Imec the Netherlands

Nanoparticles are the focus of much attention due to their astonishing properties and numerous possibilities for applications in nanotechnology. For realising versatile functions, assembly of nanoparticles in regular patterns on surfaces and at interfaces is required. Assembling nanoparticles generates new nanostructures, which have unforeseen collective, intrinsic physical properties. These properties can be exploited for multipurpose applications in nanoelectronics, spintronics, sensors, etc. This review surveys different techniques, currently employed and being developed, for assembling nanoparticles in to ordered nanostructures. In this endeavour, the principles and methods involved in the development of assemblies are discussed. Subsequently, different possibilities of nanoparticle-based nanostructures, obtained in multi-dimensions, are presented.

This paper presents an asynchronous SAR ADC for flexible, low energy radios. To achieve excellent power efficiency for a relatively moderate resolution, various techniques are introduced to reduce the power consumption: custom-designed 0.5 fF unit capacitors minimize the analog power consumption while asynchronous dynamic logic minimizes the digital power consumption. The variability of the custom-designed capacitors is estimated by a specialized CAD tool and verified by chip measurements. An implemented 8-bit prototype in a 90 nm CMOS technology occupies 228 μm × 240 μm including decoupling capacitors, and achieves an ENOB of 7.77 bit at a sampling frequency of 10.24 MS/s. The power consumption equals 26.3 μW from a 1 V supply, thus resulting in an energy efficiency of 12 fJ/conversion-step. Moreover, the fully dynamic design, which is optimized for low-leakage, leads to a standby power consumption of 6 nW. In that way, the energy efficiency of this converter can be maintained down to very low sampling rates.

In this paper, a design method for the co-design and integration of a CMOS rectifier and small loop antenna is described. In order to improve the sensitivity, the antenna-rectifier interface is analyzed as it plays a crucial role in the co-design optimization. Subsequently, a 5-stage cross-connected differential rectifier with a 7-bit binary-weighted capacitor bank is designed and fabricated in standard 90 nm CMOS technology. The rectifier is brought at resonance with a high-Q loop antenna by means of a control loop that compensates for any variation at the antenna-rectifier interface and passively boosts the antenna voltage to enhance the sensitivity. A complementary MOS diode is proposed to improve the harvester's ability to store and hold energy over a long period of time during which there is insufficient power for rectification. The chip is ESD protected and integrated on a compact loop antenna. Measurements in an anechoic chamber at 868 MHz demonstrate a -27 dBm sensitivity for 1 V output across a capacitive load and 27 meter range for a 1.78 W RF source in an office corridor. The end-to-end power conversion efficiency equals 40% at -17 dBm.

Convolutional neural networks (CNNs) have become the dominant neural network architecture for solving many state-of-the-art (SOA) visual processing tasks. Even though graphical processing units are most often used in training and deploying CNNs, their power efficiency is less than 10 GOp/s/W for single-frame runtime inference. We propose a flexible and efficient CNN accelerator architecture called NullHop that implements SOA CNNs useful for low-power and low-latency application scenarios. NullHop exploits the sparsity of neuron activations in CNNs to accelerate the computation and reduce memory requirements. The flexible architecture allows high utilization of available computing resources across kernel sizes ranging from 1×1 to 7×7. NullHop can process up to 128 input and 128 output feature maps per layer in a single pass. We implemented the proposed architecture on a Xilinx Zynq field-programmable gate array (FPGA) platform and presented the results showing how our implementation reduces external memory transfers and compute time in five different CNNs ranging from small ones up to the widely known large VGG16 and VGG19 CNNs. Postsynthesis simulations using Mentor Modelsim in a 28-nm process with a clock frequency of 500 MHz show that the VGG19 network achieves over 450 GOp/s. By exploiting sparsity, NullHop achieves an efficiency of 368%, maintains over 98% utilization of the multiply-accumulate units, and achieves a power efficiency of over 3 TOp/s/W in a core area of 6.3 mm2. As further proof of NullHop's usability, we interfaced its FPGA implementation with a neuromorphic event camera for real-time interactive demonstrations.

As the dimensions of conductors shrink into the nanoscale, their electrical conductivity becomes dependent on their size even at room temperature. Although the behavior varies dramatically as temperatures increase from nanokelvins to hundreds of kelvins, the effect is generally to increase the resistivity above that of bulk material. As such, the underlying size-dependent phenomena have become increasingly important as advanced technologies have shifted their focus first from macro- to microscale and more recently from micro- to nanoscale dimensions. Indeed, the size-dependent increase of electrical resistivity that results from electron scattering on external and internal surfaces of copper conductors has already become technology limiting in modern microelectronics. This article summarizes the phenomena that underlie size effects, focusing on conduction in copper lines in particular. Attention is given to describing key innovations in both theoretical and experimental assessments that have significantly modified, facilitated, or advanced understanding.

Inspired by detailed modelling of biological neurons, spiking neural networks (SNNs) are investigated as biologically plausible and high-performance models of neural computation. The sparse and binary communication between spiking neurons potentially enables powerful and energy-efficient neural networks. The performance of SNNs, however, has remained lacking compared with artificial neural networks. Here we demonstrate how an activity-regularizing surrogate gradient combined with recurrent networks of tunable and adaptive spiking neurons yields the state of the art for SNNs on challenging benchmarks in the time domain, such as speech and gesture recognition. This also exceeds the performance of standard classical recurrent neural networks and approaches that of the best modern artificial neural networks. As these SNNs exhibit sparse spiking, we show that they are theoretically one to three orders of magnitude more computationally efficient compared to recurrent neural networks with similar performance. Together, this positions SNNs as an attractive solution for AI hardware implementations. The use of sparse signals in spiking neural networks, modelled on biological neurons, offers in principle a highly efficient approach for artificial neural networks when implemented on neuromorphic hardware, but new training approaches are needed to improve performance. Using a new type of activity-regularizing surrogate gradient for backpropagation combined with recurrent networks of tunable and adaptive spiking neurons, state-of-the-art performance for spiking neural networks is demonstrated on benchmarks in the time domain.

Physiological signals have shown to be reliable indicators of stress in laboratory studies, yet large-scale ambulatory validation is lacking. We present a large-scale cross-sectional study for ambulatory stress detection, consisting of 1002 subjects, containing subjects' demographics, baseline psychological information, and five consecutive days of free-living physiological and contextual measurements, collected through wearable devices and smartphones. This dataset represents a healthy population, showing associations between wearable physiological signals and self-reported daily-life stress. Using a data-driven approach, we identified digital phenotypes characterized by self-reported poor health indicators and high depression, anxiety and stress scores that are associated with blunted physiological responses to stress. These results emphasize the need for large-scale collections of multi-sensor data, to build personalized stress models for precision medicine.

Conventional gel electrodes are widely used for biopotential measurements, despite important drawbacks such as skin irritation, long set-up time and uncomfortable removal. Recently introduced dry electrodes with rigid metal pins overcome most of these problems; however, their rigidity causes discomfort and pain. This paper presents dry electrodes offering high user comfort, since they are fabricated from EPDM rubber containing various additives for optimum conductivity, flexibility and ease of fabrication. The electrode impedance is measured on phantoms and human skin. After optimization of the polymer composition, the skin-electrode impedance is only ~10 times larger than that of gel electrodes. Therefore, these electrodes are directly capable of recording strong biopotential signals such as ECG while for low-amplitude signals such as EEG, the electrodes need to be coupled with an active circuit. EEG recordings using active polymer electrodes connected to a clinical EEG system show very promising results: alpha waves can be clearly observed when subjects close their eyes, and correlation and coherence analyses reveal high similarity between dry and gel electrode signals. Moreover, all subjects reported that our polymer electrodes did not cause discomfort. Hence, the polymer-based dry electrodes are promising alternatives to either rigid dry electrodes or conventional gel electrodes.

In this paper, we present our vision of what kind of wearable devices and how they can be powered by the heat of human beings and by using ambient light. The basic principles of designing body-powered devices and ways of their hybridizing with photovoltaic cells are discussed. The mechanisms of thermoregulation in humans and the laws of thermodynamics enable placing a distinct boarder between realistic targets and the science fiction. These allow prediction of application areas for wearable energy harvesters accounting for competitive batteries with long service life. The existing family of body-powered wearable devices and new technologies for thermopiles are discussed. The theory and practice point at the necessity of using microelectronic and microelectromechanical system technologies for the target application area. These technologies for thermopiles offer the possibility of reduced production cost. Therefore, autonomous systems powered thermoelectrically could be successfully marketed. The related aspects of design and fabrication are discussed.

Previous works on sequential learning address the problem of forgetting in discriminative models. In this paper we consider the case of generative models. In particular, we investigate generative adversarial networks (GANs) in the task of learning new categories in a sequential fashion. We first show that sequential fine tuning renders the network unable to properly generate images from previous categories (i.e. forgetting). Addressing this problem, we propose Memory Replay GANs (MeRGANs), a conditional GAN framework that integrates a memory replay generator. We study two methods to prevent forgetting by leveraging these replays, namely joint training with replay and replay alignment. Qualitative and quantitative experimental results in MNIST, SVHN and LSUN datasets show that our memory replay approach can generate competitive images while significantly mitigating the forgetting of previous categories.

A wireless sensor network is proposed for monitoring buildings to assess earthquake damage. The sensor nodes use custom-developed capacitive microelectromechanical systems strain and 3-D acceleration sensors and a low power readout application-specified integrated circuit for a battery life of up to 12 years. The strain sensors are mounted at the base of the building to measure the settlement and plastic hinge activation of the building after an earthquake. They measure periodically or on-demand from the base station. The accelerometers are mounted at every floor of the building to measure the seismic response of the building during an earthquake. They record during an earthquake event using a combination of the local acceleration data and remote triggering from the base station based on the acceleration data from multiple sensors across the building. A low power network architecture was implemented over an 802.15.4 MAC in the 900-MHz band. A custom patch antenna was designed in this frequency band to obtain robust links in real-world conditions. The modules have been validated in a full-scale laboratory setup with simulated earthquakes.

Early mental stress detection can prevent many stress related health problems. This study aimed at using a wearable sensor system to measure physiological signals and detect mental stress. Three different stress conditions were presented to a healthy subject group. During the procedure, ECG, respiration, skin conductance, and EMG of the trapezius muscles were recorded. In total, 19 physiological features were calculated from these signals. After normalization of the feature values and analysis of correlations among these features, a subset of 9 features was selected for further analysis. Principal component analysis reduced these 9 features to 7 principal components (PCs). Using these PCs and different classifiers, a consistent classification accuracy between stress and non stress conditions of almost 80% was found. This suggests that a promising feature subset was found for future development of a personalized stress monitor.

The increasing popularity of ultra-wideband (UWB) technology for location-based services, such as access control and real-time indoor track&tracing, as well as UWB support in new consumer devices such as smartphones, has resulted in the availability of multiple new UWB radio chips. However, due to this increase in UWB device availability, the question of which (industry) standards and configuration factors impact UWB interoperability and compatibility becomes increasingly important. In this paper, the fundamentals of UWB compatibility are investigated by first giving an overview of different UWB radio chips on the market. After that, an overview of UWB standards and organizations is given. Next, this overview is used to discuss the focus of these different standards and to identify the differences between them. We describe compatibility issues and associated interoperability aspects related to physical (PHY), medium-access-control (MAC) and upper layers. For the PHY layer, compatibility is possible for all UWB radio chips if the correct settings are configured. For the MAC layer, the implementation of the multiple access scheme as well as the localization technique is mostly proprietary. For the device discovery, several standards are currently being drafted. Finally, future challenges related to UWB interoperability are discussed.

Neuromorphic hardware implements biological neurons and synapses to execute a spiking neural network (SNN)-based machine learning. We present SpiNeMap, a design methodology to map SNNs to crossbar-based neuromorphic hardware, minimizing spike latency and energy consumption. SpiNeMap operates in two steps: SpiNeCluster and SpiNePlacer. SpiNeCluster is a heuristic-based clustering technique to partition an SNN into clusters of synapses, where intracluster local synapses are mapped within crossbars of the hardware and intercluster global synapses are mapped to the shared interconnect. SpiNeCluster minimizes the number of spikes on global synapses, which reduces spike congestion and improves application performance. SpiNePlacer then finds the best placement of local and global synapses on the hardware using a metaheuristic-based approach to minimize energy consumption and spike latency. We evaluate SpiNeMap using synthetic and realistic SNNs on a state-of-the-art neuromorphic hardware. We show that SpiNeMap reduces average energy consumption by 45% and spike latency by 21%, compared to the best-performing SNN mapping technique.

A brief comprehensive overview is provided of the elements constituting the burden of kidney disease [chronic kidney disease (CKD) and acute kidney injury]. This publication can be used for advocacy, emphasizing the importance and urgency of reducing this heavy and rapidly growing burden. Kidney diseases contribute to significant physical limitations, loss of quality of life, emotional and cognitive disorders, social isolation and premature death. CKD affects close to 100 million Europeans, with 300 million being at risk, and is projected to become the fifth cause of worldwide death by 2040. Kidney disease also imposes financial burdens, given the costs of accessing healthcare and inability to work. The extrapolated annual cost of all CKD is at least as high as that for cancer or diabetes. In addition, dialysis treatment of kidney diseases imposes environmental burdens by necessitating high energy and water consumption and producing plastic waste. Acute kidney injury is associated with further increases in global morbidity, mortality and economic burden. Yet investment in research for treatment of kidney disease lags behind that of other diseases. This publication is a call for European investment in research for kidney health. The innovations generated should mirror the successful European Union actions against cancer over the last 30 years. It is also a plea to nephrology professionals, patients and their families, caregivers and kidney health advocacy organizations to draw, during the Decade of the Kidney (2020-30), the attention of authorities to realize changes in understanding, research and treatment of kidney disease.

Sleep apnea is one of the most common sleep disorders and the consequences of undiagnosed sleep apnea can be very severe, ranging from increased blood pressure to heart failure. However, many people are often unaware of their condition. The gold standard for diagnosing sleep apnea is an overnight polysomnography in a dedicated sleep laboratory. Yet, these tests are expensive and beds are limited as trained staff needs to analyze the entire recording. An automated detection method would allow a faster diagnosis and more patients to be analyzed. Most algorithms for automated sleep apnea detection use a set of human-engineered features, potentially missing important sleep apnea markers. In this paper, we present an algorithm based on state-of-the-art deep learning models for automatically extracting features and detecting sleep apnea events in respiratory signals. The algorithm is evaluated on the Sleep-Heart-Health-Study-1 dataset and provides per-epoch sensitivity and specificity scores comparable to the state of the art. Furthermore, when these predictions are mapped to the apnea-hypopnea index, a considerable improvement in per-patient scoring is achieved over conventional methods. This paper presents a powerful aid for trained staff to quickly diagnose sleep apnea.

This paper reviews the state of the arts and trends of the AI-Based biomedical processing algorithms and hardware. The algorithms and hardware for different biomedical applications such as ECG, EEG and hearing aid have been reviewed and discussed. For algorithm design, various widely used biomedical signal classification algorithms have been discussed including support vector machine (SVM), back propagation neural network (BPNN), convolutional neural networks (CNN), probabilistic neural networks (PNN), recurrent neural networks (RNN), Short-term Memory Network (LSTM), fuzzy neural network and etc. The pros and cons of the classification algorithms have been analyzed and compared in the context of application scenarios. The research trends of AI-Based biomedical processing algorithms and applications are also discussed. For hardware design, various AI-Based biomedical processors have been reviewed and discussed, including ECG classification processor, EEG classification processor, EMG classification processor and hearing aid processor. Various techniques on architecture and circuit level have been analyzed and compared. The research trends of the AI-Based biomedical processor have also been discussed.

An envelope detector-based wake-up receiver front-end is presented. A double-sampling technique is implemented to suppress the 1/f noise and DC offset, resulting in a lower output noise level and therefore a better SNR for a given input level. The receiver consumes 51 μW and occupies 0.36 mm2in 90 nm CMOS. For 10 kb/s 00K reception it achieves -69 dBm and -80 dBm sensitivity at 2.4 GHz and 915 MHz respectively.

This paper describes an 8-channel gel-free EEG/electrode-tissue impedance (ETI) acquisition system, consisting of nine active electrodes (AEs) and one back-end (BE) analog signal processor. The AEs amplify the weak EEG signals, while their low output impedance suppresses cable-motion artifacts and 50/60 Hz mains interference. A common-mode feed-forward (CMFF) scheme boosts the CMRR of the AE pairs by 25 dB. The BE post-processes and digitizes the analog outputs of the AEs, it also can configure them via a single-wire pulse width modulation (PWM) protocol. Together, the AEs and BE are capable of recording 8-channel EEG and ETI signals. With EEG recording enabled, ETIs of up to 60 kΩ can be measured, which increases to 550 kΩ when EEG recording is disabled. Each EEG channel has a 1.2 GΩ input impedance (at 20 Hz), 1.75 µVrms (0.5–100 Hz) input-referred noise, 84 dB CMRR and ±250 mV electrode offset rejection capability. The EEG acquisition system was implemented in a standard 0.18 µm CMOS process, and dissipates less than 700 µW from a 1.8 V supply.

Nanowire-based devices show great promise for next generation (bio)chemical sensors as evidenced by the large volume of high-quality publications. Here, a nanoscale gas sensing device is presented, based on gold-free grown vertical InAs nanowire arrays. The nanowires are contacted Ohmically in their as-grown locations using an air bridge construction, leaving the nanowire surface free for gas adsorption. Noise measurements were performed to determine the measurement resolution for gas detection. These devices are sensitive to NO(2) concentrations well below 100 ppb at room temperature. NO(2) exposure leads to both a reduction in carrier density and electron mobility.