Journal of Low Power Electronics and Applications

The primary functions of an energy harvesting system include the harvesting, transformation, management, and storage of energy. Until now, various types of energy, with different power levels, have been harvested and stored by the energy harvesting system. In low-power scenarios, such as microwaves, sound, friction, and pressure, a specific low-power energy harvesting system is required. Due to the absence of an external power supply in such systems, cold-start circuits play a crucial role in igniting the low-power energy harvesting system, ensuring a reliable start-up from the initial state. This paper reviews the categorization and characteristics of energy harvesting systems, with a focus on the design and performance parameters of cold-start circuits. A tabular comparison of existing cold-start strategies is presented herein. The study demonstrates that resonance-based integrated cold-start methods offer significant advantages in terms of conversion efficiency and dynamic range, while ring oscillator-based integrated cold-start methods achieve the lowest start-up voltage. Additionally, the paper discusses the challenges of self-starting and future research directions, highlighting the potential role of emerging technologies, such as artificial intelligence (AI) and neural networks, in optimizing the design of energy harvesting systems.

Demand side management is a critical issue in the energy sector. Recent events such as the global energy crisis, costs, the necessity to reduce greenhouse emissions, and extreme weather conditions have increased the need for energy efficiency. Thus, accurately predicting energy consumption is one of the key steps in addressing inefficiency in energy consumption and its optimization. In this regard, accurate predictions on a daily, hourly, and minute-by-minute basis would not only minimize wastage but would also help to save costs. In this article, we propose intelligent models using ensembles of convolutional neural network (CNN), long-short-term memory (LSTM), bi-directional LSTM and gated recurrent units (GRUs) neural network models for daily, hourly, and minute-by-minute predictions of energy consumptions in smart buildings. The proposed models outperform state-of-the-art deep neural network models for predicting minute-by-minute energy consumption, with a mean square error of 0.109. The evaluated hybrid models also capture more latent trends in the data than traditional single models. The results highlight the potential of using hybrid deep learning models for improved energy efficiency management in smart buildings.

Low-power portable devices play a major role in IoT applications, where the analog-to-digital converters (ADCs) are very important components for the processing of analog signals. In this paper, a 5-bit two-step flash ADC with a low-power double-tail dynamic comparator (DTDC) using the control switching technique is presented. The most significant bit (MSB) in the proposed design is produced by only one low-power DTDC in the first stage, and the remaining bits are generated by the flash ADC in the second stage with the help of an auto-control circuit. A control circuit produced reference voltages with respect to the control input and mid-point voltage (Vk). The proposed design and simulations are carried out using 90 nm CMOS technology. The result shows that the peak differential non-linearity (DNL) and integral non-linearity (INL) are +0.60/−0.69 and +0.66/−0.40 LSB, respectively. The signal-to-noise and distortion ratio (SNDR) for an input signal having a frequency of 1.75 MHz is found to be 30.31 dB. The total power consumption of the proposed design is significantly reduced, which is 439.178 μW for a supply voltage of 1.2 V. The figure of merit (FOM) is about 0.054 pJ/conversion step at 250 MS/s. The present design provides low power consumption and occupies less area compared to the existing works.

Power integrity is a critical aspect of microcontroller (MCU) system design. The present tendency of increasing current density and operating frequency, along with decreasing operating voltage, significantly diminishes voltage margins. Given the cost efficiency required for MCU systems, this context places important constraints on the design of the power distribution network (PDN), which directly impacts power supply noise. Therefore, characterizing the PDN is necessary. This paper introduces a cost-effective measurement and modeling method to estimate the die-package resonance frequency of the PDN, a major threat to power integrity. The method, applied to two 32-bit MCUs from STMicroelectronics with varying PDN configurations, enables the identification of the die-package resonance frequency. The results lead to the refinement of the die capacitance model for both cases, with a maximum relative error of less than 7%. The final objective is to implement the measurement system in the die in order to adjust the PDN if necessary.

Ionoacoustic tomography exploits the acoustic signal generated by the fast energy deposition along the path of pulsed particle beams to reconstruct with sub-mm precision the dose deposition, with promising envisioned applications in hadron therapy treatment monitoring. State-of-the-art ionoacoustic detectors mainly rely on single-channel sensors and time-of-flight measurements to provide 1D localization of the maximum dose deposition at the so-called Bragg peak. This work investigates the design challenges of multichannel sensors for ionoacoustic tomography in terms of their ability to accurately reconstruct the dose deposition of a 200 MeV clinical proton beam, highlighting the impact of the number of channels in the array and their directivity. A complete acoustic model of the sensors and environment has been developed and used to find an optimum tradeoff between accuracy, evaluated numerically through the gamma index, and hardware complexity due to higher channel numbers, thus minimizing the system-level power consumption of the detector.