Recent advances in the design and implementation of wearable resistive, capacitive, and optical strain sensors are summarized herein. Wearable and stretchable strain sensors have received extensive research interest due to their applications in personalized healthcare, human motion detection, human–machine interfaces, soft robotics, and beyond. The disconnection of overlapped nanomaterials, reversible opening/closing of microcracks in sensing films, and alteration of the tunneling resistance have been successfully adopted to develop high‐performance resistive‐type sensors. On the other hand, the sensing behavior of capacitive‐type and optical strain sensors is largely governed by their geometrical changes under stretching/releasing cycles. The sensor design parameters, including stretchability, sensitivity, linearity, hysteresis, and dynamic durability, are comprehensively discussed. Finally, the promising applications of wearable strain sensors are highlighted in detail. Although considerable progress has been made so far, wearable strain sensors are still in their prototype stage, and several challenges in the manufacturing of integrated and multifunctional strain sensors should be yet tackled.

Wearable and Stretchable Strain Sensors: Materials, Sensing Mechanisms, and Applications

With the fast development of the fifth-generation cellular network technology (5G), the future sensors and microelectromechanical systems (MEMS)/nanoelectromechanical systems (NEMS) are presenting a more and more critical role to provide information in our daily life. This review paper introduces the development trends and perspectives of the future sensors and MEMS/NEMS. Starting from the issues of the MEMS fabrication, we introduced typical MEMS sensors for their applications in the Internet of Things (IoTs), such as MEMS physical sensor, MEMS acoustic sensor, and MEMS gas sensor. Toward the trends in intelligence and less power consumption, MEMS components including MEMS/NEMS switch, piezoelectric micromachined ultrasonic transducer (PMUT), and MEMS energy harvesting were investigated to assist the future sensors, such as event-based or almost zero-power. Furthermore, MEMS rigid substrate toward NEMS flexible-based for flexibility and interface was discussed as another important development trend for next-generation wearable or multi-functional sensors. Around the issues about the big data and human-machine realization for human beings’ manipulation, artificial intelligence (AI) and virtual reality (VR) technologies were finally realized using sensor nodes and its wave identification as future trends for various scenarios.

Development Trends and Perspectives of Future Sensors and MEMS/NEMS.

Hydrogels offer tissue-like compliance, stretchability, fracture toughness, ionic conductivity and compatibility with biological tissues. However, their electrical conductivity ( 350 S cm−1) and is capable of delivering direct current while maintaining soft compliance (Young’s modulus < 10 kPa) and deformability. Micrometre-sized silver flakes are suspended in a polyacrylamide–alginate hydrogel matrix and, after going through a partial dehydration process, the flakes form percolating networks that are electrically conductive and robust to mechanical deformations. To illustrate the capabilities of our silver–hydrogel composite, we use the material in a stingray-inspired swimmer and a neuromuscular electrical stimulation electrode. A hydrogel composite that consists of micrometre-sized silver flakes suspended in a polyacrylamide–alginate hydrogel matrix exhibits a high electrical conductivity of over 350 S cm−1 and a low Young’s modulus of less than 10 kPa.

An electrically conductive silver–polyacrylamide–alginate hydrogel composite for soft electronics

With the prospect of a smart society in the foreseeable future, humans are experiencing an increased link to electronics in the digital world, which can benefit our life and productivity drastically. In recent decades, advances in the Human Machine Interface (HMI) have improved from tactile sensors, such as touchpads and joysticks, to now include the accurate detection of dexterous body movements in more diversified and sophisticated devices. Advancements in highly adaptive machine learning techniques, neural interfaces, and neuromorphic sensing have generated the potential for an economic and feasible solution for next-generation applications such as wearable HMIs with intimate and multi-directional sensing capabilities. This review offers a general knowledge of HMI technologies beginning with tactile sensors and their piezoresistive, capacitive, piezoelectric, and triboelectric sensing mechanisms. A further discussion is given on how machine learning, neural interfaces, and neuromorphic electronics can be used to enhance next-generation HMIs in an upcoming 5 G infrastructure and advancements in the internet of things and artificial intelligence of things in the near future. The efficient interactions with kinetic and physiological signals from human body through the fusion of tactile sensor and neural electronics will bring a revolution to both the advanced manipulation and medical rehabilitation.

Technologies toward next generation human machine interfaces: From machine learning enhanced tactile sensing to neuromorphic sensory systems

Electronic tattoos have great potential in health and movement sensing applications on the skin. However, existing electronic tattoos cannot be conformal, sticky, and multilayered at the same time. Here, we have achieved multilayered integration of the electronic tattoo that is highly stretchable (800%), conformal, and sticky. This electronic tattoo can enable the crease amplification effect, which can amplify the output signal of integrated strain sensors by three times. The tattoo can be transferred to different surfaces and form a firm attachment, where no solvent or heat is needed. The tattoo fabrication is straightforward and scalable; a layer-by-layer strategy and two materials (metal-polymer conductors and the elastomeric block copolymer) are used to fabricate the circuit module with desirable numbers of layers within the tattoo. A three-layered tattoo integrating 1 heater and 15 strain sensors is developed for temperature adjustment, movement monitoring, and remote control of robots.

Multilayered electronic transfer tattoo that can enable the crease amplification effect

Wearable electronic circuits based on the films of gallium and its alloys offer promising implementations in health monitoring and gaming sensing applications. However, the complex rheology of liquid metals makes it challenging to manufacture thin gallium‐based devices with reliable, reproducible, and stable properties over time. Herein, the surface coating and topography of silicone substrates are engineered to enable precisely defined, micrometer‐thick liquid metal patterns over large (>10 cm2) surface areas, and high design versatility. Control over the film microstructure meets manufacturing conditions that enable gallium films with a precise, repeatable, and durable electromechanical performance. The robustness and applicability of this technology in a virtual‐reality scenario is demonstrated by implementing thin, soft, and stretchable gallium‐based sensors to accurately monitor human hand kinematics.

https://onlinelibrary.wiley.com/doi/pdf/10.1002/aisy.201900079

Gallium‐Based Thin Films for Wearable Human Motion Sensors

Analog in Memory Computing (AiMC) based neural network acceleration is a promising solution to increase the energy efficiency of deep neural networks deployment. However, the quantization requirements of these analog systems are not compatible with state-of-the-art neural network quantization techniques. Indeed, while the quantization of the weights and activations is considered by modern deep neural network quantization techniques, AiMC accelerators also impose the quantization of each Matrix Vector Multiplication (MVM) result. In most demonstrated AiMC implementations, the quantization range of MVM results is considered a fixed parameter of the accelerator. This work demonstrates that dynamic control over this quantization range is possible but also desirable for analog neural networks acceleration. An AiMC compatible quantization flow coupled with a hardware aware quantization range driving technique is introduced to fully exploit these dynamic ranges. Using CIFAR-10 and ImageNet as benchmarks, the proposed solution results in networks that are both more accurate and more robust to the inherent vulnerability of analog circuits than fixed quantization range based approaches.

/pdf/dynamic-quantization-range-control-for-analog-in-memory-2men41yt.pdf

Dynamic Quantization Range Control for Analog-in-Memory Neural Networks Acceleration

Recent years have seen a growing trend of deploying deep neural network-based applications on edge devices. Many of these applications, such as biometric identification, activity tracking, user preference learning, etc., require fine-tuning of the trained networks for user personalization. One way to prepare these models to handle new, unseen tasks, is to pre-train them on a distribution of known tasks. This observation has led to increasing research into meta-learning based few-shot learning techniques. However, basic meta-learning approaches do not account for the limited memory and computational resources during on-chip training. We propose a modified meta-learning algorithm that enables quantized fine-tuning to optimally condition the models for on-chip few shot learning. The modification involves the inclusion of target hardware constraints upfront in the meta-learning process. Block floating point datatypes with low precision mantissa bits are utilized in the forward and backward passes, to allow hardware-friendly adaptation. Experiments show that our algorithm provides better initializations than conventional algorithms, more suitable for efficient quantized fine-tuning. This allows the few-shot learner to achieve better convergence, in terms of accuracy and speed. Extensive experiments are also performed to analyze the impact of initialization on quantized fine-tuning and further corroborate the benefits of our method.

Learn to Learn on Chip: Hardware-aware Meta-learning for Quantized Few-shot Learning at the Edge

Analog in Memory Computing (AiMC) is a promising solution to achieve energy efficient deployment of Deep Neural Networks (DNNs). However, to this day, the accuracies of DNNs deployed using AIMC do not compare favorably with digital alternatives. In this work, we illustrate that these lacking performance can be partly attributed to mismatched arithmetic assumption. We show that correcting those assumptions results in more robust and accurate DNNs on analog platforms. By further leveraging insights from AiMC related methods, we question modern neural network quantization techniques, revealing, in the process, alternative multiplication-less modes of executions for DNNs.

Nathan Laubeuf

Papers

Gallium‐Based Thin Films for Wearable Human Motion Sensors

Dynamic Quantization Range Control for Analog-in-Memory Neural Networks Acceleration

Learn to Learn on Chip: Hardware-aware Meta-learning for Quantized Few-shot Learning at the Edge

Analog Compute in Memory and Breaking Digital Number Representations