FULL PAPER

www.advintellsyst.com

Liquid Metal-Based Strain Sensor with Ultralow Detection

Limit for Human–Machine Interface Applications
Yuanzhao Wu, Youlin Zhou, Waqas Asghar, Yiwei Liu,* Fali Li, Dandan Sun,
Chao Hu, Zhenguang Wu, Jie Shang, Zhe Yu, Run-Wei Li,* and Huali Yang

are realizing such innovative applications

Flexible strain sensors play vital role in human–machine interaction. Despite their that are impossible for rigid HMIs.
vast development, a strain sensor having broad strain sensing range, ultralow Flexible strain sensors are important
detection limit, and negligible hysteresis still remains a challenge. Herein, a liquid component of soft HMI systems.[3] To date,
significant efforts are made to develop flex-
metal (LM)-based stretchable resistive strain sensor, prepared by selective wetting
ible strain sensors with ultralow detection
and transferring process to attain improved compatibility between LM and limit and broad sensing range. Generally,
polydimethylsiloxane substrate, is reported. This sensor exhibits broad strain the performance of a strain sensor is
sensing range (105%), with ultralow detection limit (0.05%), minimal hysteresis, improved by choosing stretchable polymers
fast response time (58 ms), and excellent repeatability. When practically dem- or by designing novel sensing structures.[3–5]
onstrated, this sensor successfully monitored various human activities, such as Bao et al. have reported carbon nanotube
(CNT)-based spiral-structured stretchable
blink motion monitoring, voice intensity differentiation, heartbeat and wrist pulse
strain sensor that can accommodate stretch-
monitoring. Moreover, when the current sensor is used in the form of smart ing strain up to 150%.[6] In contrast, a vari-
kneecap and smart glove, it efficiently detected various human gestures, which ety of structures are designed to enhance
confirms its great application potential in the fields of human health and motion the performance of strain sensors. For
monitoring, human–machine interface, and virtual reality applications. instance, Kang et al. have developed strain
sensors by generating cracks in the plati-
num (Pt) film, which demonstrated the
detection limit of 0.1% and a gauge factor
1. Introduction of over 2000, within 0–2% strain range.[7] Peng et al. manufac-
tured a strain sensor by depositing CNTs on spiral-shaped poly-
Human–machine interaction (HMI) enables two-way informa- propylene fiber and achieved wide sensing range of 200% strain
tion transfer between humans and machines (such as computers with ultralow detection limit of 0.01%.[8] Despite these advance-
or robots).[1] Conventional rigid HMIs are widely used in various ments, a strain sensor having precise displacement–resistance
applications, such as rehabilitation robots, motion gesture response with high resolution, ultralow detection limit, and broad
monitoring, and biological health monitoring. However, their sensing range still remains a challenge.
rigid nature leads to discomfort and unstable signal problem Gallium-based LM alloys are considered highly suitable mate-
during the motion of human body. Alternatively, soft HMIs pro- rial for stretchable strain sensors because of their capability of
vide all the sensing functions of conventional rigid HMIs but retaining high electrical conductivity even under extreme stretch
with the ability of being stretched, compressed, twisted, bent, condition.[9,10] These LMs exhibit low viscosity and fluidic nature,
and deformed into arbitrary shapes.[2] Nowadays, soft HMIs which enables LM to follow the deformation path of soft

Dr. Y. Wu, Y. Zhou, W. Asghar, Prof. Y. Liu, F. Li, D. Sun, C. Hu, Z. Wu, Prof. Y. Liu, Prof. J. Shang, Prof. R.-W. Li
Prof. J. Shang, Z. Yu, Prof. R.-W. Li, Dr. H. Yang Center of Materials Science and Optoelectronics Engineering
CAS Key Laboratory of Magnetic Materials and Devices University of Chinese Academy of Sciences
Ningbo Institute of Materials Technology and Engineering Beijing 100049, China
Chinese Academy of Sciences
Ningbo 315201, P. R. China Prof. Y. Liu, Prof. R.-W. Li
E-mail: [email protected]; [email protected] School of Future Technology
University of Chinese Academy of Sciences
The ORCID identification number(s) for the author(s) of this article Beijing 100049, P. R. China
can be found under https://doi.org/10.1002/aisy.202000235.
Dr. Y. Wu, Y. Zhou, W. Asghar, Prof. Y. Liu, F. Li, D. Sun, C. Hu, Z. Wu,
© 2021 The Authors. Advanced Intelligent Systems published by Wiley- Prof. J. Shang, Z. Yu, Prof. R.-W. Li, Dr. H. Yang
VCH GmbH. This is an open access article under the terms of the Zhejiang Province Key Laboratory of Magnetic Materials and Application
Creative Commons Attribution License, which permits use, distribution Technology
and reproduction in any medium, provided the original work is Ningbo Institute of Materials Technology and Engineering
properly cited. Chinese Academy of Sciences
Ningbo 315201, P. R. China
DOI: 10.1002/aisy.202000235

Adv. Intell. Syst. 2021, 2000235 2000235 (1 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
www.advancedsciencenews.com www.advintellsyst.com

substrate and minimizes the problem of hysteresis. This type of on sensor. The mathematical proof of aforementioned equation
behavior is absent in other conductive materials (silver, gold, is given in the Supporting Information.
carbon, etc.) because of their rigid nature.[10] However, there
exists a week compatibility between LM and elastomeric sub- 2.2. Preparation Process of the Strain Sensor
strate, which influences the final properties of the strain sensor.
Therefore, it is important to improve the compatibility of LM and Figure 1a shows the fabrication process of LM-based strain sen-
elastic substrate. Selective wetting and transferring process is sor in which a selective wetting and transferring approach is used
found useful to precisely pattern LM’s complex geometries to pattern LM Galinstan on PDMS substrate. This patterning
and to attain improved compatibility between LM and elasto- approach gives high compatibility between the LM and elasto-
meric substrate.[9,11] meric substrate of sensor.[12] Process starts by submerging strain
In this article, we report a LM-based stretchable resistive strain gauge-shaped copper foil in NaOH solution. NaOH removes the
sensor, prepared by using selective wetting and transferring gallium oxide layer of LM and makes it suitable for further proc-
process to attain improved compatibility between LM and poly- essing. Then, Galinstan is injected on the top of copper foil by
dimethylsiloxane (PDMS) substrate. Our sensor exhibits broad using syringe, which results in the formation of strain gauge-
strain sensing range of 105%, with ultralow detection limit shaped LM film on the top of copper foil. Galinstan exhibits
(0.05%) both under normal and prestretch condition, negligible hydrophilic behavior toward copper and hydrophobic (repulsive)
hysteresis, high mechanical robustness, fast response time behavior toward plastic that is why Galinstan thin film is only
(58 ms), and excellent repeatability. When capability of sensor formed on copper foil rather than on plastic supporting the cop-
is practically demonstrated, our sensor successfully detected var- per foil (Figure S1, Supporting Information). LM film deposited
ious human activities, such as blink motion monitoring, voice copper foil is cleaned in deionized water, followed by drying it in
intensity differentiation, heartbeat and wrist pulse monitoring. air. Afterward, liquid PDMS is poured on copper foil, dispersed
Moreover, strain sensor incorporated smart glove and smart by a glass slide, and then cured at 80  C for 30 min. After
kneecap are used to detect various human gestures, which con- complete curing, PDMS is peeled off, which results in the
firms the potential of our sensor in the field of human–machine transfer of strain gauge-shaped LM film from copper foil to
interaction. the PDMS substrate. During peeling operation, LM and
PDMS remain stick with each other due to the presence of
LM oxide layer, formed during prior exposure of LM film to
2. Results and Discussion air. This oxide layer improves the interfacial interaction
between LM and PDMS, as shown in Figure S2, Supporting
2.1. Working Principle of Strain Sensor Information. Finally, we deposited another layer of PDMS for
packaging purpose, which completes the formation of LM-
The design concept of our strain sensor is based on based strain sensor. Figure 1b shows the initial and stretched
deformability-dependent resistive sensing mechanism. The rela- state of strain sensor. Figure 1c shows the high stretchability
tive change in resistance (ΔR/R0) of our sensor during stretching of sensor.
is expressed by following equation
ΔR 2.3. Electromechanical Properties of the Strain Sensor
¼ ε2 þ 2ε (1)
R0
Figure 2 shows the electromechanical properties of sensor. It is
where ΔR and R0 show the respective resistances of sensor evident from Figure 2a that our sensor exhibits stable perfor-
during stretched and normal states, and ε shows the strain acting mance under repeated strains of 20%, 40%, 60%, 80%, and

Figure 1. Fabrication of LM-embedded strain sensor. a) Schematic illustration showing the fabrication process of strain sensor. LM Galinstan is
patterned on PDMS substrate by using selective wetting and transferring approach. b) Figure showing the normal and stretched state of strain sensor.
c) Figure showing the high stretchability of sensor.

Adv. Intell. Syst. 2021, 2000235 2000235 (2 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
www.advancedsciencenews.com www.advintellsyst.com

Figure 2. Electromechanical characterization of strain sensor. a) Sensor’s performance under repeated strains of 20%, 40%, 60%, 80%, and 100%
(rate ¼ 100 mm min1). Sensor exhibits high gauge factor of 2.33 and linear response to applied strains. b) Relative resistance variation of sensor under
100% cyclic strain, indicating low DH. c) Sensor’s response to cyclic strains under normal condition. d) Sensor’s response to cyclic strain under prestretch
condition (90%).

100%. Relative resistance of sensor increases with the increase in strain. Low hysteresis of sensor is attributed to the fluidic nature
strain, which confirms the existence of linear relationship of LM combined with viscoelasticity of PDMS, which makes our
between relative resistance and applied strain. When theoretical sensor highly responsive to stretching and releasing strain.
and experimental performance of sensor is compared (inset of The hysteresis curves of sensor under 20% and 60% stretching
Figure 2a), theoretical curve slightly deviates from the experimen- strain are shown in Figure S4, Supporting Information.
tal curve. This occurs because only uniform deformation of fluidic Figure 2c shows the dynamic response of sensor under increas-
channel is considered in the derivation of Equation (1). ing cyclic strains of 0.05%, 0.10%, 0.15%, 0.20%, and 0.25%. Our
Deformation of the cross-sectional area along the longitudinal sensor exhibits stable response to increasing cyclic strain while
direction can also occur unevenly depending on the sensor
geometry. In addition, the nonuniform deformation can give rise
to a different sensitivity from the aforementioned theoretical
prediction. Under the strain range of 0–100%, our sensor
depicted gauge factor of 2.33, which is calculated by the following
formula
ΔR=R0
GF ¼ ¼2þε (2)
ε
In addition, our sensor exhibits Young’s modulus of
1.1 MPa and typical elastomeric behavior, which is evident
from the sensor’s stress–strain curves (Figure S3, Supporting
Information). If A shows the area under hysteresis curve then
degree of hysteresis (DH) of sensor is quantified by the following
equation[13]
Aloading  Aunloading
DH ¼  100% (3)
Aloading

where Aloading and Aunloading are the area of loading and unload-
ing response curves, respectively. Figure 2b shows that our sen- Figure 3. Detection limit comparison of current and previously reported
sor exhibits negligible DH (0.07) even under 100% stretching strain sensors, as a function of their stretchability.

Adv. Intell. Syst. 2021, 2000235 2000235 (3 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
www.advancedsciencenews.com www.advintellsyst.com

its resistance increases regularly with the applied load. Moreover, Figure 3 shows the detection limit comparison of current and
sensor is fully capable to detect strains below 0.1%. Afterward, we previously reported strain sensors, as a function of their stretch-
prestretched our sensor to 30%, 60%, 90%, and 105% strains and ability.[4,7,14] Figure confirms that our sensor exhibits least detec-
again evaluated its performance under increasing cyclic strains tion limit (0.05%) within the 0–105% strain range, when
(Figure 2d and Figure S5, Supporting Information). compared with previously reported strain sensors. Due to this
Impressively, our sensor again showed a stable response toward factor, our sensor exhibits superior capability of measuring
repeated strains, which confirms that current sensor exhibits and distinguishing slight motions, even under the prestretched
high resolution and is fully capable of detecting minute and large strain of 105%.
applied strain, both in normal and prestretched condition. For various strain-sensing applications, stability, reproducibil-
Excellent elasticity of PDMS combined with high conductivity ity, response time, and durability of sensor also play important
of LM are the factors, responsible for ultralow detection limit role. Figure 4a shows the resistance variation of sensor with
of sensor. This demonstrates the potential applications of sensor respect to stretching cycles. Sensor’s resistance increases with
in the field of humanmachine interaction. the increase of stretching strain, but resistance variation remains

Figure 4. a) Sensor’s resistance variation as a function of stretching cycles. b) Relative resistance variation under increasing cyclic strain, confirming the
good repeatability of sensor. c,d) Sensor’s loading and unloading response time, evaluated under 0.05% strain (rate ¼ 2400 mm min1). Sensor exhibits a
response time of (58  5) ms. e) Durability of the sensor evaluated for 8000 cycles under 50% applied strain, showing sensor’s excellent mechanical
robustness. Inset shows the detailed resistance signals from 7961 to 7972 cycles.

Adv. Intell. Syst. 2021, 2000235 2000235 (4 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
www.advancedsciencenews.com www.advintellsyst.com

constant during specific strain, which demonstrates the steady sensor. So, we attached our sensor on different locations of
response of sensor. Sensor’s response toward dynamic cyclic human body (such as canthus, throat, chest, and wrist) to moni-
strain is shown in Figure 4b in which sensor’s resistance tor various human activities (Figure 5). First, the strain sensor
remained constant during the holding operation. Overall, our was fixed on the canthus to see its response to the lower move-
sensor presented good repeatability within the strain range of ments induced by blink (Figure 5a). Our sensor successfully
90%, which satisfies the requirements for human–machine detected the subtle motion of blink, in real time. Furthermore,
interface applications.[15] To determine the sensor’s response the sensor was attached on human throat to detect vibration sig-
time, we applied 0.05% strain on sensor by using computer- nals, produced due to speaking of different words such as
controlled material testing machine (Instron 5943, USA) “Liquid,” “Metal,” “Sensor,” and “NIMTE” (Figure 5b). Our sen-
(2400 mm min1) and response time is obtained by calculating sor has efficiently distinguished the various vibration signals of
the time lag between resistance and strain curves. Sensor vocal cords produced during human speech, which demonstrate
presents a response time of (58  5) ms (Figure 4c,d), which the potential applications of our sensor in the field of phonation
is comparable to the response time of previously reported strain rehabilitation exercise and humanmachine interaction. In addi-
sensors.[8] In addition, our sensor endured 8000 loading– tion, the sensor was attached on the chest for heartbeat monitor-
unloading cycles under 50% strain (rate ¼ 100 mm min1), ing and it was clearly detected by our sensor (Figure 5c). In
without showing any obvious fatigue (Figure 4e). This confirms addition, when we attached our sensor on human wrist, it also
the high mechanical robustness of our sensor. Inset of efficiently detected the wrist pulse (as expected). (Figure 5d).
Figure 4e shows that resistance signal keeps its amplitude The wrist pulse waves with ten peaks were recorded for 5 s, indi-
and waveform at different stages, confirming the high stability cating a pulse rate of 120 times per minute during exercise.
of the device. The superior performance of our LM strain We developed sensors incorporated smart glove to track the
sensor, especially in terms of wide range strain sensing motion of each human finger. Smart glove contains a printed
(105%), ultralow detection limit (0.05%), minimal hysteresis circuit board chip, which is connected to robotic hand via
(DH ¼ 0.07), excellent mechanical robustness (8000 cycles), Bluetooth. We mounted a smart glove on human hand and a
and fast response time (58 ms) makes the current manufacturing remotely placed robotic hand was operated from distant position,
method very useful, when compared with previously published by using the various gestures of glove mounted human hand
literture.[16] (Figure 6). Smart glove precisely controlled the robotic fingers,
without using any additional control algorithms for grasping
2.4. Application Demo of the Strain Sensor and releasing of objects (Movie S1, Supporting Information).
This proves that current smart glove can be successfully used
The ultralow-detection limit, broad sensing range combined with in virtual reality applications, to track the activities of various
good stretchability, greatly broadens the application of strain human body parts.

Figure 5. Application demo of our sensor for the monitoring of various human activities. a) Sensor’s resistance variation caused due to subtle motion of
blink. Sensor is attached on canthus. b) Resistance variation of strain sensor attached to human throat when the wearer speaks “Liquid,” “Metal,”
“Sensor,” and “NIMTE.” c) Response of sensor during the human heartbeat monitoring. d) Sensor’s resistance change induced by the wrist pulse
during exercise.

Adv. Intell. Syst. 2021, 2000235 2000235 (5 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
www.advancedsciencenews.com www.advintellsyst.com

Figure 6. Application demo of strain sensor. Robotic hand operated from a remote position, by using the various gestures of glove mounted human
hand.

We prepared a smart kneecap by attaching our sensor on it, its great potential to be applied in the field of human–machine
and then used that smart kneecap to record knee’s bending interaction.
motion. During the running motion, sensor’s resistance changed Furthermore, the sensor mounted kneecap was used for the
accordingly with the straightening and bending motion purpose of virtual reality and interactive gaming (Movie S3,
of knee (Figure 7). When knee returns to its original position, Supporting Information). The movie shows the gesture-driven
the resistance also returns to its baseline. More interestingly, video game, operated due to the movement of sensor mounted
our sensor successfully distinguished the repeated bending of kneecap. It is evident from movie that sensor successfully detects
knee joint, with different amplitude and frequency (Figure 7 the minute signals arising from joint motion-induced skin
and Movie S2, Supporting Information), which demonstrates stretch, which proves its full capability to be used in the field
of virtual reality.

3. Conclusion
In this article, LM-based stretchable strain sensor is prepared by
using selective wetting and transferring process, to get improved
compatibility between LM and PDMS substrate. Our sensor
exhibits broad strain sensing range (105%) with ultralow detec-
tion limit (0.05%), minimal hysteresis, high mechanical robust-
ness (8000 cycles), fast response time (58 ms), and excellent
repeatability. Outstanding combinations of LM’s high electrical
conductivity with excellent elastic properties of PDMS are the fac-
tors responsible for superior behavior of our sensor. As LM
exhibits fluidic nature, so it deformed similar to elastic PDMS
substrate without showing any elastic mismatch, and enabled
our sensor to detect minute and larger strains at the same time,
with negligible hysteresis. When demonstrated practically, our
sensor successfully detected various human activities, such as
blink motion monitoring, voice intensity differentiation, and
heartbeat and wrist pulse monitoring. Moreover, the detection
Figure 7. Sensor’s real-time response to amplitude of the human knee of various human gestures by strain sensor illustrates the
motion. After 140 s, the frequency of signal is increased due to fast move- capability of sensor for effective human health and motion
ment of human knee. monitoring, virtual reality, and interactive gaming applications.

Adv. Intell. Syst. 2021, 2000235 2000235 (6 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
www.advancedsciencenews.com www.advintellsyst.com

As practically demonstrated, our sensor can be utilized either as a Data Availability Statement
standalone device or key component in electronic skin and smart
Research data are not shared.
textiles for the applications of HMI, human motion monitoring,
robotics, and beyond.

4. Experimental Section Keywords

Materials and Chemicals: High purity metals gallium, indium, and tin broad range strain sensing, flexible strain sensors, human–machine
(99.99%, Beijing Founde Star Sci. & Technol. Co., Ltd) were mixed interaction, liquid metal, ultralow detection limit
together in the ratio of 67.3:19.2:13.5 by mass. Then, the aforementioned
Received: October 15, 2020
mixture was heated and stirred at 60  C for 30 min to obtain LM Galinstan
(Ga67.3In19.2Sn13.5). Sodium hydroxide (NaOH) was purchased from Revised: December 27, 2020
Sinopharm Chemical Reagent Co., Ltd. Strain gauge-shaped copper foil Published online:
and PDMS (Sylgard 184, Dow Corning, USA) were purchased from
market. Cu foil was already deposited on plastic board and after
cleaning, it was used without any further modification. PDMS was pre- [1] a) J. X. Wang, M. F. Lin, S. Park, P. S. Lee, Mater. Today 2018, 21, 508;
pared in a 10:1 w/w ratio (base to curing agent) and then degassed to b) R. Herbert, J.-H. Kim, Y. S. Kim, H. M. Lee, W.-H. Yeo, Materials
remove air bubbles. 2018, 11, 187; c) H. Wang, X. Ma, Y. Hao, Adv. Mater. Interfaces
Fabrication of Strain Sensors: Strain gauge-shaped Cu foil was cleaned 2017, 4, 1600709.
and submerged in NaOH solution (250 mmol L1) and Galinstan was [2] a) J. A. Rogers, T. Someya, Y. Huang, Science 2010, 327, 1603;
poured on it by using syringe. LM stuck on Cu patterns (copper patterns)
b) Y. Huang, Y. Ding, J. Bian, Y. Su, J. Zhou, Y. Duan, Z. Yin,
due to its hydrophilic behavior toward copper. Afterward, PDMS was
spin coated on the Cu patterns and cured at 70  C for 1 h, followed by Nano Energy 2017, 40, 432.
peeling it with transfer printing machine. Finally, the package layer of [3] M. Amjadi, K.-U. Kyung, I. Park, M. Sitti, Adv. Funct. Mater. 2016,
PDMS was deposited to complete the formation of sensor. Prior to the 26, 1678.
deposition of PDMS package layer, square-shaped flexible copper electro- [4] M. Amjadi, A. Pichitpajongkit, S. Lee, S. Ryu, I. Park, ACS Nano 2014,
des were attached with LM, which later served as electrodes. During 8, 5154.
manufacturing of sensor, deposited PDMS layers were cured in an electric [5] a) G. Cai, J. Wang, K. Qian, J. Chen, S. Li, P. S. Lee, Adv. Sci. 2017, 4,
oven. The final size of sensor was 3.5 cm  1 cm  1 mm and dimensions 1600190; b) S. Gong, D. T. H. Lai, Y. Wang, L. W. Yap, K. J. Si,
of patterned LM include, 3 cm length, 300 μm line width, and 30 μm Q. Q. Shi, N. N. Jason, T. Sridhar, H. Uddin, W. L. Cheng, ACS
thickness. Appl. Mater. Interfaces 2015, 7, 19700; c) O. Kanoun, C. Muller,
Characterization and Performance Testing: Mechanical tests were
A. Benchirouf, A. Sanli, T. N. Dinh, A. Al-Hamry, L. Bu,
conducted by using computer-controlled material testing machine
C. Gerlach, A. Bouhamed, Sensors 2014, 14, 10042.
(Instron 5943, USA). Electromechanical tests were performed at room
temperature by using a two-probe configuration. The electrical current [6] D. J. Lipomi, M. Vosgueritchian, B. C. K. Tee, S. L. Hellstrom, J. A. Lee,
was provided by current source device (Keithley 237, Keithley C. H. Fox, Z. Bao, Nat. Nanotechnol. 2011, 6, 788.
Instruments, USA), whereas the voltage was measured by voltmeter [7] D. Kang, P. V. Pikhitsa, Y. W. Choi, C. Lee, S. S. Shin, L. Piao, B. Park,
(Keithley 6517A, Keithley Instruments, USA). K.-Y. Suh, T.-I. Kim, M. Choi, Nature 2014, 516, 222.
The experiments involving human subjects have been performed with [8] L. J. Lu, Y. J. Zhou, J. Pan, T. Q. Chen, Y. J. Hu, G. Q. Zheng, K. Dai,
the full, informed consent of the volunteers. C. T. Liu, C. Y. Shen, X. M. Sun, H. S. Peng, ACS Appl. Mater. Interfaces
2019, 11, 4345.
[9] M. D. Dickey, Adv. Mater. 2017, 29, 1606425.
Supporting Information [10] Y.-H. Wu, R.-M. Zhen, H.-Z. Liu, S.-Q. Liu, Z.-F. Deng, P.-P. Wang,
S. Chen, L. Liu, J. Mater. Chem. C 2017, 5, 12483.
Supporting Information is available from the Wiley Online Library or from
[11] a) N. Kazem, T. Hellebrekers, C. Majidi, Adv. Mater. 2017, 29,
the author.
1605985; b) C. Pan, E. J. Markvicka, M. H. Malakooti, J. Yan,
L. Hu, K. Matyjaszewski, C. Majidi, Adv. Mater. 2019, 31, e1900663.
[12] Y. Jiang, S. Su, H. Peng, H. S. Kwok, X. Zhou, S. Chen, J. Mater. Chem.
Acknowledgements C 2017, 5, 12378.
Y.W. and Y.Z. contributed equally to this work. This research was partially [13] S. G. Yoon, H. J. Koo, S. T. Chang, ACS Appl. Mater. Interfaces 2015,
supported by the National Natural Foundation of China (61704177, 7, 27562.
61774161, 51525103, 51971233, 51701231, and 51931011), National [14] a) S. Chen, Y. Wei, S. Wei, Y. Lin, L. Liu, ACS Appl. Mater. Interfaces
Key Technologies R&D Program of China (2016YFA0201102), the 2016, 8, 25563; b) L. Lu, X. Wei, Y. Zhang, G. Zheng, K. Dai, C. Liu,
External Cooperation Program of Chinese Academy of Sciences C. Shen, J. Mater. Chem. C 2017, 5, 7035; c) J. Lee, S. Kim, J. Lee,
(174433KYSB20190038), Chinese Academy of Sciences Youth D. Yang, B. C. Park, S. Ryu, I. Park, Nanoscale 2014, 6, 11932;
Innovation Promotion Association (2018334), CAS President’s
d) M. Amjadi, M. Turan, C. P. Clementson, M. Sitti, ACS Appl.
International Fellowship Initiative (PIFI) (2019PE0019), Public Welfare
Mater. Interfaces 2016, 8, 5618; e) Y.-H. Wu, H.-Z. Liu, S. Chen,
Technical Applied Research Project of Zhejiang Province
(LGG19F010006), and Ningbo Scientific and Technological Innovation X.-C. Dong, P.-P. Wang, S.-Q. Liu, Y. Lin, Y. Wei, L. Liu, ACS Appl.
2025 Major Project (2018B10057, 2019B10127). Mater. Interfaces 2017, 9, 20098; f ) J. Y. Kim, S. Ji, S. Jung,
B.-H. Ryu, H.-S. Kim, S. S. Lee, Y. Choi, S. Jeong, Nanoscale 2017,
9, 11035; g) Q. Liu, J. Chen, Y. Li, G. Shi, ACS Nano 2016, 10,
7901; h) E. Roh, B.-U. Hwang, D. Kim, B.-Y. Kim, N.-E. Lee, ACS
Conflict of Interest Nano 2015, 9, 6252; i) Y. Cheng, R. Wang, J. Sun, L. Gao, Adv.
The authors declare no conflict of interest. Mater. 2015, 27, 7365.

Adv. Intell. Syst. 2021, 2000235 2000235 (7 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH
www.advancedsciencenews.com www.advintellsyst.com

[15] T. Yamada, Y. Hayamizu, Y. Yamamoto, Y. Yomogida, A. Izadi- S.-H. Jung, J. H. Jung, H. J. Sung, H. W. Lee, H. M. Lee, ACS
Najafabadi, D. N. Futaba, K. Hata, Nat. Nanotechnol. 2011, Appl. Mater. Interfaces 2017, 9, 1770; c) Y. Zhou, Y. Wu,
6, 296. W. Asghar, J. Ding, X. Su, S. Li, F. Li, Z. Yu, J. Shang, Y. Liu,
[16] a) J. Chen, J. Zhang, Z. Luo, J. Zhang, L. Li, Y. Su, X. Gao, Y. Li, R.-W. Li, ACS Appl. Electron. Mater. 2019, 1, 1866; d) P. Cao,
W. Tang, C. Cao, Q. Liu, L. Wang, H. Li, ACS Appl. Mater. Y. Liu, W. Asghar, C. Hu, F. Li, Y. Wu, Y. Li, Z. Yu, S. Li, J. Shang,
Interfaces 2020, 12, 22200; b) D. Y. Choi, M. H. Kim, Y. S. Oh, X. Liu, R.-W. Li, Adv. Eng. Mater. 2020, 22, 1901239.

Adv. Intell. Syst. 2021, 2000235 2000235 (8 of 8) © 2021 The Authors. Advanced Intelligent Systems published by Wiley-VCH GmbH