Falls remain one of the leading causes of injury and mortality worldwide, underscoring the need for accurate and energy-efficient wearable detection systems. Existing deep neural networks achieve high recognition accuracy but require dense, always-on computation that is impractical for continuous monitoring. Meanwhile, current fall detection frameworks process all activities with the same high-resolution pipeline, wasting energy on steady daily motions that dominate most of the time. To address these challenges, we propose a conditional dual-path spiking framework based on Adaptive Leaky Integrate-and-Fire (ALIF) neurons. A steady-state path continuously monitors low-frequency daily activities at reduced sampling, while a transient-response path activates only when high-frequency motion or low-confidence predictions occur. This event-driven design dynamically allocates computation to salient motion periods, ensuring efficient and responsive operation. The proposed model achieves 98.6% overall accuracy on the KFall dataset, while reducing computation by 56.8% compared to dense ANN baselines.

Frequency-domain functional near-infrared spectroscopy (FD-fNIRS) ultilizes the propagation time and amplitude attenuation of light in tissue to quantify blood oxygen concentration. This enables more detailed monitoring of human brain activity, providing a potential solution for non-invasive and wearable brain-computer interface (BCI) devices. However, the amplitude-phase coupling effect of the comparator in the phase detector introduces phase errors, which in turn leads to inaccuracies in the measurement of light propagation time. To suppress amplitude-phase coupling effect, an amplitude control loop based on the level-crossing analog-to-digital converter (LC ADC) is employed. The utilization of the LC ADC eliminates the need for a high-speed envelope detector, thereby reducing circuit complexity and power consumption. In addition, a sub-1V transimpedance amplifier (TIA) is employed to improve energy efficiency. The proposed structure achieves a phase error of 0.11° and 12 mW power consumption.

A flexible bi-directional bio-interface design using indium zinc oxide (IZO) thin-film transistors (TFTs) to share the same microelectrode array and electrical interconnects is proposed for direct electrical stimulation (DES) and bio-potential signals sensing. The oxide TFTs present a high current driving capability (mA level) for stimulation and a high ON-OFF ratio (>10(9)) for switching with good uniformity. An active-matrix pixel design is developed to provide higher stimulation current and also enable the fabricated microelectrode array to be highly robust and tolerant to customizable cutting for various wearable applications. Finally, a bio-interface system is built based on the flexible microelectrode array for electrocardiogram (ECG) and electromyography (EMG) tests on a mouse.

This paper details the design and simulation-based evaluation of three distinct N-type-only low-temperature polycrystalline silicon thin-film transistor (LTPS-TFT) voltage reference circuits, tailored for the specific demands of flexible electronic systems. Addressing the inherent design challenges in TFT technology, these circuits employ a common strategy: generating a current with a negative temperature coefficient to compensate for proportional-to-absolute-temperature (PTAT) voltage characteristics. Comprehensive post-layout and statistical Monte Carlo simulation results highlight that the first of the proposed configurations achieves superior performance, exhibiting a temperature coefficient (TC) of 34.03 ppm/°C, a line sensitivity of 0.61 %/V, and a power supply rejection ratio (PSRR) of -42.34 dB. This study presents new circuit-level approaches for achieving reliable voltage references in flexible applications, thereby advancing the development of high-performance, flexible electronic systems.

Compact, energy-efficient sensor interface circuits capable of multimodal sensing are essential for next-generation IoT and wearable devices. Present architectures, often relying on separate readout channels, face significant challenges regarding hardware complexity, dynamic range (DR) degradation due to sensor baseline variations, and susceptibility to low-frequency flicker noise. To address these issues, this work presents a sub-mu W dual-mode reconfigurable sensor interface circuit (RSIC). By streamlining the architecture into a dual-mode topology that includes a resistive-feedback mode for V/I/R sensing and a charge-redistribution mode for C sensing, the design achieves high area efficiency with a minimal reconfigurable passive network and a shared core amplifier. To prevent saturation from sensor-dependent baseline drift, a dynamic baseline pre-compensation circuit is integrated to increase 2%-25% of the linear range for V/I/R/C sensing while maintaining a high coefficient of determination (R-2 > 0.9999). Furthermore, the autozeroing (AZ) technique is employed to suppress the dominant flicker noise components. A prototype IC was fabricated in a 55-nm CMOS process with a 1.2-V supply. Experimental results demonstrate that the prototype achieves sensing ranges of 0.2-1.1 V, 100 nA-5.5 mu A, 50 k Omega-5 M Omega, and 4-36 pF. With the AZ technique enabled, the input-referred noise floor is reduced by a factor of 4. The RSIC occupies a compact area of 0.18 mm(2) and consumes 890-1000 nW, achieving a 2.7x to 325x power reduction compared to prior art.

The small size, portability and non invasive features of functional near-infrared spectroscopy (fNIRS) system position it as one of the outstanding candidates for continuous and dynamic measurement of vascular parameters, therefore enabling blood pressure (BP) monitoring. However, the nonlinearity of optical quantities collected by multiple channels from fNIRS systems makes it difficult to convert real-time data to vascular parameters (radius and depth) efficiently. This paper proposes a numerical compact model featuring a forward/inverse fast lookup table (FLUT) and a depth-adaptive calibration scheme for source-detector (S-D) channel activation. Sensitivity analysis and optimization on finite element (FEM) simulation dataset is employed to aid the method. Simulation results demonstrate its outstanding performance with 0.6 ms single inverse operation time, 0.005 mm vascular radius resolution and a 49.8 dB dynamic SNR. Compared with the full-channel model, vascular radius reconstruction time is saved by 98%. The proposed solution provides a promising platform for continuous, real-time vascular monitoring in dedicated fNIRS applications.

This demonstration presents a reconfigurable self-regulating wearable near-infrared spectroscopy(NIRS) detection system for non-invasive real-time monitoring of hemoglobin concentration in human tissues. The system utilizes dual-wavelength LEDs at 735 nm and 850 nm as the light source, estimating hemoglobin concentration by detecting the intensity of light irradiated onto the body and reflected back. The system is highly portable and flexible, can be worn on different parts of the body, and allows users to conveniently monitor hemoglobin concentration.

The wearable electrocardiogram (ECG) sensor has immense potential for cardiovascular healthcare monitoring. However, uploading ECG signals to server-side devices for diagnosis poses privacy risks. Multi-tasking ECG processors rely on multi-beat windows and complex models, leading to high latency and energy consumption that limit deployment in resource-constrained wearable devices. In this work, we present a single-lead ECG sensor for authenticated healthcare that performs QRS complex detection (Task 1), arrhythmia detection (Task 2), bio-identification (Task 3), and bio-authentication (Task4). This sensor uses an invariant spectrum feature extractor on dual-mode single-beat ECG segments to derive normalized biometric features for these tasks, reducing the record latency to a single beat. A binary neural network (BNN) model is used for multi-task decision-making, achieving an accuracy of >94% across all tasks in the MIT-BIH database while reducing the model size to 5.725 kB. The sensor, implemented in 55-nm technology, occupies 0.50 mm(2), with the digital processing core (DPC) occupying 0.27 mm(2). Operating at a 0.85-V supply and 320-kHz clock, the DPC achieves a task latency of 1.04 s, representing a 2 & times; - 12 & times; reduction compared to state-of-the-art work, with inference energy of 0.19 mu J/task ( 3.7 & times; - 26 & times; reduction).

The temperature field across MEMS resonant accelerometers exhibits nonuniform and dynamic distributions under practical operating conditions, which makes temperature monitoring and thermal drift compensation in silicon accelerometers challenging. The ring-down driving scheme enables real-time quality factor monitoring for accurate in-situ temperature measurement, while frequency ripple induced by the amplitude-stiffening (A-S) effect degrades frequency stability. Addressing this problem, we propose a ring-down driving scheme with active ripple suppression (ARS), where the real-time monitored quality factor serves as a virtual thermometer, and the resonant frequency ripple is suppressed through electrostatic stiffness modulation. During continuous ring-down operation, the amplitude of the sensing signal is utilized as a phase stamp to continuously adjust the bias voltage of the driving signal via the dual-LUT and employ a calibration circuit based on P-type iterative learning control (ILC) to calibrate the feedback compensation relationship. The results of the experiment demonstrate that the proposed technique reduced the peak-to-peak frequency ripple from 7.3 Hz to 0.8 Hz, a suppression of 89% and achieved a harmonic suppression ratio of 41.9 dB, representing an improvement of $18.8\times $ . The proposed technique in generalization experiments achieved an average harmonic suppression ratio of 36.1 dB with a standard deviation of 5.4 dB. Additionally, it enables near in-situ temperature monitoring with low hysteresis, with a temperature resolution better than 0.2 °C within 1 Hz bandwidth through quality measurement. For applications of resonant accelerometers, the proposed ARS method achieves an acceleration resolution of 2.3 mg within a 10 Hz bandwidth and a bias stability of 1.0 mg after temperature compensation over the temperature range of 0 °C to 45 °C. The proposed system enables both real-time and accurate in-situ temperature tracking while improving acceleration resolution.[2026-0068]

This article presents a fully integrated frequency-domain (FD) functional near-infrared spectroscopy (fNIRS) detection integrated circuit (IC) designed for non-invasive measurement of tissue metabolite optical properties. Departing from traditional static architectures, a dynamic light sensing architecture is proposed, which allows for the decoupling of time-of-flight (ToF) resolution from the system power consumption through duty-cycle modulation, thereby enhancing energy efficiency. In addition, to improve the measurement precision of both ToF and intensity loss, an inter-stabilized intensity and phase-to-digital converter (IS-IPDC) is proposed to resolve coupling issues between intensity and phase quantification. The chip is implemented in a standard 180-nm CMOS process. Test results indicate that the light ToF resolution is 2.2 ps within a 10-Hz bandwidth, while consuming only 12.5 mW. Thanks to its high resolution and crosstalk-free characteristics, compared with the high-precision instrument-based reference system, the maximum measurement errors for the absorption coefficient (mu(a)) and reduced scattering coefficient (mu(')(s)) are 4.2% and 4%, respectively. Finally, comprehensive in vitro and in vivo demonstrations demonstrate the IC's capabilities for metabolic imaging and long-term monitoring.

Electroencephalography (EEG) provides a non-invasive window into brain activity, enabling Brain-Computer Interfaces (BCIs) for communication and control. However, their performance is limited by signal fidelity issues, among which the choice of re-referencing strategy is a pervasive but often overlooked preprocessing bias. Addressing controversies about its necessity and optimal choice, we adopted a quantified approach to evaluate four strategies - no re-referencing, Common Average Reference (CAR), small Laplacian, and large Laplacian - using 62-channels EEG (31 subjects, 2,520 trials). To our knowledge, this is the first study systematically quantifying their impact on single-trial P300 classification accuracy. Our controlled pipeline isolated re-referencing effects for source-space reconstruction (eLORETA with Phase Lag Index) and anatomically constrained classification. The large Laplacian resolves distributed P3b networks while maintaining P3a specificity, achieving the best P300 peak classification accuracy (81.57% hybrid method; 75.97% majority regions of interest). Performance follows a consistent and statistically significant hierarchy: large Laplacian > CAR > no re-reference > small Laplacian, providing a foundation for unified methodological evaluation.

Functional near-infrared spectroscopy (fNIRS) systems are particularly suited for psychiatric studies due to their mobility and flexibility [1]. Frequency-domain (FD) NIRS utilizes multiple source-detector distances (MD) or carrier frequencies (MF) to assess optical properties at varying tissue depths, with both methods yielding comparable results [2]. While MF FD-NIRS is more complex, it is less sensitive to light source-detector coupling variations, enabling more compact systems with fewer optical components. However, most existing work has focused on MD-based approaches [3–5], with limited exploration of MF methods [6]. Current MF FD-NIRS circuits still faces two main challenges (Fig. 1): First, although dynamic architecture has been proposed to decouple resolution from power consumption [4], using the same power across different carrier frequencies still limits energy efficiency. Second, avalanche photodiode’s (APD) intensity-phase crosstalk [7] and common offset complicate the accurate determination of metabolite concentrations, leading to diagnostic inaccuracies. Moreover, offset calibration requires additional hardware, increasing system complexity and cost.

transistor (TFT) can be fabricated on flexible substrates, offering potential for seamless, long-term wearable health monitoring. However, the limited performance, process variation, and motion artifacts hinder the widespread adoption of flexible wearable devices. This article presents a TFT-based photoplethysmography (PPG) acquisition circuit for flexible healthcare rings, integrating a voltage-controlled oscillator (VCO) based analog front-end (AFE) to overcome the performance limitations caused by the low mobility of TFT devices. Meanwhile, a dead-zone free (DZ-Free) phase-frequency detector (PFD) was proposed and integrated into the VCO transimpedance amplifier (TIA) to enhance the linearity. Additionally, a 0-1 MASH VCO-analog-to-digital conversion (ADC) structure was adopted to improve the quantization dynamic range (DR). The system is implemented using 3-mu m low-temperature poly-silicon (LTPS) TFT (LTPS-TFT) process of TIANMA Microelectronics. The measurement results show that DZ-Free PFD improves the linearity and the proposed VCO-TIA achieves a 61.5-dB DR. The proposed ADC demonstrates a 72-dB spurious-free dynamic range (SFDR), a 65.8-dB SNDR and an 80-dB DR. Under temperature sweeping from-40 degrees C to 80 degrees C and supply voltage variations of 10 +/- 2 V, the ADC maintains performance variations within 3%, while the TIA stably sustains a gain of 110 dB Omega. Finally, PPG and blood oxygen saturation (SpO2) measurements were demonstrated, including ambient light compensation. This work provides a feasibility demonstration for TFT-based fully integrated flexible electronics in bio-signal monitoring applications.

Silicon oscillating accelerometers have the potential to achieve navigation-grade precision due to their frequency-modulation nature. As a result, they present a promising alternative to quartz accelerometers in smart unmanned platforms that require satellite-free navigation. This tutorial will begin with an introduction to the fundamental principles and specification requirements of MEMS resonant sensors. It will then delve into the noise modelling and agile design methodologies essential for optimizing performance. Advancements in low-noise, highefficiency readout circuits will also be explored. Finally, the impact of long-term drift errors caused by temperature fluctuations, process variations, and environmental changes will be discussed, along with advanced compensation techniques. By addressing key challenges and solutions, this tutorial aims to provide valuable insights for sensor and system designers working on high-precision MEMS accelerometers and navigation applications. This talk is sponsored by IEEE Circuits and Systems Society.

Electroencephalography (EEG)-based wearable brain-computer interfaces (BCIs) face challenges due to low signal-to-noise ratio (SNR) and non-stationary neural activity. We introduce in this manuscript a mathematically rigorous framework that combines data-driven noise interval evaluation with advanced SNR visualization to address these limitations. Analysis of the publicly available Eye-BCI multimodal dataset demonstrates the method's ability to recover canonical P300 characteristics across frequency bands (delta: 0.5-4 Hz, theta: 4-7.5 Hz, broadband: 1-15 Hz), with precise spatiotemporal localization of both P3a (frontocentral) and P3b (parietal) subcomponents. To the best of our knowledge, this is the first study to systematically assess the impact of noise interval selection on EEG signal quality. Cross-session correlations for four different choices of noise intervals spanning from early to late pre-stimulus phases also indicate that alertness and task engagement states modulate noise interval sensitivity, suggesting broader applications for adaptive BCI systems. While validated in healthy participants, our results represent a first step towards providing clinicians with an interpretable tool for detecting neurophysiological abnormalities and provides quantifiable metrics for system optimization.

Healthcare Education and Labeling for Cardiopulmonary Sounds, HEALSound, is an interactive platform designed to enhance the identification and labeling of adventitious cardiopulmonary sounds through gamification. HEALSound enables users, particularly medical professionals, to engage in exercises that improve their knowledge of abnormal heart and lung sounds while simultaneously contributing to the labeling of raw audio data. By incorporating real-time feedback and progress tracking, the app promotes continuous learning and provides a valuable crowdsourced resource for building high-quality labeled datasets over time. This dual-purpose platform not only aids in medical education but also contributes to the advancement of machine learning models for sound classification in healthcare. The system demonstrates a new potential for significant impact in educational and clinical environments by seamlessly integrating learning with data collection, ensuring scalability and the continuous improvement of cardiopulmonary sound databases.

This paper presents a reconfigurable readout architecture for continuous-wave (CW) and frequency-domain (FD) opto-biological sensing, primarily targeting tissue and brain imaging applications. By reusing hardware resources, the proposed design supports three operating modes: single-channel CW, multi-channel CW, and FD. A duty-cycle-adjustable trans-impedance amplifier (TIA) is introduced to achieve a trade-off between bandwidth and power consumption. Measurement results demonstrate an average power consumption of only 1.99mW, which is significantly lower than that of previously reported FD-only implementations. In CW mode, the correlation coefficient between input and output signals reaches 0.9721, while in FD mode, the measured phase values exhibit a high linearity ($r^{2}=0.9882$) with respect to theoretical predictions.

The accurate and consistent translation of design rules/manuals into machine-readable Design Rule Check (DRC) formats is critical for ensuring the manufacturability and reliability of integrated circuits. However, the process is often hindered by inconsistencies in the language and structure of design rules/manuals, which lead to errors in DRC implementation across various Electronic Design Automation (EDA) tools. This paper presents a new tool D2D-LLM+, that leverages large language models to automate the translation of DRC codes into design rules/manuals compatible with leading EDA tools such as Mentor, Cadence, Synopsys, and open-source platforms like KLayout and Magic. Key challenges addressed include the hallucination and stability issues caused by inconsistent terminology and ambiguous language, as well as the visualization challenges of design rules. Additionally, we explore strategies to enhance large language models (LLM) learning through dataset augmentation, improving result accuracy, and increasing the stability and efficiency of LLM models through various prompt engineering techniques. Through extensive validation, we demonstrate the effectiveness of our approach in reducing errors and streamlining the DRC workflow, paving the way for more reliable IC design processes.

Sensing technologies continue to evolve and become ubiquitous, which greatly improves the quality of life. The multi-modal sensor systems record different types of electrical signals simultaneously while demanding small dimensions, low cost, long battery life, and lightweight requirements [1–5]. The nature of these signals can be classified into voltage, current, resistive, and capacitive (V/I/R/C). Prior multi-modal integrated circuits [2] customized their analog frontend (AFE) circuits based on the type of signals to achieve the best performance at the expense of design complexity, area, and power consumption. The reconfigurable signal acquisition circuit aims to use the least number of circuit components and configure itself to provide the correct signal conditioning interface, thereby addressing the aforementioned problems [1], [3–5].