Special Hardware Research Articles

Self-Organized Operational Neural Networks for Early Glaucoma Diagnosis in Digital Fundus Images

Glaucoma leads to permanent vision disability by damaging the optical nerve that transmits visual images to the brain. The fact that glaucoma does not show any symptoms as it progresses and cannot be stopped at the later stages, makes it critical to be diagnosed in its early stages. Although various deep learning models have been applied for detecting glaucoma from digital fundus images, due to the scarcity of labeled data, their generalization performance was limited along with high computational complexity and special hardware requirements. In this study, compact Self-Organized Operational Neural Networks (Self-ONNs) are proposed for early detection of glaucoma in fundus images and their performance is compared against the conventional (deep) Convolutional Neural Networks (CNNs) over three benchmark datasets: ACRIMA, RIM-ONE, and ESOGU. The experimental results demonstrate that Self-ONNs not only achieve superior detection performance but can also significantly reduce the computational complexity making it a potentially suitable network model for biomedical datasets especially when the data is scarce.

Single-layer digitized-counterdiabatic quantum optimization for p-spin models

Abstract Quantum computing holds the potential for quantum advantage in optimization problems, which requires advances in quantum algorithms and hardware specifications. Adiabatic quantum optimization is conceptually a valid solution that suffers from limited hardware coherence times. In this sense, counterdiabatic quantum protocols provide a shortcut to this process, steering the system along its ground state with fast-changing Hamiltonian. In this work, we take full advantage of a digitized-counterdiabatic quantum optimization algorithm to find an optimal solution of the p-spin model up to four-local interactions. We choose a suitable scheduling function and initial Hamiltonian such that a single-layer quantum circuit suffices to produce a good ground-state overlap. By further optimizing parameters using variational methods, we solve with unit accuracy two-spin, three-spin, and four-spin problems for 100%, 93%, and 83% of instances, respectively. As a particular case of the latter, we also solve factorization problems involving 5, 9, and 12 qubits. Due to the low computational overhead, our compact approach may become a valuable tool towards quantum advantage in the NISQ era.

Comparison of preclinical PET scanners performance with varying axial fields-of-view

In preclinical Positron Emission Tomography (PET) imaging, understanding a PET system's performance is crucial for comparing its results with other systems’. In multicentre studies, the performance of the participating systems is essential when planning the experiments [1,2]. In this study, a comprehensive performance evaluation of three PET systems was carried out. The Mediso PET systems were installed in 2023, 2022, and 2014. The newest system is a three-ring system, with 150 mm axial field-of-view (FOV), while the others are two-ring systems. The reconstruction software has undergone multiple updates over time for each system. Other hardware specifications and the crystal matrix are consistent across the three systems. The NEMA NU-4 2008 assessments [3], and linearity measurements were performed. The resolution was similar across the three systems, with filtered-back-projection (FBP) and with iterative reconstruction at the centre of the FOV. The absolute sensitivity within the 250-750 keV range was highest in the newest 3-ring system, while it was similar with the other 2-ring scanners. The mouse 250-750 keV NEC peak was considerably higher with the newest system. Image quality results are presented in Figure 1. The linearity of all systems was within acceptable limits in the 0.5-120 MBq range. All results of the newest system are preliminary, before publication. The greatest differences arose from the extended FOV of the newest system. The impact of software advancements and system ageing is evident in our results, particularly in the image quality phantom. Although the systems are similar, their performances vary significantly. When designing a study involving these systems, factors such as ageing, software version, and the length of the FOV must be considered. Please click on the 'PDF' for the full abstract!

Mobile Devices in Forest Mensuration: A Review of Technologies and Methods in Single Tree Measurements

Mobile devices such as smartphones, tablets or similar devices are becoming increasingly important as measurement devices in forestry due to their advanced sensors, including RGB cameras and LiDAR systems. This review examines the current state of applications of mobile devices for measuring biometric characteristics of individual trees and presents technologies, applications, measurement accuracy and implementation barriers. Passive sensors, such as RGB cameras have proven their potential for 3D reconstruction and analysing point clouds that improve single tree-level information collection. Active sensors with LiDAR-equipped smartphones provide precise quantitative measurements but are limited by specific hardware requirements. The combination of passive and active sensing techniques has shown significant potential for comprehensive data collection. The methods of data collection, both physical and digital, significantly affect the accuracy and reproducibility of measurements. Applications such as ForestScanner and TRESTIMATM have automated the measurement of tree characteristics and simplified data collection. However, environmental conditions and sensor limitations pose a challenge. There are also computational obstacles, as many methods require significant post-processing. The review highlights the advances in mobile device-based forestry applications and emphasizes the need for standardized protocols and cross-device benchmarking. Future research should focus on developing robust algorithms and cost-effective solutions to improve measurement accuracy and accessibility. While mobile devices offer significant potential for forest surveying, overcoming the above-mentioned challenges is critical to optimizing their application in forest management and protection.

CMN: a co-designed neural architecture search for efficient computing-in-memory-based mixture-of-experts

Artificial intelligence (AI) has experienced substantial advancements recently, notably with the advent of large-scale language models (LLMs) employing mixture-of-experts (MoE) techniques, exhibiting human-like cognitive skills. As a promising hardware solution for edge MoE implementations, the computing-in-memory (CIM) architecture collocates memory and computing within a single device, significantly reducing the data movement and the associated energy consumption. However, due to diverse edge application scenarios and constraints, determining the optimal network structures for MoE, such as the expert’s location, quantity, and dimension on CIM systems remains elusive. To this end, we introduce a software-hardware co-designed neural architecture search (NAS) framework, CIM-based MoE NAS (CMN), focusing on identifying a high-performing MoE structure under specific hardware constraints. The results of the NYUD-v2 dataset segmentation on the RRAM (SRAM) CIM system reveal that CMN can discover optimized MoE configurations under energy, latency, and performance constraints, achieving 29.67× (43.10×) energy savings, 175.44×(109.89×) speedup, and 12.24× smaller model size compared to the baseline MoE-enabled Visual Transformer, respectively. This co-design opens up an avenue toward high-performance MoE deployments in edge CIM systems.

Custom RISC-V architecture incorporating memristive in-memory computing

Due to the rise in data-intensive applications, the von Neumann bottleneck is increasingly restricting modern computer architectures, resulting to latency and energy consumption. Addressing this challenge necessitates a CMOS-compatible solution with high energy efficiency and significant parallelism. Utilizing resistive switching components within a 1T1R crossbar array and the application of Stanford RRAM model, this paper suggests an original method for in-memory computing. Moreover, this work shows a new way to advance the popular RISC-V architecture by including memristive crossbar array. It does this by adding a custom instruction set, special hardware blocks, and the Scouting Logic Scheme. These modifications serve both as a comprehensive testbed for the memory system and a proof of concept for the future integration of memristors in computing architectures. The proposed design undergoes extensive testing and power analysis to validate its functionality and performance under various conditions. The results demonstrate significant improvements in computational efficiency and energy savings, highlighting the potential of memristor-based in-memory computing systems to overcome current architectural limitations.

Intermediate-grained kernel elements pruning with structured sparsity

Neural network pruning provides a promising prospect for the deployment of neural networks on embedded or mobile devices with limited resources. Although current structured strategies are unconstrained by specific hardware architecture in the phase of forward inference, the decline in classification accuracy of structured methods is beyond the tolerance at the level of general pruning rate. This inspires us to develop a technique that satisfies high pruning rate with a small decline in accuracy and has the general nature of structured pruning. In this paper, we propose a new pruning method, namely KEP (Kernel Elements Pruning), to compress deep convolutional neural networks by exploring the significance of elements in each kernel plane and removing unimportant elements. In this method, we apply a controllable regularization penalty to constrain unimportant elements by adding a prior knowledge mask and obtain a compact model. In the calculation procedure of forward inference, we introduce a sparse convolution operation which is different from the sliding window to eliminate invalid zero calculations and verify the effectiveness of the operation for further deployment on FPGA. A massive variety of experiments demonstrate the effectiveness of KEP on two datasets: CIFAR-10 and ImageNet. Specially, with few indexes of non-zero weights introduced, KEP has a significant improvement over the latest structured methods in terms of parameter and float-point operation (FLOPs) reduction, and performs well on large datasets.

Critical infrastructure monitoring in CBRNe scenarios: a reliable and robust communication network for distributed multimodal sensors

The malicious misuse of CBRNe agents can inflict extensive damage to critical infrastructure and terrorize public society. The effects of such attacks can range from financial or structural drawbacks to significant casualties. Recent terror attacks in Cucuta (2021) or Kabul (2021) demonstrate the need to protect infrastructures such as airports, power plants, and transportation infrastructure. In the face of these modern-day threats, the need for a reliable and effective monitoring system for critical infrastructure has become increasingly important. In a first contribution to the long-term development, we present an initial version of a concept that requires the implementation paradigm of diverse redundancy and secure communication using MQTTS relying on the included cipher suite. An initial set of node types is defined and assigned requirements for implementation. The software tools architecture we propose based on these requirements is designed to support network operators and developers by providing standard features for network management and sensor node implementation. This includes the definition of a secure semi-automatic onboarding process for new sensor nodes, which is presented in detail. We strive for independency from specific hardware platforms, software frameworks, and network technologies to enable an open standard for communication within a critical infrastructure sensor network and also between such networks in the future.

Mentor: A Memory-Efficient Sparse-dense Matrix Multiplication Accelerator Based on Column-Wise Product

Sparse-dense matrix multiplication (SpMM) is the performance bottleneck of many high-performance and deep-learning applications, making it attractive to design specialized SpMM hardware accelerators. Unfortunately, existing hardware solutions do not take full advantage of data reuse opportunities of the input and output matrices or suffer from irregular memory access patterns. Their strategies increase the off-chip memory traffic and bandwidth pressure, leaving much room for improvement. We present Mentor , a new approach to designing SpMM accelerators. Our key insight is that column-wise dataflow, while rarely exploited in prior works, can address these issues in SpMM computations. Mentor is a software-hardware co-design approach for leveraging column-wise dataflow to improve data reuse and regular memory accesses of SpMM. On the software level, Mentor incorporates a novel streaming construction scheme to preprocess the input matrix for enabling a streaming access pattern. On the hardware level, it employs a fully pipelined design to unlock the potential of column-wise dataflow further. The design of Mentor is underpinned by a carefully designed analytical model to find the trade-off between performance and hardware resources. We have implemented an FPGA prototype of Mentor . Experimental results show that Mentor achieves speedup by geomean 2.05 × (up to 3.98 ×), reduces the memory traffic by geomean 2.92 × (up to 4.93 ×), and improves bandwidth utilization by geomean 1.38 × (up to 2.89 ×), compared to the state-of-art hardware solutions.

Optimization of algorithms for effective management of the measurement process in electrical impedance tomography and ultrasonic tomography

The article discusses the design of advanced embedded algorithms that aim to manage and control the measurement process using the Electrical Impedance Tomography and Ultrasonic Tomography methods. The project aims to develop solutions that optimize the performance of embedded devices that must operate on limited resources such as memory, computing power, or bandwidth. Minimizing energy consumption is an important aspect, especially for battery-powered devices. Algorithms must also be adapted to specific hardware constraints, such as low RAM or CPU limitations, which require complex engineering and optimization. Additionally, the project assumes the implementation of algorithms that consider security aspects, which is crucial in protecting data processed and transmitted by the device. It also requires the development of effective communication methods between the embedded device and other systems, including appropriate communication protocols.

Design of an image processing system for heterogeneous computer architectures

Embedded real-time image processing systems are widely applied in various fields, enhancing visual experience and making life more convenient. With the development of embedded processors and operating systems, as well as in-depth research into image processing algorithms, the effects and speed of image processing have been significantly improved. This paper proposes an embedded image processing system design based on the current technological developments, centered around the Zybo processor. This includes a collaborative hardware and software design, specific hardware circuit design, and the design of a system software platform. The paper also discusses the establishment of a development environment for the Zybo system, the transplantation of the Linux operating system, the design of the file system, and the application of OpenCV library functions in image processing.

Software Optimization and Design Methodology for Low Power Computer Vision Systems

This tutorial paper addresses a low power computer vision system as an example of a growing application domain of neural networks, exploring various technologies developed to enhance accuracy within the resource and performance constraints imposed by the hardware platform. Focused on a given hardware platform and network model, software optimization techniques, including pruning, quantization, low-rank approximation, and parallelization, aim to satisfy resource and performance constraints while minimizing accuracy loss. Due to the interdependence of model compression approaches, their systematic application is crucial, as evidenced by winning solutions in the Lower Power Image Recognition Challenge (LPIRC) of 2017 and 2018. Recognizing the typical heterogeneity of processing elements in contemporary hardware platforms, the effective utilization through parallelizing neural networks emerges as increasingly vital for performance enhancement. The paper advocates for a more impactful strategy—designing a network architecture tailored to a specific hardware platform. For detailed information on each technique, the paper provides corresponding references.

Combining Machine Learning techniques and Genetic Algorithm for predicting run times of High Performance Computing jobs

This study proposes a novel approach combining Machine Learning (ML) techniques and Genetic Algorithms (GA) for predicting High-Performance Computing (HPC) job run times. The objective is to create a prediction method universally applicable to any HPC system, irrespective of workload characteristics, application specific parameters, user behavior, or hardware architecture. Since user-supplied run time estimates are often inaccurate, we aim to categorize job runtimes into several classes, allowing users to select appropriate classes for their jobs. A Genetic Algorithm is developed to optimally define these runtime classes, determining both the number of classes and the time intervals represented by them. Four Machine Learning algorithms (K Nearest Neighbours, Support Vector Regression, Extreme Gradient Boosting and Deep Neural Networks) are implemented for run time prediction. A unique set of features extracted from historical job data serves as input to the Machine Learning models. The generalized nature of our method is demonstrated by validating its performance on data from six clusters with distinct configurations, applications and runtime distributions. Our results illustrate the superior performance of Machine Learning models incorporating GA-defined runtime classes for all datasets. Across all six datasets, our method achieves R2 scores exceeding 0.8, and accuracy greater than 0.7.

Nonintrusive and Effective Volume Reconstruction Model of Swimming Sturgeon Based on RGB-D Sensor.

The sturgeon is an important commercial aquaculture species in China. The measurement of sturgeon mass plays a remarkable role in aquaculture management. Furthermore, the measurement of sturgeon mass serves as a key phenotype, offering crucial information for enhancing growth traits through genetic improvement. Until now, the measurement of sturgeon mass is usually conducted by manual sampling, which is work intensive and time consuming for farmers and invasive and stressful for the fish. Therefore, a noninvasive volume reconstruction model for estimating the mass of swimming sturgeon based on RGB-D sensor was proposed in this paper. The volume of individual sturgeon was reconstructed by integrating the thickness of the upper surface of the sturgeon, where the difference in depth between the surface and the bottom was used as the thickness measurement. To verify feasibility, three experimental groups were conducted, achieving prediction accuracies of 0.897, 0.861, and 0.883, which indicated that the method can obtain the reliable, accurate mass of the sturgeon. The strategy requires no special hardware or intensive calculation, and it provides a key to uncovering noncontact, high-throughput, and highly sensitive mass evaluation of sturgeon while holding potential for evaluating the mass of other cultured fishes.

A single‐channel transmitarray antenna with abilities of super resolution and digital beamforming

AbstractA single‐channel transmitarray antenna with abilities of super resolution and digital beamforming (DBF) is proposed. The amplitude and phase distributions on the array aperture are reconstructed through beam switching with a single channel. By finding and solving relations between distributions and received signals of beams, aperture distributions can be reconstructed. The ill‐conditioned problem caused by limited scanning range and angular interval is solved using the regularization method. The obtained distributions can then be delivered to super‐resolution algorithm to obtain super resolution, or used to produce shaped beams by DBF. On the basis of the proposed theory, a bifocal beam‐reconfigurable transmitarray is designed and tested as a single‐channel DBF antenna without special and additional hardware design. It has the outstanding advantages of low cost, simple structure, low energy consumption, and high working frequency.

Optimization of acquisition patterns for establishing inter CubeSat optical communications

As commercially available CubeSats with up to six standardized units cannot achieve the precision required for an instantaneous establishment of a low-divergence optical inter-satellite link, search patterns are used to scan the remaining field of uncertainty. This analysis optimizes the simultaneously executed search pattern combinations of the two laser communication terminals involved. Based on a Monte Carlo simulation, the perturbations on these links are investigated, and the corresponding key performance parameters such as mean acquisition time and success rate are calculated. The results are penalized by the hardware specifications, including actuator and sensor bandwidths, given by their design. Residual attitude error components imply a significant influence on the acquisition process and are therefore presented within this work. The pattern pairs are fed through an automated optimization algorithm to tune and analyze them. In this particular scenario of two CubeISL models, the mean duration for a first detected acquisition hit is within a pattern period of 3.2 s for the best performing pairs spiral-rose and lissajous-rose. Assuming an uncertainty field of ±0.2deg due to limited attitude knowledge, success rates between 82.3% and 99.9% are achieved.

Revisiting Database Indexing for Parallel and Accelerated Computing: A Comprehensive Study and Novel Approaches

While the importance of indexing strategies for optimizing query performance in database systems is widely acknowledged, the impact of rapidly evolving hardware architectures on indexing techniques has been an underexplored area. As modern computing systems increasingly leverage parallel processing capabilities, multi-core CPUs, and specialized hardware accelerators, traditional indexing approaches may not fully capitalize on these advancements. This comprehensive experimental study investigates the effects of hardware-conscious indexing strategies tailored for contemporary and emerging hardware platforms. Through rigorous experimentation on a real-world database environment using the industry-standard TPC-H benchmark, this research evaluates the performance implications of indexing techniques specifically designed to exploit parallelism, vectorization, and hardware-accelerated operations. By examining approaches such as cache-conscious B-Tree variants, SIMD-optimized hash indexes, and GPU-accelerated spatial indexing, the study provides valuable insights into the potential performance gains and trade-offs associated with these hardware-aware indexing methods. The findings reveal that hardware-conscious indexing strategies can significantly outperform their traditional counterparts, particularly in data-intensive workloads and large-scale database deployments. Our experiments show improvements ranging from 32.4% to 48.6% in query execution time, depending on the specific technique and hardware configuration. However, the study also highlights the complexity of implementing and tuning these techniques, as they often require intricate code optimizations and a deep understanding of the underlying hardware architecture. Additionally, this research explores the potential of machine learning-based indexing approaches, including reinforcement learning for index selection and neural network-based index advisors. While these techniques show promise, with performance improvements of up to 48.6% in certain scenarios, their effectiveness varies across different query types and data distributions. By offering a comprehensive analysis and practical recommendations, this research contributes to the ongoing pursuit of database performance optimization in the era of heterogeneous computing. The findings inform database administrators, developers, and system architects on effective indexing practices tailored for modern hardware, while also paving the way for future research into adaptive indexing techniques that can dynamically leverage hardware capabilities based on workload characteristics and resource availability.

Protecting Instant Messaging Notifications against Physical Attacks: A Novel Instant Messaging Notification Protocol Based on Signal Protocol

Over the years, there has been a significant surge in the popularity of instant messaging applications (IMAs). However, the message notification functionality in IMAs exhibits certain limitations. Some IMAs fail to alert users about new messages after their phone restarts unless they unlock the phone. This is a consequence of end-to-end encryption (E2EE) and the app not knowing the message is in the queue until the app decrypts it. This approach using E2EE is used to prevent offline attacks, as the key is unavailable to decrypt the notification messages. In this paper, we introduce a novel design and implementation of a message notification protocol for IMAs based on the Signal protocol. The proposed protocol aims to securely display notifications on a locked device and ensures that cryptographic keys are stored in a location that is isolated from the user’s device to prevent offline attacks. This approach enhances the security of private key storage, safeguarding private keys against various external threats. The innovative design strengthens the off-site key management system, rendering it resilient against offline attacks and mitigating the risk of key compromise. Additionally, the proposed protocol is highly efficient, requiring no specialized hardware for implementation. It offers confidentiality of cryptographic keys and protection against offline attacks, further enhancing the overall security of the system. We evaluate the protocol’s effectiveness by analyzing multiple independent implementations that pass a suite of formal tests via ProVerif.

Runtime support for CPU-GPU high-performance computing on distributed memory platforms

IntroductionHardware heterogeneity is here to stay for high-performance computing. Large-scale systems are currently equipped with multiple GPU accelerators per compute node and are expected to incorporate more specialized hardware. This shift in the computing ecosystem offers many opportunities for performance improvement; however, it also increases the complexity of programming for such architectures.MethodsThis work introduces a runtime framework that enables effortless programming for heterogeneous systems while efficiently utilizing hardware resources. The framework is integrated within a distributed and scalable runtime system to facilitate performance portability across heterogeneous nodes. Along with the design, this paper describes the implementation and optimizations performed, achieving up to 300% improvement on a single device and linear scalability on a node equipped with four GPUs.ResultsThe framework in a distributed memory environment offers portable abstractions that enable efficient inter-node communication among devices with varying capabilities. It delivers superior performance compared to MPI+CUDA by up to 20% for large messages while keeping the overheads for small messages within 10%. Furthermore, the results of our performance evaluation in a distributed Jacobi proxy application demonstrate that our software imposes minimal overhead and achieves a performance improvement of up to 40%.DiscussionThis is accomplished by the optimizations at the library level and by creating opportunities to leverage application-specific optimizations like over-decomposition.

Detection of Lowering in Sport Climbing Using Orientation-Based Sensor-Enhanced Quickdraws: A Preliminary Investigation.

Climbing gyms aim to continuously improve their offerings and make the best use of their infrastructure to provide a unique experience for their clients, the climbers. One approach to achieve this goal is to track and analyze climbing sessions from the beginning of the ascent until the climber's descent. Detecting the climber's descent is crucial because it indicates when the ascent has ended. This paper discusses an approach that preserves climber privacy (e.g., not using cameras) while considering the convenience of climbers and the costs to the gyms. To this aim, a hardware prototype has been developed to collect data using accelerometer sensors attached to a piece of climbing equipment mounted on the wall, called a quickdraw, which connects the climbing rope to the bolt anchors. The sensors are configured to be energy-efficient, making them practical in terms of expenses and time required for replacement when used in large quantities in a climbing gym. This paper describes the hardware specifications, studies data measured by the sensors in ultra-low power mode, detects sensors' orientation patterns during descent on different routes, and develops a supervised approach to identify lowering. Additionally, the study emphasizes the benefits of multidisciplinary feature engineering, combining domain-specific knowledge with machine learning to enhance performance and simplify implementation.