Journal of Low Power Electronics and Applications

The building block computation system is constructed by stacking various chips three-dimensionally. The stacked chips incorporate the same TCI IP (Through Chip Interface Intellectual Property) but cannot provide identical characteristics, requiring adjustments in power supply and bias voltage. However, providing characteristics measurement hardware for all chips is difficult due to the limitation of chip area or pin numbers. To address this problem, we developed TCI Tester, a small chip to measure electric characteristics by stacking on TCI of every chip. By stacking two TCI Tester chips, it appears that the up-directional data transfer has a stricter condition than down directional one on power supply voltage and operational frequency. Also, the transfer performance is poorer than designed. Similar measurement results are obtained by stacking TCI Tester on other chips with TCI IP. To investigate the reason, we analyzed the power grid resistance of various chips with the TCI IP. Results also showed that the chips with higher resistance have a narrow operational condition and poorer performance. The results suggest that the power grid design is important for keeping the performance through the TCI channel. Full article

A programmable, energy-efficient analog hardware implementation of a multilayer perceptron (MLP) is presented featuring a highly programmable system that offers the user the capability to create an MLP neural network hardware design within the available framework. In addition to programmability, this implementation provides energy-efficient operation via analog/mixed-signal design. The configurable system is made up of 12 neurons and is fabricated in a standard 130 nm CMOS process occupying approximately 1 mm² of on-chip area. The system architecture is analyzed in several different configurations with each achieving a power efficiency of greater than 1 tera-operations per watt. This work offers an energy-efficient and scalable alternative to digital configurable neural networks that can be built upon to create larger networks capable of standard machine learning applications, such as image and text classification. This research details a programmable hardware implementation of an MLP that achieves a peak power efficiency of 5.23 tera-operations per watt while consuming considerably less power than comparable digital and analog designs. This paper describes circuit elements that can readily be scaled up at the system level to create a larger neural network architecture capable of improved energy efficiency. Full article

Light Fidelity (LiFi) technology has gained attention and is growing rapidly today. Utilizing light as a propagation medium allows LiFi to promise a wider bandwidth than existing Wireless Fidelity (WiFi) technology and enables the implementation of cellular technology to improve bandwidth utilization. In addition, LiFi is very attractive because it can utilize lighting facilities consisting of light-emitting diodes (LEDs). A LiFi system that uses intensity modulation and direct detection requires the signal of orthogonal frequency division multiplexing (OFDM) to have a real and non-negative value; therefore, certain adjustments must be made. The proposed methods for generating unipolar signals vary from adding a direct current, clipping the signal, superposing several unipolar signals, and hybrid methods as in DC-biased optical (DCO)-OFDM, asymmetrically clipped optical (ACO)-OFDM, layered ACO (LACO)-OFDM, and asymmetrically clipped DC-biased optical (ADO)-OFDM, respectively. In this paper, we review and compare various modulation techniques to support the implementation of LiFi systems using commercial LEDs. The main objective is to obtain a modulation technique with good energy efficiency, efficient spectrum utilization, and low computational complexity so that it is easy for us to apply it in experiments on a laboratory scale. Full article

Structures for the evaluation of fast Fourier transforms are important components in several signal-processing applications and communication systems. Their capabilities play a key role in the performance enhancement of the whole system in which they are embedded. In this paper, a novel implementation of the discrete Fourier transform is proposed, based on a bit-slice approach and on the exploitation of the input sequence finite word length. Input samples of the sequence to be transformed are split into binary sequences and each one is Fourier transformed using only complex sums. An FPGA-based solution characterized by low latency and low power consumption is designed. Simulations have been carried out, first in the Matlab environment, then emulated in Quartus IDE with Intel. The hardware implementation of the conceived system and the test for the functional accuracy verification have been performed, adopting the DE2-115 development board from Terasic, which is equipped with the Cyclone IV EP4CE115F29C7 FPGA by Intel. Full article

New mission-critical applications, such as autonomous vehicles and life-support systems, set a high bar for the reliability of modern microprocessors that operate in highly challenging conditions. However, while cutting-edge integrated circuit (IC) technologies have intensified microprocessors by providing remarkable reductions in the silicon area and power consumption, they also introduce new reliability challenges through the complex design rules they impose, creating a significant hurdle in the design process. In this paper, we focus on electromigration (EM), which is a crucial factor impacting IC reliability. EM refers to the degradation process of IC metal nets when used for both power supply and interconnecting signals. Typically, EM concerns have been addressed at the backend, circuit, and layout levels, where EM rules are enforced assuming extreme conditions to identify and resolve violations. This study presents new techniques that leverage architectural features to mitigate the effect of EM on the memory hierarchy of modern microprocessors. Architectural approaches can reduce the complexity of solving EM-related violations, and they can also complement and enhance common existing methods. In this study, we present a comprehensive simulation analysis that demonstrates how the proposed solution can significantly extend the lifetime of a microprocessor’s memory hierarchy with minimal overhead in terms of performance, power, and area while relaxing EM design efforts. Full article

This paper presents a multiplexer-based extended range multi-modulus divider (ER-MMD) technique for multi-band phase locked loop (PLL). The architecture maintains a modular structure by using conventional $2/3$ divider cells and a multiplexer without adding any extra logic circuitry. The area and power overhead is minimal. The $2/3$ divider cells are designed using true single phase clock (TSPC) logic for ER-MMD to operate in the sub-10 GHz range. A division range of 2 to 511 is achieved using this logic. The ER-MMD operates at a maximum frequency of 6 GHz with a worst-case current of 625 $\mathsf{\mu }$A when powered with a 1 V supply. A dual voltage controlled oscillator (VCO), $L5$/S band PLL for Indian Regional Navigation Satellite System (IRNSS) application is designed, which incorporates an ER-MMD based on the proposed approach as a proof of concept. This technique achieves the best power efficiency of 12 GHz/mW, among the state-of-the-art ER-MMD designs. Full article

Semantic segmentation is a machine learning task that is seeing increased utilization in multiple fields, from medical imagery to land demarcation and autonomous vehicles. A real-time autonomous system must be lightweight while maintaining reasonable accuracy. This research focuses on leveraging the fusion of long-wave infrared (LWIR) imagery with visual spectrum imagery to fill in the inherent performance gaps when using visual imagery alone. This approach culminated in the Fast Thermal Fusion Network (FTFNet), which shows marked improvement over the baseline architecture of the Multispectral Fusion Network (MFNet) while maintaining a low footprint. Full article

A contactless position sensor based on an array of magnetically coupled resonators and an external single coil cell is discussed for both stationary and dynamic applications. The simple structure allows the sensor to be adapted to the system in which it is installed and can be used to detect the positions of objects in motion that bear an external resonator coil that does not necessitate a supply. By exploiting the unique behaviour of the array input impedance, it is possible to identify the position of the external resonator by exciting the first array cell with an external voltage source and measuring the resulting input current. The system is robust and suitable for application in harsh environments. The sensitivity of the measured input impedance to the space variation is adjustable with the definition of the array geometry and is analysed. Different configurations of the array and external resonator are considered, and the effects of various termination conditions and the resulting factor of merit after changing the coil resistance are discussed. The proposed procedure is numerically validated for an array of ten identical magnetically coupled resonators with 15 cm side lengths. Simulations carried out for a distance of up to 20 cm show that, with a quality factor lower than 100 and optimal terminations of both the array and external coil, it is possible to detect the position of the latter. Full article

Convolutional Neural Networks (CNNs) have been incredibly effective for object detection tasks. YOLOv4 is a state-of-the-art object detection algorithm designed for embedded systems. It is based on YOLOv3 and has improved accuracy, speed, and robustness. However, deploying CNNs on embedded systems such as Field Programmable Gate Arrays (FPGAs) is difficult due to their limited resources. To address this issue, FPGA-based CNN architectures have been developed to improve the resource utilization of CNNs, resulting in improved accuracy and speed. This paper examines the use of General Matrix Multiplication Operations (GEMM) to accelerate the execution of YOLOv4 on embedded systems. It reviews the most recent GEMM implementations and evaluates their accuracy and robustness. It also discusses the challenges of deploying YOLOv4 on autonomous vehicle datasets. Finally, the paper presents a case study demonstrating the successful implementation of YOLOv4 on an Intel Arria 10 embedded system using GEMM. Full article

Tuberculosis (TB) has long been recognized as a significant health concern worldwide. Recent advancements in noninvasive wearable devices and machine learning (ML) techniques have enabled rapid and cost-effective testing for the real-time detection of TB. However, small datasets are often encountered in biomedical and chemical engineering domains, which can hinder the success of ML models and result in overfitting issues. To address this challenge, we propose various data preprocessing methods and ML approaches, including long short-term memory (LSTM), convolutional neural network (CNN), Gramian angular field-CNN (GAF-CNN), and multivariate time series with MinCutPool (MT-MinCutPool), for classifying a small TB dataset consisting of multivariate time series (MTS) sensor signals. Our proposed methods are compared with state-of-the-art models commonly used in MTS classification (MTSC) tasks. We find that lightweight models are more appropriate for small-dataset problems. Our experimental results demonstrate that the average performance of our proposed models outperformed the baseline methods in all aspects. Specifically, the GAF-CNN model achieved the highest accuracy of 0.639 and the highest specificity of 0.777, indicating its superior effectiveness for MTSC tasks. Furthermore, our proposed MT-MinCutPool model surpassed the baseline MTPool model in all evaluation metrics, demonstrating its viability for MTSC tasks. Full article

This paper presents a capacitively coupled chopper instrumentation amplifier (CCIA) with ultra-low power consumption and programmable bandwidth for biomedical applications. To achieve a flexible bandwidth from 0.2 to 10 kHz without additional power consumption, a programmable Miller compensation technique was proposed and used in the CCIA. By using a Squeezed inverter amplifier (SQI) that employs a 0.2-V supply, the proposed CCIA addresses the primary noise source in the first stage, resulting in high noise power efficiency. The proposed CCIA is designed using a 0.18 µm CMOS technology process and has a chip area of 0.083 mm². With a power consumption of 0.47 µW at 0.2 and 0.8 V supply, the proposed amplifier architecture achieves a thermal noise of 28 nV/√Hz, an input-related noise (IRN) of 0.9 µV_rms, a closed-loop gain (A_V) of 40 dB, a power supply rejection ratio (PSRR) of 87.6 dB, and a common-mode rejection ratio (CMRR) of 117.7 dB according to post-simulation data. The proposed CCIA achieves a noise efficiency factor (NEF) of 1.47 and a power efficiency factor (PEF) of 0.56, which allows comparison with the latest research results. Full article

The dark-silicon challenge poses a design problem for future many-core systems. As a result of this, several techniques have been introduced to improve the number of processing elements that can be powered on. One of the techniques employed by many is Task Migration. In this paper, an Ageing Task Migration Aware for High-Performance Computing (AMA) is proposed to improve the lifetime of nodes. The proposed method determines which clusters applications are mapped to and migrates high-demand tasks amongst nodes to improve the lifetime at every epoch. Experimental results show that the proposed method outperforms state-of-the-art techniques by more than 10%. Full article

This work details the design and experimental characterization of a 2D rectenna for scavenging radio frequency energy at 2.45 GHz (WiFi band), fabricated on polylactic acid polymer (PLA) using a plastronics approach. PLA is the RF substrate of both antenna and rectifier. The two transmission line (TTL) approach is used to characterize the substrate properties to be considered during design. A linearly polarized patch antenna with microstrip transmission feeding is connected to a single series diode rectifier through a T-matching network. The antenna has simulated and measured gain of 7.6 dB and 7.5 dB, respectively. The rectifier has a measured DC output power of 0.96 $\mathsf{\mu }$W at an optimal load of 2 k$\Omega$ under RF input power of −20 dBm at 2.45 GHz. The power conversion efficiency is 9.6% in the latter conditions for a 54 × 36 mm patch antenna of a 1.5 mm thick PLA substrate obtained from additive manufacturing. The power conversion efficiency reaches a value of 28.75% when the input power is −10 dBm at 2.45 GHz. This corresponds to a peak DC power of 28.75 $\mathsf{\mu }$W when the optimal load is 1.5 k$\Omega$. The results compare significantly with the ones of a similar rectenna circuit manufactured on preferred RF substrate. Full article

In this paper, a novel dynamic body-driven ultra-low voltage (ULV) comparator is presented. The proposed topology takes advantage of the back-gate configuration by driving the input transistors’ gates with a clocked positive feedback loop made of two AND gates. This allows for the removal of the clocked tail generator, which decreases the number of stacked transistors and improves performance at low $V_{DD}$. Furthermore, the clocked feedback loop causes the comparator to behave as a full CMOS latch during the regeneration phase, which means no static power consumption occurs after the outputs have settled. Thanks to body driving, the proposed comparator also achieves rail-to-rail input common mode range (ICMR), which is a critical feature for circuits that operate at low and ultra-low voltage headrooms. The comparator was designed and optimized in a 130-nm technology from STMicroelectronics at $V_{DD}=0.3$ V and is able to operate at up to 2 MHz with an input differential voltage of 1 mV. The simulations show that the comparator remains fully operational even when the supply voltage is scaled down to 0.15 V, in which case the circuit exhibits a maximum operating frequency of 80 kHz at $V_{id}=1$ mV. Full article

The shrinking of technology nodes allows higher performance, but susceptibility to soft errors increases. The protection has been implemented mainly by lockstep or hardened process techniques, which results in a lower frequency, a larger area, and higher power consumption. We propose a protection technique that only slightly affects the maximal frequency. The area and power consumption increase are comparable with dual lockstep architectures. A reaction to faults and the ability to recover from them is similar to triple modular redundancy architectures. The novelty lies in applying redundancy into the processor’s pipeline and its separation into two sections. The protection provides fast detection of faults, simple recovery by a flush of the pipeline, and allows a large prediction unit to be unprotected. A proactive component automatically scrubs a register file to prevent fault accumulation. The whole protection scheme can be fully implemented at the register transfer level. We present the protection scheme implemented inside the RISC-V core with the RV32IMC instruction set. Simulations confirm that the protection can handle the injected faults. Synthesis shows that the protection lowers the maximum frequency by only about 3.9%. The area increased by 108% and power consumption by 119%. Full article

In this work, a first-order low-pass filter is proposed as suitable for time-mode PWM signal processing. In time-mode PWM signal processing, the pulse width of a rectangular pulse is the processing variable. The filter is constructed using basic time-mode building blocks such as time registers and time adders and so it is characterized by low complexity which can lead to the modular and versatile design of higher-order filters. All the building blocks of the filter were designed and verified in a TSMC 65 nm technology process. The sampling frequency was 5 MHz, the gain of the filter at low frequencies was at −0.016 dB, the cut-off frequency was 1.2323 MHz, and the power consumption was around 59.1 μW. Full article

Drinking water is becoming increasingly scarce as the world’s population grows and climate change continues. However, there is great potential to improve drinking water pipelines, as 30% of fresh water is lost between the supplier and consumer. While systematic process monitoring could play a crucial role in the early detection and repair of leaks, current practice requires manual inspection, which is both time-consuming and costly. This project envisages maintenance-free measurements at numerous locations within the underground infrastructure, a goal that is to be achieved through the use of a harvesting device mounted on the water pipe. This device extracts energy from the temperature difference between the water pipe and the soil using a TEG (thermoelectric generator), takes sensor measurements, processes the data and transmits it wirelessly via LoRaWAN. We built 16 harvesting devices, installed them in four locations and continuously evaluated their performance throughout the project. In this paper, we focus on two devices of a particular type. The data for a full year show that enough energy was available on 94% of the days, on average, to take measurements and transmit data. This study demonstrates that it is possible to power highly constrained sensing devices with energy harvesting in underground environments. Full article

Biologging refers to the use of animal-borne recording devices to study wildlife behavior. In the case of audio recording, such devices generate large amounts of data over several months, and thus require some level of processing automation for the raw data collected. Academics have widely adopted offline deep-learning-classification algorithms to extract meaningful information from large datasets, mainly using time-frequency signal representations such as spectrograms. Because of the high deployment costs of animal-borne devices, the autonomy/weight ratio remains by far the fundamental concern. Basically, power consumption is addressed using onboard mass storage (no wireless transmission), yet the energy cost associated with data storage activity is far from negligible. In this paper, we evaluate various strategies to reduce the amount of stored data, making the fair assumption that audio will be categorized using a deep-learning classifier at some point of the process. This assumption opens up several scenarios, from straightforward raw audio storage paired with further offline classification on one side, to a fully embedded AI engine on the other side, with embedded audio compression or feature extraction in between. This paper investigates three approaches focusing on data-dimension reduction: (i) traditional inline audio compression, namely ADPCM and MP3, (ii) full deep-learning classification at the edge, and (iii) embedded pre-processing that only computes and stores spectrograms for later offline classification. We characterized each approach in terms of total (sensor + CPU + mass-storage) edge power consumption (i.e., recorder autonomy) and classification accuracy. Our results demonstrate that ADPCM encoding brings 17.6% energy savings compared to the baseline system (i.e., uncompressed raw audio samples). Using such compressed data, a state-of-the-art spectrogram-based classification model still achieves 91.25% accuracy on open speech datasets. Performing inline data-preparation can significantly reduce the amount of stored data allowing for a 19.8% energy saving compared to the baseline system, while still achieving 89% accuracy during classification. These results show that while massive data reduction can be achieved through the use of inline computation of spectrograms, it translates to little benefit on device autonomy when compared to ADPCM encoding, with the added downside of losing original audio information. Full article