Applications in Electronics Pervading Industry, Environment and Society

This article aims to sketch the figure of Alessandro De Gloria, a professor who dedicated his entire life, with generosity and enthusiasm, to engineering and scientific research. He designed one of the first chips in an Italian university and then a set of digital system architectures that contributed to shaping the then-fledgling field of microprocessors. Also, by founding the Applepies conference (International Conference on Applications in Electronics Pervading Industry, Environment and Society), he promoted the intuition of the value for the whole population of the applications of electronic systems. He pioneered the field of mobile apps and started a fruitful dialogue with the humanistic culture, particularly exploiting virtual reality technologies. In the last few years, he also contributed to machine learning and big data, particularly in the domain of automated driving. He founded an MSc program on strategy, intending to shape education in that field from a rigorous engineering point of view. We believe that his generous and enthusiastic academic life can be an outstanding example in the face of today’s challenges of Electronic Engineering and higher education in the broader societal context.

Among all the fault tolerance (FT) techniques developed over the years, Dual-core lock-step techniques have emerged as an effective approach to enhance the fault tolerance capabilities of these systems. However, they cannot withstand Hard Errors occurring in the architecture, and they have certain drawbacks for checkpoints and restore methodologies necessary to save and restore the correct state of the core. This paper shows an execution paradigm within a new architectural approach to overcome the disadvantages related to the long checkpointing/restoring procedures that affect the classical lock-step architectures, also improving the hard, destructive faults resilience, leveraging the advantages of the already published dynamic-TMR technique inside the Klessydra-dfT03 architecture.

This paper explores the impact of wire bonding solutions on high-speed serial links, focusing on the co-design between on-chip and off-chip parameters to achieve a boost in system bandwidth. Despite traditional limitations, wire bonding can outperform flip-chip techniques when properly designed. This study investigates various cases with different in-band ripples and describes how to optimize the on-chip and bonding parameters accordingly. Additionally, the adoption of shunt peaking techniques is examined, which allows us to achieve a bandwidth boost of up to 3.82 times. While on-chip parameters can be well controlled, the limited mechanical precision of wire bonding may lead to suboptimal wire lengths. An analysis of the variation in bandwidth boost and ripple is presented, indicating that wire length variations below 10% result in minimal impact on boosting effects and ripple. In conclusion, wire bonding offers advantages for high-speed serial links, and co-design optimizes system performance. Properly designed, wire bonding is an attractive choice for high-speed applications, outperforming flip-chip techniques in bandwidth enhancement.

CRYSTALS-Kyber is a lattice-based key encapsulation mechanism (KEM) recognized as one of the finalist algorithms in NIST’s post-quantum cryptography (PQC) standardization process. Polynomial multiplications and hash functions, as essential operations in lattice-based PQC schemes, pose a significant time consumption challenge with respect to nowadays cryptographic protocols. This work addresses these computational efforts by incorporating LOKI, an accelerator, into a RISC-V microcontroller. By leveraging the accelerator, the performance can be enhanced, contributing to the overall efficiency of Kyber in the fundamental tasks of key generation, encryption, and decryption operations. Through empirical evaluations and benchmarking, the effectiveness and practicality of the proposed hardware architectures are demonstrated, highlighting their potential to advance the field of post-quantum cryptography.

Efficient and cost-effective functional verification strategies are more and more essential in digital integrated system design. This paper presents a low-cost approach to meet this challenge, introducing a universal hardware emulator designed for verifying various Intellectual Property (IP) cores on FPGA. We demonstrate the practicality and performance of this emulator through a use-case of verifying an advanced Integer Arithmetic Logic Unit (ALU) for the RISC-V ISA vector extension. The process of integration and the results obtained from the verification are presented and discussed. Preliminary findings indicate that the emulator can perform more than 1.1 million tests per second. This research contributes to the advancement of hardware verification techniques, providing researchers, universities, and small businesses with an accessible and effective solution.

The miniaturization of electronic devices and the improved operating speeds increase the likelihood of single event faults. Differently from Single Event Upset (SEU) faults, Single Event Transient (SET) faults generally affect combinational logic, making all voting systems vulnerable to errors. The proposed work uses an ad-hoc fault-simulation campaign employing signal glitching to identify SET vulnerabilities inside a RISC-V core already equipped with resilience logic against Single Event Upset (SEU) faults. The faults target the majority voting logic structures, highlighting how they can be susceptible to faults depending on the width of the injected pulses, and showing how the use of Buffered Triple Modular Redundancy (BTMR) allows decreasing the total failure probability due to erroneous majority voters.

Latest technological improvements and investments from government agencies and private companies pushed to the limits the requirements related to both data rate speed and security of the communication links in space applications. The high volume of data and the continuous integration of services opened the path to hackers for new and increasingly diffused cyberattacks. Governmental agencies are attempting to stem this problem by issuing and updating accordingly a series of reports and standards through the Consultative Committee for Space Data Systems (CCSDS). In this work, we present the implementation of an Advanced Encryption Standard—Cipher-based Message Authentication Code (AES-CMAC) core on space-grade FPGAs, that is compliant with the latest CSSDS security standards and outperforms the state-of-the-art in terms of resource efficiency.

The Advanced Encryption Standard (AES) is widely accepted as the de-facto standard for symmetric-key encryption, and it is going to be used in the coming decades because of its resistance against Post-Quantum Cryptography. For this reason, it is the subject of many research works, and almost all converge on the usage of composite/tower fields for the hardware implementation of the S-box, the most expensive circuit in terms of both area and critical delay. Anyway, the debate is still open on applying isomorphic fields also to the other AES algorithm operations. In the attempt to give an answer, it is analyzed the application of the two approaches to the most recent and performing solutions from the state-of-the-art with the synthesis of the corresponding circuits on a 7 nm standard-cell technology. In addition, the presented work constitutes also a guideline for implementing hardware AES modules that execute all operations over composite/tower fields.

This work presents the acceleration of a Bounded Algorithmic Number (BAN) library exploiting vector instructions in general-purpose processors. With the use of this encoding, it is possible to represent non-Archimedean numbers that are not only finite (like real numbers) but also infinite or infinitesimal. The tremendous growth in non-Archimedean numerical computations over the past 20 years and the resulting applications spurred this study’s development. Enabling acceleration of BANs processing can significantly increase the throughput of non-Archimedean numerical computations, enlarging the spectrum of possible applications to industrial and real-time ones.

The necessity of efficient comparators in modern ADCs calls for new two-stages topologies that overcome the Strongarm limitations in terms of common-mode offset and gain dependency. The latest fashion is represented by dynamic bias integrators coupled with low power and low noise latches. The dynamic bias significantly reduces the overall power consumption, a feature which the second stage has to maintain; also, a CMOS implementation is usually adopted to gain more robustness. Starting from the comparator with dynamic floating inverter amplifier [1], a new structure has been derived and explained. The architecture has been simulated in 5 nm FinFET technology and compared to the state-of-the-art for very stringent power, noise and speed targets. Nonetheless, the proposed topology matches various applications thanks to its multiple degrees of freedom which allow the designer plenty of room for further improvements. The continuously scaling technology will favour this CMOS dynamic bias implementation even more.

We propose a Q-Learning hardware accelerator for a RISC-V platform. In particular, our work focuses on the Klessydra processor. To the best of our knowledge, this is the first work in the literature that addresses this topic. We implemented the system on an AMD-Xilinx ZedBoard development board using a small amount of hardware resources and requiring a limited dynamic power of 1.528 W. The data we obtained are compatible with the future implementation of more accelerators on the same device to enhance the capabilities of the system. Compared to a standard software version of the algorithm, our accelerator allows a speed-up of $$\times 36$$ × 36 in convergence time and an energy saving of $$\times 34$$ × 34 . The results obtained prove how our proposed system is suitable for high-speed and low-energy applications like Edge Machine Learning and embedded IoT systems.

Single-Flux-Quantum (SFQ) logic is a digital electronic technology known for its very low-power consumption (nW-µW) and high operating frequency (up to 100 GHz). Like any other device, SFQ-based logic circuits suffer from manufacturing process issues, specifically concerning variations in the determined values of individual components such as the critical current of a Josephson junction and inductances. This leads to the need for a deep understanding of the circuit performances, its tolerance range and, furthermore, an optimization tool to improve it achieving a certain margin for each component. In this regard, the present article delves into the techniques and the development of a new design parameter optimization algorithm, whose main goal is to increase the critical margin of the circuit. By using such a simple and efficient technique, failures due to the fabrication are avoided and performance enhancement is achieved.

Recently, there has been a growing interest in Physically Unclonable Functions (PUFs). These electronic circuits possess several key characteristics such as unpredictability and uniqueness that make them particularly attractive for security applications. PUFs offer an appealing solution for secure boot applications, providing a hardware-based mechanism for generating unique cryptographic keys. These keys can be used to encrypt the bootloader and operating system, thereby enhancing security. In this paper, we propose an innovative, secure boot scheme that leverages the functionality and characteristics of a PUF. Our approach eliminates the need for physical storage of the encryption key of the boot code, which enhances security and provides the possibility of securely updating the firmware. We will present an architecture that comprises essential components, along with a demo board on FPGA. The demo board features a general-purpose 64-bit RISC-V-based system that leverages the proposed PUF-based secure architecture, enabling secure boot and firmware update functionalities.

Data analysis has become a common practice in professional and amateur sport activities, to monitor the player state and enhance performance. In tennis, performance analysis requires detecting and recognizing the different types of shots. With the advances in microcontrollers and machine learning algorithms, this topic becomes ever more considerable. We propose a 1-D convolutional neural network (CNN) model and an embedded system based on Arduino-Nano system for real-time shot classification. The network is trained through a dataset composed of three different tennis shot types, with 6 features recorded by an inertial device placed on the racket. Results demonstrate that the proposed model is able to discriminate the tennis shots with high accuracy, also generalizing to different users. The network has been deployed on a low-cost Arduino nano 33 IoT model, with an inference time of 65 ms.

Machine learning allows designing intelligent sensing networks capable to perform automatic inferences about the integrity of technical facilities. Compression techniques decrease significantly energy requirements of the sensing networks proving essential when sensing nodes are not supported by constant power sources. Existing schemes pass through the reconstruction of the original time series data before moving to the diagnosis phase. However, this passage can be avoided, i.e., inference can be performed directly in the compressed domain, by exploiting the specific information retained in the compressed patterns. This paper fulfills the goal above in the context of vibration-based structural health monitoring by proving, from an empirical perspective, that Convolutional Neural Networks (CNNs) can be used to predict the structural health status directly in the compressed domain when properly combined with adapted Compressed Sensing mechanisms. Importantly, the study analyses the effect of the intrinsic noise that affects digital accelerometer sensors. Results confirm that CNNs can mine information in the compressed domain even in presence of strong noise components, i.e., accuracy remains above 94% even for ultra-low-cost solutions featuring a signal-to-noise-ratio below 20 dB.

Driving a motorized wheelchair is not without risk and requires high cognitive effort to obtain a good environmental perception to complete tasks such as obstacle avoidance and path planning. Nonetheless, completely autonomous wheelchairs release the users from any cognitive and muscular effort, resulting in a progressive degradation of their residual capacities. Thus, we developed a collaborative driver assistance system that engages selectively, specifically when it detects a positive or negative height object within a defined safety zone. The system first reduces linear speed to assist and prioritize the user in making a decision and ultimately calculates a steering direction that avoids obstacles. An extensible mathematical model is devised by relying on lidar and ultrasonic sensors. Simulations have been carried out, confirming the effectiveness of this proof-of-concept.

Traditional tiny machine learning systems are widely employed because of their limited energy consumption, fast execution, and easy deployment. However, such systems have limited access to labelled data and need periodic maintenance due to the evolution of data distribution (i.e., context drift). Continual machine learning algorithms can enable continuous learning on embedded systems by updating their parameters, addressing context drift, and allowing neural networks to learn new categories over time. However, the availability of labelled data is scarce, limiting such algorithms in supervised settings. This paper overcomes this limitation with an alternative approach which combines supervised deep learning with unsupervised clustering to enable unsupervised continual machine learning on the edge. Tiny Neural Deep Clustering (TinyNDC) is deployed in an OpenMV Cam H7 Plus and tested with the MNIST dataset reaching a classification accuracy of 92.3% and a frame rate of 44 FPS.

This research explores an emerging approach, the adversarial policy learning paradigm, that aims to increase safety and robustness in deep reinforcement learning models for automated driving. We propose an iterative procedure to train an adversarial agent acting in a highway-simulated environment to attack a victim agent that is to be improved. Each training iteration consists of two phases. The adversarial agent is first trained to disrupt the victim-agent policy. The victim model is then trained to overcome the defects observed by the attack from the adversarial agent. The experimental results demonstrate that the victim agent trained with adversarial attacks outperforms the original agent.

This paper evaluates an AI video surveillance application on diverse high-performance computing (HPC) architectures. AI-powered video surveillance has emerged as a vital tool for security and monitoring, relying on hardware infrastructure for efficient processing. We present a benchmark of an AI application based on the YOLO object dection framework to track downed pepole in critical scenarios. This study investigates the impact of different architectural designs, including CPUs and GPUs on video analysis performance. Evaluation metrics encompass computational speed, power consumption and resource utilisation.

This paper presents a finite element (FE) modeling and frequency response analysis of an instrumented wheelset for rail applications. The wheelset was equipped with strain gauges to estimate contact forces at the interface between the wheels and the rails. The objective is to showcase the simulation results of the wheelset state-space reduced model, which represents the dynamic behavior of the FE wheelset. The frequency response results of the FE and state-space models are compared. The paper demonstrates how the reduced model accurately represents the behavior of the FE system, resulting in significantly improved simulation and calculation speed, with a great reduction in memory usage. Furthermore, the state-space representation provides the flexibility to augment and manage data points for enhanced post-processing manipulation.

This research explores the use of optoelectronic tweezers (OET) as a method of manipulating microparticles. It focuses on how the dielectrophoretic (DEP) force can be used to trap solder beads with different concentrations of solution in an OET device, achieving velocities of up to 1350 μm/s. This study analyzes new techniques for creating complex patterns and examines how they are affected by various parameters, including light intensity, pattern size, and solution conductivity, to manipulate particles into desired structures. The freezing-drying technique has been used to fix solder beads with a surface mount technology (SMT) resistor to construct an electronic circuit, and we found that the freezing stage was sufficient when using Peltier to fix the assembled circuit. The paper discusses the temperature and time required for the heating process to connect the circuit by partially melting the solder and measures and analyzes the I-V characteristics to determine the manufactured device's resistance compared to the installed resistor. Overall, this study presents a promising result for microparticle manipulation and demonstrates its potential application in micro-electronic circuit construction or reworking.

The Internet of Things (IoT) paradigm is nowadays pervasive in a variety of applications. Distributed computing, i.e., local processing of data inside IoT nodes is mandatory to reduce power consumption and data communication cost, too. An evaluation of the needed computing power and of the hardware requirements is then as much useful as it is conducted in an early stage of the product development. The possibility to perform on-the-field experiments at an early stage is a desirable feature, too. In this paper, a low cost and open system for IoT evaluation is described. It contains basic blocks of an IoT/IIoT system, and software and hardware structures aimed to assess its performances with negligible overhead. Power consumption and timing characteristics of firmware (MCU) and hardware (FPGA) IoT nodes components can be collected and analyzed, to obtain real system requirements.

We present a microcontroller-based in-flow airborne spectrometer to quantify the presence and activity of radioactive isotopes in particulate matter (PM). It is based on a hollow CsI(Tl) scintillator inside which air flows, thanks to a fan, and PM is captured by a micrometric filter placed in the middle, thus allowing full solid angle capture of gamma rays. The instrument is very compact, fitting in a 10 cm cube and thus being compatible with several unmanned flying vehicles, including an experimental rocket inside which it will be launched as scientific payload to evaluate the robustness of its design during flight. Here we report the mechanical and electronics (for SiPM readout) design, as well as a preliminary characterization. Simulations have allowed to find the optimal compromise in the number and position of SiPMs with an expected energy resolution better than 7% at 662 keV.

This work discusses the flash ADC-based front-end designed for future, high-rate pixel detector applications.The paper includes a description of the front-end blocks within the illustration of the submitted for fabrication prototype chip configuration blocks and the readout structure.

In this article, we report the design and initial testing of a device which paves the path to the realization of a low-cost wearable device for detecting heart and lung tones for telemedical and long-term monitoring applications. In the current practise, the capability to assess chest and lung sound characteristics heavily depends on the experience of the clinician and a device which minimizes the human factor is highly in demand. In particular, this work addresses the development of a smart device based on the novel approach of digital Micro Electro-Mechanical Systems (MEMS) microphones capable of operating in a wide frequency range and low noise ratio. The aim is to propose in future developments Artificial intelligence-based algorithms for patient monitoring, a medical diagnostic decision support system, belonging to the Internet of Medical Things framework.

The rapid expansion of Internet of Things (IoT) applications necessitates the utilization of efficient Low Power Wide Area Network (LPWAN) technologies. Among these technologies, Narrowband IoT (NB-IoT) has emerged as a highly promising solution due to its energy efficiency, long-range communications, and anticipated scalability. In this study, we conducted simulations to examine the impact of various factors on energy usage in NB-IoT devices. Taking into consideration the significance of energy efficiency in IoT devices, particularly in hard-to-reach locations, we study the effects of data compression on energy consumption. The obtained results revealed that implementing data compression techniques can significantly reduce energy consumption, specifically in areas with limited transmission coverage. Additionally, we study the benefits of expanding data transmission intervals to conserve energy. The findings of this paper emphasize the importance of adopting energy-efficient practices in the development of IoT technology.

Exposure to microgravity poses challenges to the astronaut's musculoskeletal system, causing temporary issues that can be addressed through exercise routines. For long-duration space missions, wearable technology with sensors to measure and stimulate muscular activity is crucial. The proposed active suit system is a custom-made intra-vehicular wearable designed to counteract muscle and bone mass loss in microgravity. It consists of an anti-allergic, antibacterial suit with differential compression capabilities, an electronic system to detect body joint movements, an electrostimulation system, and an electronic control unit (ECU) for motion data analysis and stimulation regulation. The suit aims to reduce muscle atrophy by providing intelligent electrostimulation, simulating Earth-like gravity conditions. System calibration on Earth and adaptive neuromuscular stimulation in space are essential steps to maintain astronauts’ muscle tone during extended space missions.

Lithium-ion battery cells are considered the best solution for energy storage systems in Battery Electric Vehicles (BEVs), due to their high energy density, high lifecycle, and low aging impact. Battery performances are strictly connected to operating conditions, especially temperature. Modern Battery Thermal Management Systems (BTMSs) optimize battery usage in terms of temperature to achieve high performance and high safety of the battery. In this paper an electrochemical-thermal coupled 3D model will be developed and discussed. The electrochemical model used is a pseudo-bidimensional model (P2D) which will output the heat source for the thermal model in order to get from this one the temperature of the battery when charging and discharging. The model data will be validated through experimental data coming from a 30 Ah NMC pouch cell used in automotive applications.

Characterization data are essential to improve state estimation algorithms for lithium-ion batteries. Unfortunately, only bigger companies can afford extensive test campaigns, as they require expensive and specific equipment. A Low-Cost, Open-Design, and highly configurable electronic load for battery testing is proposed in this paper. The developed equipment is able to sink a desired current profile configured by using a host computer Python interface. A preliminary experimental validation of the designed electronic load is performed obtaining promising results. This equipment could help smaller companies and laboratories to perform battery characterization tests and increase the amount of available data for state estimation algorithms improvement. For this reason, it is released as Open Hardware/Software system and is freely available to the research community.

Life expectancy in recent years has sensibly increased and age related problems in elderly people have followed a similar trend. Being able to find innovative solutions to enable senior population to maintain their quality of life despite the presence of chronic illnesses has become crucial for high quality ageing. The opportunities offered by the technological advancement with remote assistance applications, wearable devices, and Artificial-Intelligence-Of-Things (AIoT) architectures are of paramount importance to improve the services in healthcare facilities by adding the power of Artificial Intelligence to Internet-Of-Things devices. An experimental framework has been deployed to two residential homes in collaboration with two italian companies to collect and analyze data in order to actively monitor the vital signs of their guests, predict critical situations and identify significant clusters or communities.

Menopause is the permanent cessation of menstruation occurring naturally in women's aging. The most frequent symptoms associated with menopausal phases are mucosal dryness, increased weight and body fat, and changes in sleep patterns. Oral symptoms in menopause derived from saliva flow reduction can lead to dry mouth, ulcers, and alterations of taste and swallowing patterns. However, the oral health phenotype of postmenopausal women has not been characterized. The aim of the study was to determine postmenopausal women's oral phenotype, including medical history, lifestyle, and oral assessment through artificial intelligence algorithms. We enrolled 100 postmenopausal women attending the Dental School of the University of Seville were included in the study. We collected an extensive questionnaire, including lifestyle, medication, and medical history. We used an unsupervised k-means algorithm to cluster the data following standard features for data analysis. Our results showed the main oral symptoms in our postmenopausal cohort were reduced salivary flow and periodontal disease. Relying on the classical assessment of the collected data, we might have a biased evaluation of postmenopausal women. Then, we used artificial intelligence analysis to evaluate our data obtaining the main features and providing a reduced feature defining the oral health phenotype. We found 6 clusters with similar features, including medication affecting salivation or smoking as essential features to obtain different phenotypes. Thus, we could obtain main features considering differential oral health phenotypes of postmenopausal women with an integrative approach providing new tools to assess the women in the dental clinic.

With the advancement in sensor technology and the trend shift of health measurement from treatment after diagnosis to abnormalities detection long before the occurrence, the approach of turning private spaces into diagnostic spaces has gained much attention. In this work, we designed and implemented a low-cost and compact form factor module that can be deployed on the steering wheel of cars as well as most frequently touch objects at home in order to measure physiological signals from the fingertip of the subject as well as environmental parameters. We estimated the heart rate and SpO2 with the error of 2.83 bpm and 3.52%, respectively. The signal evaluation of skin temperature shows a promising output with respect to environmental recalibration. In addition, the electrodermal activity sensor followed the reference signal, appropriately which indicates the potential for further development and application in stress measurement.

Portable diagnostic tools enable the next generation of in-home sensing and biosensing devices for screening of pathologies and monitoring of physiological parameters. In this context, the reduction in cost of off-the-shelf electronics components is rapidly advancing the development of affordable and easy-to-use portable platforms. In this work we describe the development of a multisensing platform capable of simultaneously or concurrently performing three electrochemical measurements that can provide information in real time about the levels of important physiological markers, namely pH, antioxidant capacity and conductivity of biofluids. The developed device can be wirelessly powered via an inductive charging module and an onboard battery and transmits data via Bluetooth to a mobile device for rapid and simple data sharing and processing. By integrating multiple types of electrochemical sensors into an inexpensive, pocket-sized diagnostic tool, we demonstrate a low-cost approach to point-of-care health monitoring using open-source hardware that can be readily deployed in resource-limited settings. These types of inexpensive sensing modalities have the potential to provide actionable health information as initial screening markers for the onset of several systemic diseases.

Motor deficits in the lower limbs are common in people with multiple sclerosis (MS), impacting mobility and quality of life. Objective and quantitative metrics are crucial for effective identification and monitoring of motor deficits. Recent advancements in computer vision and human pose estimators allow for automatic extraction of movement information from video data, offering potential insights into human motion patterns. This exploratory study investigates the use of Gray-Code Kernels (GCKs) in characterizing gait patterns in individuals with advanced-stage MS compared to age- and sex-matched unimpaired controls. The preliminary results obtained demonstrate the promising potential of combining GCKs and pose estimators in characterizing gait patterns, warranting further investigation in this area.

Auscultation of cracks in asphalt pavements plays a fundamental role for ensuring transportation infrastructure maintenance and longevity. This paper presents a system that combines neural network algorithms to detect, classify, and segment pavement cracks in order to produce reports of the concentration of cracks in asphalt pavements. The proposed approach generates a choropleth map that allows road inspectors to quickly determine the state of the pavement and get more information on the type of cracks present on the pavement. By implementing this system, continuous auscultation of asphalt pavements becomes feasible, contributing to effective infrastructure management and maintenance practices.

This work presents the SensiTag, a novel Battery Assisted RFID Tag with multi-sensing capabilities enabled by a recently introduced versatile integrated sensor interface (the Sensiplus). The suggested tag enables a variety of sensing capabilities by using both the built-in sensors within the Sensiplus and external commercial sensors that can be connected through the interface. This study demonstrates, for the first time, the Sensiplus platform ability to conduct temperature measurements and communication using commercial RFID systems at the UHF band. The SensiTag performance was assessed through experimental testing on the prototype.

Digital twins provide virtual replicas of medical systems, reshaping the E-Health industry towards smart, intelligent, and proactive services. With most existing biosensor solutions, patient data is transmitted over Bluetooth to a gateway that in turn connects via Ethernet to a file server (or cloud) for storage and analysis. In this work, a performance twin of a biosensor is developed, generating self-similar ECG data at increased rates compared to the actual device. In addition, we design a biosensor digital twin representation in the open-source Eclipse Ditto twin management system. We evaluate the scalability of the traditional file server and the digital twin approach. Our results indicate that the file server scales slightly better than the digital twin since Ditto operates in soft real-time only for rates below 4096 pulses/s. The digital twin's energy costs from the client/server viewpoints are also much higher.

This paper presents the design of tiny deeply quantized neural networks suitable for super integration within the Time-of-Flight, low-resolution, image sensor. They were aimed to achieve ultra low complexity with adequate classification accuracy. First floating-point models were studied to process 8-bits images 2 $$\times $$ × 2 pixel resolution, down-sampled from 8 $$\times $$ × 8 images of a public dataset. Data were acquired with an off-the-shelf Time-of-Flight sensor. Next, many deeply quantized networks were designed from scratch by using QKeras quantization training-aware schema. 8 and 6 bits pixel depth were generated by the sensor. Ternary, 6, and 8 bits quantizers were used along with convolutions and dense layers. Experimental results have shown that the proposed deeply quantized models achieved in maximum an accuracy of 77.79% compared to 88.48% for the floating-point models. The model size of the latters ranged from 92,320 bytes for the largest floating-point model to 151 bytes for the tiniest 6-bits deeply quantized model.

In this summary, we introduce a new design of the Latched Ring Oscillator (LRO), true-random-number-generator (TRNG) on a 7-Series FPGA, demonstrating the portability of the architecture. The novel design, combined with a novel sampling strategy, allows to enhance performance of the previous work, speeding up the throughput of about 265 times. The proposed LRO-TRNG effectively utilizes both meta-stability, manufacturing variations and accumulated jitter as sources of entropy, resulting in excellent levels of unpredictability and randomness. The TRNG’s performance has been validated with considering NIST and AIS-31 tests. Measurement results indicate that the LRO-TRNG achieves an estimated entropy of approximately 7.99984 per byte (as per the T8 test of the AIS-31) and a throughput of 201 Mbits/s using a 450MHz clock. When compared to existing TRNGs, the LRO-TRNG surpasses most of previously published designs in terms of the throughput attaining a good trade-off among speed and FPGA resources usage.

In this paper we present a circuit simulator based on MATLAB, developed in a dedicated way for the analysis of simple circuit topologies given by the series between a memristor described by the Stanford model and a current limiter. The simulator is built taking into account critical aspects regarding the numerical integration of the Stanford model. The performed simulations demonstrate the ability of the simulator to manage the considered critical aspects, confirming the reliability of the results it provides.

We investigate the possible application of Artificial Neural Networks (ANNs) to monitor and test hardware True Random Number Generators (TRNGs). The preliminary results have been obtained developing original investigation and design tools, including the characterization of optimized statistical estimators, considering a class of TRNGs based on binary Markov Chains. Due to the large amount of data required during the design of the ANN we developed an original flexible Database Management Systems to manage and generate training and testing data samples organized in a Object Oriented Database.

In the last decades, the space sector has been the subject of significant technological improvements and investments from both government agencies and private companies, generating an increase in data rates and volumes of exchanged data. Accordingly, the security threats and the number of documented cyberattacks have grown. In order to meet the requirements of space applications, the Consultative Committee for Space Data Systems (CCSDS) has issued and maintained a series of reports and recommendations over the years, including a set of standards aimed at efficiently exploiting the communication channels. In this work, we present the implementation of an Advanced Encryption Standard – Galois/Counter Mode (AES-GCM) core on space-grade FPGAs, that is compliant with the latest CCSDS security standards and outperforms the state-of-the-art in terms of resource efficiency.

Pedestrian detection has extensive applications in computer vision, such as intelligent transportation and security surveillance. YOLOv7 is an object detection model that achieves real-time object detection by dividing the image into grids and predicting bounding boxes and classes for each grid. This study explores and optimizes pedestrian detection based on the YOLOv7 model. Firstly, a large-scale pedestrian dataset is used to train and fine-tune the model to improve detection accuracy and robustness. The YOLOv7 model demonstrates powerful pedestrian detection performance, achieving a high average precision (AP) of 0.92 and a recall rate of 0.85. Experimental results indicate that the pedestrian detection method based on YOLOv7 achieves good results in terms of accuracy and speed, and it has high practical value and application prospects.

Computer vision and motion capture have gradually developed, and the detection of moving objects has always been very important. Vibe is a simple and efficient algorithm with low computation, good real-time performance, fast speed. There will be ghosting when detecting images in the foreground, which allows people to observe the trajectory of objects, but it will also affect the image display at the next moment to a certain extent. Vibe is divided into two steps: initializing the background model and updating the background model. By adjusting the secondary sampling factor, very few sample values can cover all background samples, store a set of values for each pixel that used to be at the same location and its neighbors. And then compare the pixel values in this collection with the current pixel values, to determine if the pixel belongs to the background and adapt the model by randomly selecting which values to replace from the background model. Finally, when a pixel is found to be part of the background, its value is propagated to the background model of neighboring pixels. The results of the vibe test are not affected by the speed at which the object moves, but by light.

Public places such as parks, pedestrian lanes, and dangerous construction sites are often prohibited for vehicles to enter, but there are still some cars that cause hidden safety hazards due to driver negligence or deliberate entry. At the same time, there are problems of high cost and low efficiency in car detection and warning in public places completely relying on manpower. Therefore, in view of the security risks caused by blind car driving in some public places, this paper builds a car recognition model based on directional gradient histogram (HOG) and support vector machine (SVM) to realize the car recognition function. This model is used to judge whether there is a car entering the recognition range, and is applied to the car recognition in a specific public environment, so as to warn cars entering the forbidden area and feed back to the staff for corresponding treatment. Through the experiment and the analysis of sample data confusion matrix, the accuracy and specificity of the model are up to 92.7% and 92.6%, respectively, which proves that the model has a high recognition rate and can be well applied to the automobile recognition task.

Infrared thermography has shown great promise as a diagnostic method for health care, providing useful information on a person's physiological, and pathological state. Recently, the use of artificial intelligence combined with infrared technology has boosted the adoption of thermal imaging in various applications and has been proposed to recognize human emotion by measuring facial skin temperature. However, its application has been limited to laboratory settings due to demanding computational and hardware resources. In this scenario, this work presents the design and development of a portable system based on a low power microcontroller implementing an optimized Edge-AI solution for binary emotional state classification using minimal hardware resources. The recognition of happiness and sadness emotional states induced by audiovisual stimuli serves as a case-study for feasibility assessment. Thermal images, produced by an uncooled and low-cost thermal sensor, along with electrocardiogram, are acquired and processed with an Arm® Cortex®-M4 microcontroller. A simple, yet effective neural network has been developed, optimized, and deployed to run the emotion detection algorithm in real time. The complete system has been experimentally verified and results in terms of accuracy and hardware constraints are discussed. Specifically, by employing a dataset consisting of 60 infrared videos, an accuracy of 80% was achieved with a resource occupation of 3.4 kB of RAM and 76.4 kB of flash memory.

The monitoring of a patient's heart rate (HR) is critical in the diagnosis of diseases. In the detection of sleep disorders, it also plays an important role. Several techniques have been proposed, including using sensors to record physiological signals that are automatically examined and analysed. This work aims to evaluate using a contactless HR monitoring system based on an accelerometer sensor during sleep. For this purpose, the oscillations caused by chest movements during heart contractions are recorded by an installation mounted under the bed mattress. The processing algorithm presented in this paper filters the signals and determines the HR. As a result, an average error of about 5 bpm has been documented, i.e., the system can be considered to be used for the forecasted domain.

The perception of the amount of stress is subjective to every person, and the perception of it changes depending on many factors. One of the factors that has an impact on perceived stress is the emotional state. In this work, we compare the emotional state of 40 German driving students and present different partitions that can be advantageous for using artificial intelligence and classification. Like this, we evaluate the data quality and prepare for the specific use. The Stress Perceived Questionnaire (PSQ20) was employed to assess the level of stress experienced by individuals while participating in a driving simulation for 5 and 25 min. As a result of our analysis, we present a categorisation of various emotional states into intervals, comparing different classifications and facilitating a more straightforward implementation of artificial intelligence for classification purposes.

The electrical energy generation is one of the main pollution causes as it is still largely based on oil and coal generators. Renewable energy sources are the most mature alternative but their production is intermittent and then require energy storage systems able to store a high quantity of energy to increase their usability. The second-life batteries of electric vehicles and low-cost battery chemistries are the best candidates to compose low-cost energy storage systems for these applications. These batteries must be parallel-connected to obtain the required capacity. A novel parallelization approach based on Input Parallel/Output Serial DC/DC Partial Power converter series-connected to each battery is presented in this paper. This approach is applied to different case studies showing that controlling only the 15% of the battery power allows the DC/DC converter to control the sharing of the grid current among the batteries. The reduction of the controlled power leads to a reduction of the cost of the energy storage system.

This paper focuses on developing a Deep Reinforcement Learning (DRL)—based agent for real-time trajectory planning and tracking in a simulated parking environment, specifically low-speed maneuvers in a parking area with comb-shaped spaces and a random distribution of non-player vehicles. We rely on CARLA as a virtual driving simulator due to its realistic graphics and physics simulation capabilities, and on the Gymnasium and Stable-Baselines3 toolkits for training the agent. We show that the agent is able to achieve a success rate of 97% in reaching the target parking lot without collisions. However, integrating CARLA with DRL frameworks poses challenges, such as determining suitable environment and neural network update frequencies. Despite these issues, the results demonstrate the potential of DRL agents in developing automated driving functions.

Analyses of the presence of heavy metals are particularly important in assessing water quality. The monitoring of environmental parameters leads to the collection and analysis of a large set of data that may be used to reduce polluting actions, suggesting that Internet of Things (IoT) technology may provide an efficient contribution to the mitigation of environmental issues. The objective of this paper is to choose the best Machine Learning (ML) model, which can be loaded into a low-power microcontroller, using the sensed data obtained through the Electrochemical Impedance Spectroscopy (EIS) technique as a dataset. The real and imaginary parts of the impedance for each frequency value occurring within the range of 0.18 and 1 Hz are used as features. Five different lead concentrations are used as outputs. The features and the output constitute the dataset. Sixteen distinct scenarios were examined to train the different models in order to investigate ways to reduce the number of features. With the aim of creating a low-energy device that can conduct measurements and predict outcomes locally, the emphasis is on training a variety of models, including qualitative (classification) and quantitative (regression) approaches.

This paper presents a novel tool for generating driving scenario datasets, that are a key asset to advance research and development in automated driving and driver assistance systems. The tool relies on the MATLAB. Automated Driving Toolbox and focuses on the overtaking maneuver. It uses simulated vehicular data, without relying on camera-equipped real-world vehicles, thus providing a low-cost solution, while allowing to abstract the main action features, that are very important for the pre-training of machine learning models. The tool has been designed to target customization (in terms, e.g., of road curvature radii), in order to allow meeting specific requirements, while its interoperability (e.g., multiple-format export) supports integration with other development environments. A preliminary analysis of the first scenarios generated with the tool confirms the validity of the system under development.

Approximate computing is a well-established technique to mitigate power consumption in error-tolerant domains such as image processing and machine learning. When paired with reconfigurable hardware, it enables dynamic adaptability to each specific task with improved power-accuracy trade-offs. In this work, we present a design methodology to enhance the energy and error metrics of a signed multiplier. This novel approach reduces the approximation error by leveraging a statistic-based truncation strategy. Our multiplier features 256 dynamically configurable approximation levels and run-time selection of the result precision. Our technique improves the mean-relative error by up to 34% compared to the zero truncation mechanism. Compared with an exact design, we achieve a maximum of 60.1% power saving for a PSNR of 10.3dB on a 5 $$\,\times \,$$ × 5 Sobel filter. Moreover, we reduce the computation energy of LeNet by 31.5%, retaining 89.4% of the original accuracy on FashionMNIST.

This paper presents a cycle-accurate verification environment for the Crypto-Tile, a cryptographic accelerator integrated into the EPI General Purpose Processor. The focus of this work is to provide a robust methodology for validating the functionality and performance of the Crypto-Tile. The verification environment includes an in-depth examination of the internal architecture and operational aspects of the Crypto-Tile, allowing for accurate modelling of hardware components and emulation of Direct Memory Access (DMA) operations. Developers can leverage this environment to simulate and verify their C-Code implementations, utilizing the functions available in the Crypto-Tile library or creating custom libraries. The verification process involves using the 32-bit AXI4 interface for communication between the processor and the Crypto-Tile while emulating DMA operations to ensure accurate testing.

State of Charge estimation is very important to deliver essential information about battery charge and aging level of Li-ion batteries in Electric Vehicles. This paper applies the Deep Leaning and Machine Learning approaches comparing decision tree and Long Short-Term Memory for estimating the State of Charge. The datasets for the training and the evaluation have been generated with a Digital Twin model applying driving cycles at different ambient temperature. The proposed Digital Twin model includes non-linear phenomena.

Online testing of computer systems is crucial in contexts such as the safety-critical domain, where the software is usually made of functional code, which is the code implementing the application-specific functionalities, and non-functional code, which implements auxiliary functionalities, e.g., test routines. By periodically running a test routine it is possible to satisfy the high dependability requirements mandated by regulators, and defined in safety standards such as ISO26262, IEC61508, and CENELEC EN 5012X. Self-Test Libraries (STLs) are a form of software-based self-test, widely used in safety-related applications. The main drawback of this safety mechanism is the overhead imposed on the execution of the functional code, and reducing this overhead is a well-known challenge in research. We propose here HUSTLE, a Hardware Unit for STL Efficient execution, which can be integrated into the chip design with no modification to the CPU’s internal logic. We also propose a scheduling mechanism that allows HUSTLE to efficiently execute self-tests, by exploiting the CPU’s idle time. This is achieved by storing test code in a separate memory and sending instructions to the CPU, bypassing the Instruction Cache, thus allowing to reduce the overall execution time and the cache interference of STL, while CPU utilization increases.

Context-aware classrooms augment traditional classrooms with unprecedented sensing and communication capabilities to monitor and analyse large amounts of contextual data. Enriching classrooms with these data opens up new opportunities to enhance the learning experience in education centres. However, implementing and deploying context-aware classrooms is not straightforward. This article describes the challenges addressed by the ACTUA project, which aims to advance the development of context-aware classrooms in primary schools.

Data-driven algorithms, such as the neural network ones, seem very appealing and accurate solutions to estimate the lithium-ion battery’s State of Charge. Their accuracy is strongly related to the amount of data used in their training phase. Therefore, huge experimental campaigns are needed to effectively train the neural network used to State of Charge estimation. The main idea behind this paper is to mitigate this drawback by training the algorithm with synthetic datasets generated from simulations of a model of the battery, instead of experimentally collected data. Two instances of the same Long-Short-Term-Memory neural network architecture designed for battery State of Charge estimation are trained, one with an experimental dataset, and the other with a synthetic one. The two neural network instances are then evaluated with the same test dataset derived from experimental data and their estimation accuracies are compared. Results show that the performances of the two networks are comparable. The experimental trained neural network scored a RMSE of only $$0.3\,\%$$ 0.3 % lower than the RMSE of the synthetic trained one. These results suggest the possibility of fruitfully using a synthetic training dataset to speed up and reduce the complexity and cost of the training phase of neural network algorithm for battery state of charge estimation.

This work explores energy harvesting through kinetic energy capture from human steps. The proposed smart floor system, consisting of multiple smart tiles, offers a promising solution for energy generation and data acquisition in high foot-traffic areas, such as shopping centers. The smart tile incorporates an energy generation and storage system, along with a data acquisition and transmission system. The use of only an accelerometer for both step and energy data acquisition minimizes power impact. An edge computing approach processes acceleration data directly on the tile, transmitting essential information, such as event steps and generated energy, to the cloud via the tile WiFi connection. This information can be used for floor optimization and commercial uses based on customer tracking. Sustainability analysis indicates that for real-time monitoring the current smart tile system requires around 15540 steps per tile for 10 h for sustainability, but this can be reduced to 362 steps by implementing power-saving techniques if a real-time feature is not required. Further research can lead to practical and commercially viable applications, contributing to a greener future.

Recognition of driving scenarios is getting ever more relevant in research, especially for assessing performance of advanced driving assistance systems (ADAS) and automated driving functions. However, the complexity of traffic situations makes this task challenging. In order to improve the detection rate achieved through state-of-the-art deep learning models, we have investigated the use of the YoloP fully convolutional neural network architecture as a pre-processing step to extract high-level features for a residual 3D convolutional neural network We observed thar this approach reduces computational complexity, resulting in optimized model performance, also in terms of generalization from training on a synthetic dataset to testing in a real-world one.

An open source remote diagnostics platform for custom electronics in experimental acquisition setups has been developed, targeting nuclear and high-energy physics (HEP) applications. We aim at enabling remote access to instrumentation hardware prototypes located in radiation-controlled areas by using existent network infrastructures. The platform relies on two components: a graphical user interface (GUI) developed in GNURadio and a remote hardware bridge (HB). The GUI was designed using the GNURadio’s optimized building blocks, ensuring a fluid visualization of real-time pulse traces and energy spectrum. No third-party libraries are used, turning our solution into an easy drop-in tool for instrumentation diagnostics in HEP and radiation-related experiments. Reliable communication between the remote instrument and the GUI is accomplished using TCP/IP, carried out by the HB in case of remote instruments without network interface. A Raspberry PI Zero-W is tested as the HB to demonstrate the few computational resources required for this end.

Multi-modal battery-operated trains are gaining an increasing favor and diffusion, as a sustainable alternative to Diesel-Electric and Diesel-Hydraulic Powertrain that are conventionally used on not electrified or partially electrified lines. In this work authors examine how current improvements on battery technologies are going to increase performances autonomy and diffusion of this technology also proposing a model and a power management architecture that can be used to simulate and accelerate the development of this technology.

The deployment of deep neural networks for visual affordance segmentation on wearable robots poses may prove critical, due to some conflicting aspects of the problem. On one hand, affordance segmentation requires high-level abstraction capabilities, that typically involve large-size models. On the other hand, computing resources hosted on wearable robots prevent to run large-size models in real-time. The paper presents an analysis of the role of the segmentation head in the trade-off between generalization performance and compute cost. The obtained models outperform modern baseline solutions in well-known, real-world datasets while meeting low computing requirements.

Research in the Internet of Things (IoT) have paved the way to a new generation of applications and services that collect huge quantities of data from the field and do a significant part of the processing on the edge. This requires availability of efficient and effective methodologies and tools for a workflow spanning from the edge to the cloud. This paper presents a generic, complete workflow and relevant system architecture for field data collection and analysis with a focus on the human physical activities. The data source is given by a low-cost embedded system that can be placed on the user body to collect heterogeneous data on the performed movements. The system features a 9 DoF IMU sensor, to ensure a high level of configurability, connected to a custom board equipped with a rechargeable battery for wireless data collection. Data are transmitted via Bluetooth Low Energy (BLE) to a smartphone/tablet app, which manages the data transfer to Measurify, a cloud-based open-source framework designed for building measurement-oriented applications. Results from a preliminary functional experiment confirm the ability of the proposed end-to-end system architecture to efficiently implement the whole targeted edge-cloud workflow.

In the Internet of Medical Things (IoMT) applications, time synchronization assumes a relevant role as in every distributed system. Various protocols and algorithms have been developed for this purpose. Recently, Fractional Low-Power Synchronization Algorithm (FLSA) was proposed as a potential solution for low-power multi-board time synchronization within a Wireless Body Area Network (WBAN). FLSA is based on a fine timer resolution control thanks to a dual-modulus divider. In this paper, we aim to study the impact of the modulus control policy on time synchronization accuracy.

This Fetal electrocardiogram is a standard method to identify and diagnose fetal diseases. Therefore, effective techniques are needed to monitor fetal conditions during pregnancy and delivery. Meanwhile, obtaining the fetal electrocardiogram (FECG) signal, which contains the electrical activity of the fetal heart, has great importance, Of course, this signal is contaminated with many noises and disturbances and the most important of them is the mother's electrocardiogram signal. Inherently, the NI-ECG signal contains the maternal ECG signal, which has a larger amplitude than the fetal ECG signal. Therefore, it is not easy to detect the fetal QRS complex in order to control the condition of the fetus and prevent congenital defects. According to the above explanations, in this article, a deep learning approach based on a convolutional neural network is proposed to separate the electrocardiogram signals of the mother from the fetus without separating the mother's ECG signal. The proposed algorithm is able to reliably detect the fetal QRS complex. Also, in addition to not needing feature extraction steps, it has been able to show more suitable performance than the best methods proposed in previous research in terms of detection accuracy.

This paper presents a method for sea monitoring based on a low-cost 24 GHz off-the-shelf FMCW radar and an embedded Raspberry Pi PC. The technique relies upon the use of range-Doppler maps, which are obtained through a 2D FFT processing of the acquired scattered data. The resulting insights may offer new possibilities for monitoring and understanding sea waves, which have important applications The study's findings provide valuable insights into monitoring and understanding sea waves, which can have significant implications for predicting natural disasters and mitigating their impact based on predicted disaster scenarios.

Complex electronic systems comprise heterogeneous components operating concurrently to achieve complex goals. Electronic systems (like everything else) can fail. Hence, having resiliency in mind, components must adapt to changing circumstances to meet their goals. In this context, multi-agent systems (MAS) offer a flexible approach to managing complex problems by leveraging agents’ collective behaviour and sensory abilities. This article presents a multi-layered event-driven hierarchical MAS model, which enhances the resilience of pervasive electronic systems by adapting to events within the application domain. We report a case study with classrooms augmented for contextual sensing.

The financial sector has suffered a groundbreaking transformation with the advent of cryptocurrencies, shifting from centralised to decentralised schemes. Hardware wallets play an essential role in storing cryptocurrencies securely. However, these electronic devices generally have limited resources that open the door to attacks. In this article, we describe three attacks against them. Several wallets with funds or recent transactions have been discovered with these attacks.

This paper investigates the design, implementation, and assessment of different embedded neural network models for hand motion classification. Using an Inertial Measurement Unit (IMU) sensor, a dataset of five distinct hand motion classes has been collected from the Arduino nano BLE 33 targeting hand motion analysis. Two different machine learning models namely Convolutional Neural Network (CNN) and Multilayer Perceptron (MLP) are implemented and compared in terms of classification accuracy and hardware complexity. Experimental results show identical classification accuracy of 100% for the two models. After deployment, results show that the MLP network exhibited the lowest processing time and RAM usage, taking 4 ms and 4.0 K, respectively. In contrast, the CNN model needs 313ms and utilizes 15.9 K of RAM. Moreover, the MLP model also had the lowest flash usage with 25.3 K, while the CNN model used 67.7 K.

Bowel preparation is a crucial step in ensuring the success and accuracy of colonoscopy procedures. Adequate bowel cleansing allows for better visualization and detection of abnormalities within the colon. In this study, we present an AI tool developed to assess the quality of bowel preparation in colonoscopy procedures. The dataset used in this study consists of 350 images of toilet bowls obtained from patients at the hospital “Hôtel Dieu de France” in Beirut, Lebanon. Their images are labeled by the professionals using the Boston scores. Our methodology involves a comprehensive pre-processing phase, encompassing detection, cropping, color adjustment, and Principal Component Analysis (PCA) on the image dataset. Subsequently, we applied different machine learning (ML) models for classification, achieving a high accuracy of 92% with Gradient Boosting. This AI-based approach exhibits great potential in enhancing the efficiency and reliability of colonoscopy evaluations, ultimately leading to improved patient outcomes and early detection of gastrointestinal disorders.

Common bearing failure modes (wear, contamination, corrosion, overload, misalignment, etc.) have unique characteristics, requiring diverse identification and mitigation strategies. No single definition can encompass all contributing factors. Understanding these complexities is crucial for safety critical system to implement fault detection. Machine learning offers a data-driven and intelligent approach to fault detection to improve safety, efficiency, and cost-effectiveness. In this work we propose a supervised machine learning model using naïve bayes classifier for safety critical system fault detection using time domain vibration features. Isolation forest-based anomaly detection is used for labeling faults or healthy condition. The proposed model is tested on the PRONOSTIA dataset, and the model detects fault before their failure criteria in all eleven experiments. The models hold promise for early fault detection in safety-critical systems.

The use of brain computer interfaces (BCIs) is crucial to the development of neural prosthetics and neuroscience. BCIs assist paralyzed patients by enabling them to control computers or robots using their neural activity. Optically transparent BCIs, combined with optical imaging modalities, enable simultaneous gathering of high-resolution electrophysiological signals and imaging of neural activities. In this work, we report the investigation of the mechanical behavior of an ultra-thin glass-based brain machine interface probe when subjected to various types of forces encountered during insertion and engagement into the brain tissues. Results of probe thickness optimization show that electrodes can be 25 µm thick while maintaining a factor of safety against failure > 1.