as it changes. The method is verified using a swarm of 17 self-ballasting subsurface floats that independently drifted with currents off of the coast of San Diego, California. The changing geometry of the floats was estimated using both an acoustic localization system and the proposed approach. The two estimates show good agreement, validating our method. We believe that this new localization strategy is useful for high endurance, low power, and multi-vehicle surveys.

Hardware has become an increasingly attractive target for attackers, yet we still largely lack tools that enable us to analyze large designs for security flaws. Information flow tracking (IFT) models provide an approach to verifying a hardware design's adherence to security properties related to isolation and reachability. However, existing precise IFT models are usually too complex to actually use. Queries may fail to finish even for small designs when verifying relatively simple properties. It is possible to create less complex models, but these come at the cost of a severe loss of precision - -they frequently indicate a property fails when in fact it passes, which means verification requires extensive additional manual investigation. We present a new method to bridge the chasm between precision and complexity in a finer-grained, controlled, and disciplined manner. Our method allows using the most appropriate precision/complexity tradeoff for the design size and available computing resources, meaning it is now possible to create models that are not too complex to be usable, but which offer more precision (fewer false positives) than was previously possible.

Acoustic methods are an established technique to monitor marine mammal populations and behavior, but developments in computer science can expand the current capabilities. A central aim of these methods is the automated detection and classification of marine mammal vocalizations. While many studies have applied bioacoustic methods to cetacean calls, there has been limited success with humpback whale (Megaptera novaeangliae) social call classification, which has largely remained a manual task in the bioacoustics community. In this project, we automated this process by analyzing spectrograms of calls using PCA-based and connected-component-based methods, and derived features from relative power in the frequency bins of these spectrograms. These features were used to train and test a supervised Hidden Markov Model (HMM) algorithm to investigate classification feasibility. We varied the number of features used in this analysis by varying the sizes of frequency bins. Generally, we saw an increase in precision, recall, and accuracy for all three classified groups, across the individual data sets, as the number of features decreased. We will present the classification rates of our algorithm across multiple model parameters. Since this method is not specific to humpback whale vocalizations, we hope it will prove useful in other acoustic applications.

Measuring the ever-changing 3-dimensional (3D) motions of the ocean requires simultaneous sampling at multiple locations. In particular, sampling the complex, nonlinear dynamics associated with submesoscales (

This paper presents a demonstration of ambient acoustic noise processing on a set of free floating oceanic receivers whose relative positions vary with time. It is shown that it is possible to retrieve information that is relevant to the travel time between the receivers. With thousands of short time cross-correlations (10 s) of varying distance, it is shown that on average, the decrease in amplitude of the noise correlation function with increased separation follows a power law. This suggests that there may be amplitude information that is embedded in the noise correlation function. An incoherent beamformer is developed, which shows that it is possible to determine a source direction using an array with moving elements and large element separation. This incoherent beamformer is used to verify cases when the distribution of noise sources in the ocean allows one to recover travel time information between pairs of mobile receivers.

De novo assembly is a widely used methodology in bioinformatics. However, the conventional short-readbased de novo assembly is incapable of reliably reconstructing the large-scale structures of human genomes. Recently, a novel optical label-based technology has enabled reliable large-scale de novo assembly. Despite its advantage in large-scale genome analysis, this new technology requires a more computationally intensive alignment algorithm than its conventional counterpart. For example, the runtime of reconstructing a human genome is on the order of 10,000 hours on a sequential CPU. Therefore, in order to practically apply this new technology in genome research, accelerated approaches are desirable. In this article, we present three different accelerated approaches, multicore CPU, GPU, and FPGA. Against the sequential software baseline, our multicore CPU design achieved an 8.4× speedup, while the GPU and FPGA designs achieved 13.6× and 115× speedups, respectively. We also discuss the details of the design space exploration of this new assembly algorithm on these three different devices. Finally, we compare these devices in performance, optimization techniques, prices, and design efforts.