Technology Roadmap for Flexible Sensors

A full list of authors and affiliations appears at the end of the article.

Abstract

Humans rely increasingly on sensors to address grand challenges and to improve quality of life

in the era of digitalization and big data. For ubiquitous sensing, flexible sensors are developed

to overcome the limitations of conventional rigid counterparts. Despite rapid advancement in

bench-side research over the last decade, the market adoption of flexible sensors remains

limited. To ease and to expedite their deployment, here, we identify bottlenecks hindering the

maturation of flexible sensors and propose promising solutions. We first analyze challenges in

achieving satisfactory sensing performance for real-world applications and then summarize issues

in compatible sensor-biology interfaces, followed by brief discussions on powering and connecting

sensor networks. Issues en route to commercialization and for sustainable growth of the sector

are also analyzed, highlighting environmental concerns and emphasizing nontechnical issues such

as business, regulatory, and ethical considerations. Additionally, we look at future intelligent

flexible sensors. In proposing a comprehensive roadmap, we hope to steer research efforts towards

common goals and to guide coordinated development strategies from disparate communities.

Through such collaborative efforts, scientific breakthroughs can be made sooner and capitalized

for the betterment of humanity.

Graphical Abstract

Corresponding Author Xiaodong Chen

– Innovative Center for Flexible Devices (iFLEX), Max Planck-NTU Joint Laboratory for

Artificial Senses, School of Materials Science and Engineering, Nanyang Technological University, Singapore 639798, Singapore;

Institute of Materials Research and Engineering (IMRE), Agency for Science, Technology and Research (A*STAR), Singapore

138634, Republic of Singapore; chenxd@ntu.edu.sg.

The authors declare the following competing financial interest(s): A.M.A. and P.S.W. have a number of patent applications related to

the technologies described in this article.

HHS Public Access

Author manuscript

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Published in final edited form as:

ACS Nano

. 2023 March 28; 17(6): 5211–5295. doi:10.1021/acsnano.2c12606.

Author Manuscript

Keywords

soft materials; mechanics engineering; flexible electronics; conformable sensors; bioelectronics;

human-machine interfaces; body area sensor networks; technology translation; sustainable

electronics

Living things are equipped with biological sensory systems for light, sound, smell,

etc

. to

monitor and to adapt to the environment. In addition to the natural senses, humans use

synthetic, fabricated sensors—devices that allow users to measure the values of physical and

psychological conditions of interest using the inherent physical properties of the sensors

—

to augment our natural abilities of perceiving the world, enabling us to interact with the

environment and to improve our living conditions.

Amongst the first documented sensors in human history are the thermoscope by Philo of

Byzantium in the 3

century BCE for temperature change detection

and the seismoscope

by Zhang Heng in 132, used to detect the occurrence of earthquakes and approximate their

directions.

Sensors in this early era (what we define as Sensors 1.0, Figure 1) convert

physical quantities/events to mechanical outputs that are easily observable. Later, with the

discovery of electricity and the invention of electric generators, sensors were designed

to convert physical parameters to electric signals, enabling control function. For instance,

the electric tele-thermoscope invented by Warren Johnson in 1883 could not only monitor

temperature, but could also modulate the function of an automatic temperature control

system.

This marked the era of Sensors 2.0. Moving to Sensors 3.0, the electronics industry

promoted the miniaturization and integration of sensors with other electronic components,

giving birth to smart devices such as smartphones and smartwatches, where dozens of

sensors collectively provide an impressive user experience. In recent years, advances in

the Internet of Things (IoT), Industry 4.0, big data, artificial intelligence (AI), robotics,

and digital health

have prompted sensors to become more connected and intelligent,

entering Sensors 4.0. For instance, a large number and variety of sensors are embedded in

autonomous vehicles with wireless connectivity for adaptive self-driving, and connected IoT

sensors integrated with AI provide effective solutions to energy management of buildings

and industrial facilities.

Sensors that translate the physical world into data serve a foundational role in the era of

digital transformation. In the digital era, connectivity and decision-making rely heavily on

high-quality big data—any data bias or inaccuracy may lead to distorted conclusions and/or

incorrect decisions, and the consequences can be catastrophic.

Therefore, it is critical to

develop sensors that can acquire accurate and reliable data at large scales. This capability

would potentially expedite solutions to the grand challenges facing humanity,

such as

aging populations,

infectious diseases,

food security,

energy crises,

climate

and environment crises,

and would improve our quality of life. For example, sensors

could be employed to test patients for common diseases and even to predict them, to detect

bacterial growth in every food package, or to monitor pollution in every lake, stream, and

river.

Luo et al.

Page 2

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

However, conventional sensor technology is usually incapable of such massive-scale

ubiquitous monitoring. Being highly integrated and miniaturized, modern sensors serve

adequately as the components of smart electronics/machines, but their small and rigid

form factors restrict their usage in many applications such as healthcare wearables,

interactive robots, smart packaging, and building-integrated electronics, where flexible

sensors would enable advances (Table 1). In this Review, we define flexible sensors

broadly to include all types of sensors that can withstand mechanical deformation (>10 m

−1

bending curvature or >1% strain on a device/system) without device failure or significant

alteration in sensing performance. We include bendable, rollable, foldable, stretchable,

twistable, and conformable sensors.

Flexible sensors enable measurements on dynamic

and/or shape-changing objects and large-area non-flat surfaces,

due to their mechanical

flexibility and stretchability, shape adaptability, and fabrication scalability, with which rigid

sensors typically struggle. Flexible sensors are lightweight, thanks to the use of organic

materials and/or thin-film form factors, benefiting integration, distribution, and application.

Furthermore, some flexible sensors can be manufactured using low-cost materials and

large-scale processes such as printing,

making mass deployment economically viable.

Importantly, the use of organic materials, the thin-film form factor, and the additive

manufacturing of flexible sensors may provide more environmentally sustainable ways of

sensor production and disposal, tackling the escalating electronic waste problem.

The above features make flexible sensors well positioned for applications that have

demanding requirements in mechanical compliance, integration density and scale,

manufacturability, and cost. For example, the physiological parameters that current wearable

sensing technologies (

e.g.

, continuous glucose monitors, smartwatches) can measure are

limited.

Medicalgrade measurements of electrocardiogram and sweat metabolites require

conformal contact between the sensor and the skin, but this goal is hardly achievable by

rigid sensors without causing discomfort due to the surface micro-texture and deformation

experienced by the skin.

In contrast, soft and stretchable sensors can address these issues,

offering disruptive solutions for future healthcare.

For robots to interact safely with

humans,

high densities of sensors (>10 sensors per cm

to be comparable to human

skin

) would need to cover the curved robotic surface over large areas (~m

In this

regard, monolithically manufacturing sensor arrays on flexible substrates is much more

efficient than individually placing rigid sensor pixels.

Furthermore, smart packaging

with embedded sensors for product tracking and quality monitoring will be critical to

efficient and sustainable supply chains,

yet sensor stickers need to be manufactured at

costs as low as a few cents to realize this large-scale (and often disposable) application.

Likewise, to install sensing maps in industrial facilities such as pipes, walls, and floors,

low-cost and large-area manufacturability is pivotal. To this end, printed sensors using

solution-processable organic/carbon materials provide the most promising solutions. In all,

the advantages offered by flexible sensors target the issues of data quantity and quality in the

digital era, which will make flexible sensors one of the most valuable players in Sensors 4.0.

Flexible sensors have matured tremendously since the beginning of the 21

century. Starting

with pressure sensor arrays on plastic films,

flexible sensors now cover a wide range of

physical and chemical sensing modalities, including temperature, strain, electrophysiology,

ions, biomarkers, metabolites, gases, and many more.

Substrates are not limited to

Luo et al.

Page 3

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

plastic films but can also be ultra-thin plastic foils,

porous polymer mats/meshes,

paper,

elastomers, and hydrogels.

From single sensors to sensor arrays, from stand-

alone sensors to integrated sensing systems,

the development has been astonishing, and

fascinating applications including sweat-based stress monitoring

and remote robotic

control through skin sensor-enriched virtual reality

have been demonstrated.

Despite significant research achievements, the adoption of and market for flexible sensors

often falls short of predicted levels.

Some flexible sensors still have a long way to go

to meet the stringent demands posed by real-world problems. It is therefore timely to

identify the bottlenecks hindering flexible sensor deployment, not only technical challenges,

but also cultural and regulatory hurdles (Figure 2, left). Here, we summarize promising

on-going efforts to address these challenges and propose further plausible solutions. In

doing so, we hope to steer and to accelerate research efforts towards faster translation of

laboratory innovations and prototypes into widely used products. Additionally, we anticipate

long-term issues facing future sensor deployment (Figure 2, right). Early awareness of

these issues will prompt crafting and development of effective solutions. We conclude by

predicting what future intelligent flexible sensors will do. These challenges and prospects

are summarized into a comprehensive roadmap, in the hope of guiding collective and

cooperative development strategies towards common goals by the research community and

beyond.

SENSING PERFORMANCE

Sensing performance is of paramount importance for any sensor. Performance of flexible

sensors encompasses the basic metrics that apply to both rigid and flexible sensors,

including the classical 3S’s (stability, selectivity, and sensitivity), as well as aspects

unique to flexible sensors, including tolerance to mechanical deformation and monolithic

integration into large-area sensing arrays. We identify key issues in these three aspects: basic

metrics, mechanical performance, and array performance (Figure 3), and provide detailed

discussions on existing solutions and research gaps.

Basic Performance Metrics Comparable to Rigid Sensors.

Stability, selectivity, and sensitivity are primary metrics used to assess sensor performance.

Because of the materials, manufacturing techniques, and sensing mechanisms used in

flexible sensors, their performance often falls short of rigid sensors, even when no

mechanical deformation is involved. Here, we discuss some most prominent issues in

the 3S’s—stability being the most challenging and pressing problem for real-world

applications, particularly when trying to achieve long-term sensing. We also briefly

discuss considerations other than the 3S’s, including the dynamic responses of mechanical

sensors and sensing capabilities enabled by wearable biosensors. We summarize sensor

performance by emphasizing holistic approaches to sensor accuracy and fundamental

correlation-establishing studies.

Stability in Harsh Environments Despite the Use of Unstable Materials.—

Stability is essential for deployable sensors because it ensures repeatable and reliable usage

in changing environments, especially for long-term monitoring. Here two dimensions are

Luo et al.

Page 4

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

considered: time and stress (

e.g.

, temperature, humidity).

The challenge of stability for

flexible sensors often stems from the use of organic and polymeric materials in their

fabrication, which tend to degrade over time and whose properties are easily altered by

environmental factors. Furthermore, wearable biosensors incorporating biological receptors

face additional bio-instability. The challenge is exacerbated by the degradative environments

flexile sensors are exposed to, such as

in vivo

tissues and biofluids, the deep sea, and high

altitudes, where extreme physicochemical stresses are present.

The most straightforward approach to tackling the stability challenge is to improve the

environmental stability of sensor materials. Engineering conventional rigid sensor materials

into flexible and stretchable form factors is one effective approach, but the fabrication

complexity can limit scalability and cost effectiveness. A special case of this strategy is

the recently discovered giant magnetoelastic effect in soft systems for pressure sensing,

where inorganic magnetic particles are embedded in elastomers to induce changes in

magnetic fields under deformation.

The sensing materials and mechanism employed

are intrinsically waterproof and environmentally stable.

While materials synthesis and

modification towards environmentally stable organics

and other emerging materials

(

e.g.

, perovskites)

are a constant pursuit, this strategy is fundamentally challenging,

governed by the physicochemical nature of these materials.

Therefore, when no direct contact between the stimulus and the sensing material is required

(in mechanical, temperature, and light sensors, for example), a more viable approach is

to apply protective layers to sensitive materials and the entire device.

To this end, high-

performance humidity and oxygen barrier materials are in great demand. However, flexible

materials themselves usually have poor barrier properties (Figure 4) due to the intrinsic

free volume in polymers and defects in inorganics,

and barrier properties degrade with

repeated mechanical deformation. Such issues are most severe for the elastic packaging of

mechanical sensors and other stretchable sensors, when both elasticity and barrier properties

are required.

Adding thin-film coatings

and filler additives to packaging materials are

two frequently employed strategies,

but there remains a lack of effective methods to

improve the barrier properties of elastic substrates and packaging layers.

In addition to

specialized barrier layers, effective sealing is also critical for devices made of fluidic or

liquid-containing materials (

e.g.

, liquid metal, electrolyte, hydrogel) and devices used in

liquid environments (

e.g., in vivo

biological tissues,

sea-water

Encapsulation, and packaging in general, are of paramount importance to flexible sensors,

especially wearable sensors. Proper packaging enables users to wear devices over extended

periods with minimal noise or signal drift. This issue is often ignored by the academic

community and is more typically considered by industry. However, packaging must be

addressed in order to obtain meaningful on-body data in population studies.

Temperature is an environmental stress affecting almost all flexible sensors because most

sensing materials, including rigid ones, are sensitive to temperature changes.

This problem

cannot be solved simply through material optimization or encapsulation. Introducing

additional compensation elements such as temperature sensors and feedback circuits

is generally more effective. Alternatively, exploring sensing mechanisms that circumvent the

Luo et al.

Page 5

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

temperature-sensitive aspects of the sensing material

is viable, although not as a general

solution. Overall, to suppress temperature effects with simple device structures and high

integration remains a challenge.

Stability is a significant challenge for flexible chemical sensors, especially wearable

biosensors,

where bio-fouling and bioreceptor inactivation are two major factors that

affect long-term (several days) sensor performance. The fouling layer strongly influences

the selectivity and binding affinity of biorecognition events and results in strong background

signals as well as poor signal-to-noise ratios. One of the most commonly used strategies

to combat biofouling is drop-casting protective polymeric membranes such as Nafion

and chitosan.

Other anti-fouling coatings such as bovine serum albumin (BSA) and

poly-(ethylene glycol) are also effective.

However, a drawback of this surface-coating

strategy is the reduction or blockade of electronic conduction between the biorecognition

moiety and the transducing electrode. To tackle this problem, three-dimensional (3D)

nanocomposites composed of anti-fouling agents (

e.g.

, BSA) and conductive materials

(

e.g.

, gold nanowires, carbon nanotubes, CNTs) have been engineered.

Alternatively,

surface roughness and wettability control can also circumvent this problem. For example,

nanoporous Au electrodes minimized fouling by slowing down mass transport while

allowing efficient small-molecule exchange.

Insights from skin biology, such as materials

chemistry and surface texture, may provide inspirations to tackle biofouling.

Low stability of immobilized bioreceptors in the uncontrolled conditions of wearable

applications (

e.g.

, changing temperature and pH) is another issue. Bioreceptors such as

enzymes can easily detach from anchoring substrates/electrodes in fluidic environments

(and even more so if mechanical deformation is involved) and lose their recognition

function outside their operational windows.

To improve the long-term stability of

enzymatic sensors, a nanoporous membrane with effective enzyme immobilization was

robustly anchored to nanotextured electrodes, achieving continuous glucose sensing with

minimal signal drift for up to 20 h.

Encapsulation of enzymes within electrodes through

a monolithic 3D printing process is another way to improve stability.

Alternatively,

nanozymes,

i.e.

, artificial enzymes made of nanomaterials, can be used,

though

often-times compromising selectivity. Molecularly imprinted polymers (MIPs), known as

“artificial receptors”, can also overcome the stability issue of bioreceptors while achieving

good selectivity.

Besides biofouling and bioreceptor instability, there are other issues impairing biosensor

stability. For example, many reported electrochemical biosensors utilize a mediator layer

to reduce the potential required to trigger redox reactions for reduced interference from

other electroactive molecules,

yet the Faradaic signal could decay over time, limiting

long-term reliability. Furthermore, charge accumulation on electrode surfaces or material

interlayers can lead to signal drift, which can be mitigated by nanotextured electrodes

with larger surface areas and more robust bonding with sensing layers.

Moreover, the

interactions between the active layers and biomarkers may alter the surrounding electric

field, introducing microenvironmental changes as an interfering factor.

Luo et al.

Page 6

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

Signal drift over a relevant period of operation is an issue not only for biosensors, but

also may be the number one challenge for any sensor technology. However, this issue is

often ignored by the academic community outside of electrical engineering. The magnitude

and predictability of signal drift often determine the lowest concentration that a sensor can

accurately report over its lifetime. Flexible sensors often suffer from much larger signal

drifts than their rigid counterparts, which effectively leads to high noise levels. In this

regard, it is important for the community to report signal drift and its measurement carefully

and precisely, and to understand the exact cause for each emerging sensor technology so

that the sensor drift can be tackled effectively. For example, by designing better sensor

architectures or customizing compensation algorithms and driving circuits, sensor drift can

be reduced.

Stability is central to sensor practicality, yet it is often neglected in academic research. We

urge the research community to place more emphasis on stability to push flexible sensors

closer to commercialization. When long-term stability is not achievable, making sensors at a

low cost such that they can be frequently replaced and disposed of might be another viable

route to mass adoption.

Selectivity to Complex Mechanical and Chemical Stimuli.—

Selectivity refers

to the ability of a sensor to discriminate between the analyte of interest and possible

interferences.

It was originally defined for chemical

100

and biological

101

sensors but may

be extended to include mechanical sensors (

e.g.

, pressure sensors, strain sensors, torsion

sensors,

etc.

). In real application scenarios, a wide range of chemical species and mechanical

forces are usually present simultaneously, and they interact with sensing materials through

similar mechanisms, thus producing ambiguous sensor responses.

There are two general approaches to sensor selectivity: specific sensors and selective sensor

arrays.

101

Ideally, a specific sensor only responds to one analyte, and an array of such

sensors would tell the exact composition of a mixture without needing a great deal of data

analysis. Such specific sensors are often hard to realize. In contrast, in a selective sensor

array, each sensor responds to a collection of analytes differently, and the array response

collectively produces a fingerprint for a mixture. With proper data analysis, the mixture

composition can be accessed. These two principles are widely applied for mechanical

sensors, biosensors, and gas sensors.

Mechanical force applied on a sensor is often a mixture of pressure, tension, shear, and

torsion. Decoupling these modes is important for gesture recognition, robotic control, and

prosthetics. There have been attempts to fabricate ‘specific’ mechanical sensors.

102

106

For instance, a stiff and anisotropically resistive material was structured into micro-

meanders and encapsulated in elastic films such that the sensor was responsive to only one

direction of tensile strain and insensitive to bending and twisting.

102

Stiff platforms were

embedded underneath pyramid microstructures for pressure sensors to achieve undisturbed

performance at up to 50% tensile strain.

104

Although the insensitivity to other mechanical

stimuli of these ‘specific’ sensors is not ideal due to materials and geometric limits, their

performance is sufficient for non-critical applications or large values of a strain of interest

(

e.g.

, joint movements). ‘Specific’ sensors have been integrated to achieve multi-modal

Luo et al.

Page 7

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

mechanical sensing, where careful mechanical design is needed to isolate and distribute

different mechanical stimuli to the desired sensors so that each deformation can be sensed

independently.

107

109

The other ‘selective sensor array’ principle takes many forms for mechanical sensors. The

simplistic implementation is to fabricate deformable sensing materials

110

111

and/or design

3D sensor structures

112

116

to make the response curves different for different forms of

deformation. With proper signal analyses, the correct deformation can be identified. A

similar method is to integrate multiple sensors on a miniaturized 3D structure

117

or a two-

dimensional (2D) surface.

118

119

The response of individual sensors differs according to the

stress applied; holistic analyses of all sensor outputs derive the deformations experienced.

However, using the above methods, it may be difficult to decouple a simultaneous

combination of deformations (

e.g.

, compression plus shear) because the signals overlap.

Advanced algorithms, such as machine learning, might be able to solve this problem. A third

approach is to use materials or devices that are sensitive to several stimuli, but the stimuli

can be distinguished by different measurement modes (

e.g.

, resistance and capacitance).

111

Readout electronics will be more complex for integrated devices using this strategy, which

further increases system-level power consumption and hardware cost in parallel.

Biosensors are used to analyze complex mixtures present in biological samples, which

may contain ions, small molecules (metabolites, cytokines, lipids, neurotransmitters,

etc.

macromolecules (peptides, proteins, nucleic acids,

etc

.), and even viruses, bacteria, and

cells. Selectivity becomes critically important in analyzing such complex mixtures as closely

related interferents (

e.g.

, biological precursors and metabolites) are often present.

120

In this

regard, nature provides many biorecognition elements that offer high specificity through

interactions with metabolites and biomarkers. The utilization of bioaffinity-based receptors,

including ionophores, DNAs/RNAs, aptamers, and antibodies on flexible biosensors allows

selective

in situ

target recognition,

although sometimes at the cost of complicated

fabrication and handling, as well as relatively poor stability. In this regard, artificial

bioreceptors such as MIPs offer a more stable and easily processible option without

sacrificing binding specificity in some cases.

Effective transduction mechanisms that

transform the receptor-target binding to measurable electrical or optical signals are

critical. Some promising examples include aptamer-functionalized field-effect transistors,

120

molecular pendulum-based biomolecular sensors,

121

as well as redox probe-tagged

electrochemical aptasensors

122

and MIP-based sensors.

More complex biorecognition

elements, including cell membranes

123

and whole cells

124

126

provide improved specificity

towards some analytes, but this advantage comes with increased fabrication complexity and

storage requirements. For biosensors that do not rely on bioaffinity for sensing, careful

engineering of catalytic nanomaterials can achieve desirable selectivity in the (electro)-

chemical recognition of some analytes.

127

129

While biosensors are most often used for biofluid analyses, they can also be engineered to

detect airborne pathogens

130

and biologically relevant gases. Gas sensors are an emerging

field for flexible sensors. They provide noninvasive means of biomarker detection to inform

metabolic processes and disease progression in humans and plants,

131

137

and are thus

attractive for real-time health monitoring and point-of-care diagnostics (Figure 5a).

138

Luo et al.

Page 8

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

Biomarker volatile organic compounds (VOCs) present in complex mixtures (often with

more than a dozen of components) and a complete profile is often required for the

determination of physiological status.

133

142

Some VOCs have similar molecular structures,

making specific sensing challenging. Although there have been attempts to utilize biological

olfactory elements, such as olfactory receptors (ORs), olfactory cells, and even olfactory

organs,

143

as well as other biomolecules (

e.g.

, enzymes, antibodies, aptamers) as the

recognition moieties (Figure 5a, left),

144

145

insufficient understanding of biological

olfactory systems poses fundamental challenges for bioaffinity-based gas sensors. For

example, the pairing relationships and the binding/unbinding interactions between gas

species and ORs are largely unknown.

145

Other factors impeding the development of

bioaffinity gas sensors include design complexity for liquid-phase reactions and the high

cost and low stability associated with bioreagents, given that gas sensors are currently

primarily used for industrial and environmental applications.

The growing healthcare/medical gas sensors area

133

may provide an impetus to continue to

develop bioaffinity-based sensing.

144

In comparison, nanomaterials with tunable structures

and chemistries capable of dry-phase sensing seem to be more technically and economically

viable.

146

147

Metal–organic frameworks (MOFs) are particularly attractive because their

porous structures can

selectively

adsorb or filter gas molecules (Figure 5b).

148

However,

a limited understanding of gas–MOF interactions, as well as the structure–property

relationships of MOFs prevents generalized design methodologies for MOF-based gas

sensors to cover wide ranges of VOCs.

In contrast to specific VOC sensors, selective sensing arrays are more widely used to

recognize gas mixtures (Figure 5a, right). Combined with machine learning, this strategy

has seen commercial success in electronic noses.

Nanomaterials are also a go-to option

for selective sensing arrays due to their high sensitivity and ease of tuning surface

interactions.

149

For instance, graphene functionalized with various ligands and coupled

with Au nanoparticles was used to construct an 8-sensor array that could classify 13

individual plant VOCs at >97% classification accuracy (Figure 5c).

142

A recent approach

achieved the fabrication and utilization of an array of 108 graphene-based sensors

functionalized with 36 chemical receptors for the discrimination of 6 gas species within

a minute,

150

shedding light on rapid VOC detection using largescale sensor arrays. Overall,

recent advances in flexible room-temperature gas sensor arrays have achieved lower

power consumption, reduced fabrication cost, and greater wearability without sacrificing

sensing performance.

142

151

155

Although machine learning algorithms capable of higher

prediction accuracy can compensate for sensor selectivity short-falls,

151

156

157

improving

the specificity of each sensor remains a critical challenge.

An interesting application of selective array sensing was recently reported for

triboelectricity-based material identification.

158

An array of sensors with differential

triboelectric properties generated a fingerprint signal pattern when in contact with

a particular material. Combined with machine learning, the accuracy for materials

classification reached 97% when four sensors in an array were used. Such strategies may

find wider application in flexible sensors to enable more sophisticated sensing capabilities.

Luo et al.

Page 9

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

Sensitivity with Wide-Range Linearity and to Low-Concentration Analytes.—

High sensitivity allows sensors to detect minute changes in a stimulus, to reduce false-

negative signals, and to improve signal-to-noise ratio and accuracy. The sensitivity of most

flexible physical sensors (

e.g.

, mechanical sensors, temperature sensors, photodetectors) is

sufficient for common applications. A notable issue is the trade-off between sensitivity and

sensing range in mechanical sensors. In comparison, sensitivity is more of a concern for

chemical sensors, specifically biosensors that detect low concentrations of analytes present

in biofluids.

The trade-off between sensitivity and sensing range, and the issue of nonlinearity exist in

most mechanical sensors,

111

159

163

and are especially prominent for pressure sensors.

164

Ideally, high sensitivity across a wide force/pressure range is desirable, but is hardly

achievable in bulk piezoresistive/piezocapacitive sensors, because of the stiffening effect

of soft materials upon compression. Microstructuring is a common strategy to improve

sensitivity,

165

166

yet this approach mostly works at low pressures. There have been many

attempts to address this problem. Structure-wise, intrafillable microstructures accommodate

deformed surface structures in the underlying undercuts and grooves, thereby retarding

the saturation of porous structures.

167

Mechanism-wise, combined piezoresistivity and

piezocapacitivity significantly increase sensitivity, even at large stress of up to 50 kPa.

168

The magnetoelastic effect is useful for pressure sensing over a wide range, from 3.5 Pa to

2000 kPa,

169

and its sensitivity is comparable to those of piezoresistive and piezocapacitive

sensors.

The above methods do not solve the nonlinearity problem. One solution is using hierarchical

microstructures, such as micropillars on hemisphere arrays.

170

Adding gradient charge

distribution within the active material may be able to solve the nonlinearity issue. This

strategy has been demonstrated in a capacitive pressure sensor, reaching a record-high

linearity range up to 1000 kPa. The mechanism is gradient compressibility and dielectric

property with increasing pressure, realized by a skin-like hierarchical microstructure made

of materials of different permittivities.

171

This strategy may be extended to other types of

pressure sensors based on gradient conductivity or gradient ionogels. Another perspective on

addressing this sensitivity-range conflict is to program the sensor performance on demand

based on application requirements, since extraordinarily high sensitivity is usually required

for small pressure detection, whereas for large pressure, a wide sensing range is more

important. A stiffness memory ionogel was developed,

172

whose stiffness could be tuned by

pressure plus thermal treatment. The programmable stiffness led to programmable pressure

ranges, detection limits, and sensitivity. Although an interesting concept, the practical

applicability of such customizable sensors should be carefully evaluated, taking account

of reproducibility, calibration,

etc

Generally, for mechanical sensors and other sensors involving mechanics sensing (

e.g.

vibration sensors, ultrasound imagers

173

), sensitivity–deformability entails a balance of rigid

and soft materials in rationally designed structures—rigid materials usually lead to good

sensitivity, whereas soft materials enable large deformability. Integration density, system

complexity, and manufacturability are key factors to consider when devising wide-range

sensitive systems.

Luo et al.

Page 10

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

Highly sensitive wearable and implantable biosensors are strongly desired for on-body

and in-body chemical sensing to aid diagnostics and therapeutics, but this technology is

relatively underdeveloped. Currently reported biosensors primarily focus on biomarkers

at the levels of tens of

M or higher.

174

176

There are a number of clinically relevant

biomarkers such as proteins, peptides, hormones, small molecules, and drugs existing

in sweat or saliva at nanomolar levels and lower.

177

To enable the detection of these

biomarkers, the sensitivity of flexible biosensors needs improvement.

Various nanomaterials such as conducting polymer nano-fibers,

178

graphene,

179

nanostructured gold,

180

MOFs,

181

and transition metal nanoparticles (

e.g.

, Fe

and

NiO)

127

are often utilized on the working electrode in electrochemical sensors as they can

enhance the electrochemically active surface area and electron transfer dynamics, resulting

in higher detection signals.

176

Recent reports show that laser-engraved graphene enabled

the detection of sweat uric acid, tyrosine, and cortisol at sub-micromolar levels,

182

and dendritic gold nanostructures were successfully used to monitor micromolar levels

of vitamin C and glucose in sweat.

183

Besides nanomaterials, micro- to macro-scale

approaches can also increase electroactive surface areas, using printable ink formulations

and 3D hybrid electrode structures.

184

185

Signal transduction is important to sensitivity—effective transduction can amplify

binding events to reach measurable signals. Transistors, including field-effect transistors

(FETs)

120

186

189

and organic electrochemical transistors (OECTs),

190

192

are effective

amplification devices.

191

193

194

When the channel of a FET is reduced to the nanoscale,

the high surface-to-volume ratio enables highly sensitive detection.

195

By employing this

mechanism with an aptamer, cortisol at a concentration down to 1 pM could be selectively

detected.

In addition, reducing the molecular size of surface-bound bioreceptors, such as

using oligonucleotides in place of DNAs

187

and nanobodies in place of antibodies,

192

can

bring the target-binding event closer to the transducer and may therefore enhance sensitivity.

This consideration can also be useful in the design and selection of aptamers, to ensure

that significant conformational changes in the artificial receptor occur close to the surface

so as to gate the FET channel optimally.

120

Successful engineering of peptides

196

and

DNA

197

into semiconductors may allow the unification of analyte binding, transduction,

and amplification in a single material, offering improvement in sensitivity and response

time. Devices capable of effective amplification should be explored further for wearable

biosensors. For example, subthreshold Schottky-barrier thin-film transistors demonstrate

exceptional intrinsic gain of up to 1,100 V V

−1

198

Schottky-contacted nanowire sensors

were found to enhance the sensitivity of Ohmiccontacted sensors to light, gas, and

(bio)chemicals by orders of magnitude.

199

Colorimetric biosensors are attractive due to low cost, simplicity, and automated operation,

but their poor sensitivities call for effective signal-amplifying mechanisms. Fluorescent

biosensors could be a good alternative as fluorescence can boost sensitivity by up to

1000× that of colorimetry.

200

Nanocatalysts are also promising, previously achieving

100× amplification in antibody-based lateral flow immunoassays.

201

Careful design of

the catalytic inorganic nanoparticles with organic recognition moieties is critical in

achieving desirable sensing performance. Nevertheless, current methods for colorimetric

Luo et al.

Page 11

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

and fluorometric signal detection by the naked eye, in-built detectors, or external cameras

suffer from drawbacks such as subjectiveness, device bulkiness, and manual operation.

Simple methods to quantify colorimetric and fluorometric signals digitally from wearable

biosensors are needed.

Another potential way to enhance the sensitivity of biosensors is the preconcentration of

target analytes through ion concentration polarization

202

or dielectrophoresis.

203

Target

preconcentration has been used for wearable real-time monitoring of low-level heavy metals

in sweat.

204

205

A further strategy being explored is to amplify signals using low-noise and

high-gain circuits, such as differential amplifiers and charge-coupled devices.

206

Considerations beyond the 3S’s. Dynamic Responses of Mechanical Sensors.

—

Since mechanical deformations occur time-dependently, the dynamic responses of

mechanical sensors to varying strains and stresses critically determine sensor accuracy in

practical use. There are three major issues in this regard: hysteresis, response time, and

strain-rate dependency, which are highly interrelated.

Hysteresis refers to differing response curves between loading and unloading, presenting a

fundamental challenge for mechanical sensors. It stems from the viscoelasticity of common

soft materials (

e.g.

, elastomers, gels) used in flexible mechanical sensors,

110

especially

when doped with conducting fillers. Micro-/nano-structuring for enhanced sensitivity

adds another source of energy dissipation from interfacial contact.

207

Moreover, flexible

mechanical sensors usually possess longer response times than rigid counterparts due to

sluggish polymer chain movements. This difference precludes time-critical applications

such as in robotic control and high-frequency applications, such as motion tracking in

racing sports. Strain-rate dependency refers to the differing response curves under varying

deformation rates or frequencies, leading to inaccurate readings in many applications, since

most deformations encountered in daily life are not at constant speed. This phenomenon

is often closely related to long response times,

i.e.

, when the structural or molecular

changes in sensors cannot catch up with the macroscale exerted stress, the sensors deviate

from equilibrium states to varying extents at different strain rates. Sometimes, strain-rate

dependency is an intrinsic characteristic dictated by the sensing mechanism (for instance,

pressure sensing based on magnetoelastic generators depends on the rate of change in

magnetic flux

169

208

). An effective strategy to overcome hysteresis and related issues is

to use rigid materials with special structural designs for strain sensing, while soft materials

are still required for deformability.

209

210

Microstructuring is effective for pressure sensors

through a reduction in contact area.

164

Alternatively, careful engineering of polymeric

networks can mitigate hysteresis,

211

215

yet the materials fabrication can be complex and

thus difficult for device integration. An emerging approach leverages machine learning

to correct the errors associated with the viscoelastic properties of soft sensors for better

prediction and analyses.

216

The dynamic performance of mechanical sensors may appear trivial, yet it is critically

important to practical measurements, deserving of greater attention. For example, although

stretchable strain sensors using conductive elastomeric composites have been widely

reported, their dynamic performance has rarely been investigated. Most studies only focus

Luo et al.

Page 12

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

on quasi-static electric behavior, where the sensing performance was evaluated in a static

state or in slow stretching–releasing processes (deformation speed <30 mm min

−1

, strain

rate <10% s

−1

217

Few studies have paid attention to the signal fidelity of strain sensors

at higher deformation speeds,

215

218

which is more relevant to dynamic motions in real

life, such as limb movements and hand gestures (speed >100 mm min

−1

, strain rate

>20% s

−1

217

In monitoring these dynamic motions, strain sensors using elastomeric

composites usually experience signal distortion, which is a common yet often overlooked

problem.

219

222

Dynamic responses at high and varying strain rates should be included as an

essential performance metric when reporting mechanical sensors.

Sensing Capabilities of Wearable Biosensors.—

Wearable biosensors are still

in early stages of development and many sensing capabilities await exploration and

development.

223

225

The first area of improvement is to expand the portfolio of

biomarkers that can be detected, to approach and to exceed current clinical assays. Complex

biomarkers (

e.g.

, proteins, hormones, nucleic acids, small molecules, and pathogens)

usually require bioaffinity-based sensing, and this strategy demands the design of effective

biorecognition moieties and proper immobilization and stabilization. Sensitive and selective

aptamers are being developed for a wider array of targets and they could be deployed

for this purpose.

120

Moreover, some approaches require multi-step preparation (

e.g.

immunobiosensors using antibodies),

making them challenging to integrate into wearable

platforms. Microfluidics is one promising approach,

226

229

which helps to collect, to

contain, and to drive biofluids, as well as to deliver and to wash out unbound detection

probes or labeling reagents. In addition, to reduce the number of preparation steps and

time consumed in immunosensing, development of label-free, reagent-free, and wash-free

methods is also necessary.

200

Surface-enhanced Raman spectroscopy (SERS) has emerged

as a powerful tool in this regard, but it requires a standalone spectrometer for signal

readout.

230

231

Recent work proposed an indirect electrochemical approach based on MIPs

coupled with redox-active reporters, which enabled the detection of non-electroactive

species in sweat, including amino acids, vitamins, metabolites, lipids, hormones, and

drugs.

228

This approach may be customized to detect a more diverse range of biomarkers.

A second area worth exploring is to realize continuous monitoring of these biomarkers,

which enables real-time monitoring and prompt detection of abnormalities. Common

bioaffinity assays (

e.g.

, immunoassays) of disease biomarkers involve complex steps,

require accurately controlled sample volumes and receptor regeneration, and are not

reversible. These features make immunoassays not amenable to continuous on-body

operation. Innovative strategies need to be crafted to overcome these challenges. For

example, modulating intermolecular forces between the bioreceptor and the target using

proper stimuli such as heat, ultrasound, electric/magnetic fields, and chemical cues might

be a viable approach to sensor regeneration.

Using this strategy, the regeneration of

MIP-based electrochemical biosensors by current or voltage has been demonstrated.

228

resettable electrochemical sweat lactate sensor has been developed through reversible redox

reactions in a biofuel cell.

232

Regeneration of aptamers for cocaine sensing has been realized

through pH-modulated conformational changes.

233

Microfluidics are a promising platform

for continuous-monitoring wearable biosensors. For instance, stretchable microfluidics can

Luo et al.

Page 13

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

expel sweat from filled channels to enable multiple usage.

234

Rational design in channel

shape and wettability can accelerate sweat collection and realize continuous sampling.

235

236

High temporal resolution can be achieved through encapsulating biofluids in water-in-

oil droplets and assessing the droplets sequentially, although system compactness needs

improvement.

237

Meanwhile, safe, continuous, controllable, and quantitative biofluid

sampling is also an important aspect of continuous biosensing. Passive micro-fluidics,

238

240

porous/hydrogel absorption pads,

128

180

241

244

microneedles,

245

246

iontophoresis and

reverse iontophoresis

174

247

248

are common solutions, but none can simultaneously satisfy

all requirements. Lastly, while wearable sweat sensors are the most often studied, other

bodily fluids such as saliva, tears,

249

and wound exudate should also be explored,

224

as they

may provide biomedical insights inaccessible

via

other means.

Equally important to technological advancement, robust knowledge of the clinical and

biomedical relevance and correlation of various bodily chemicals is needed to guide the

design and engineering of practically relevant biosensors. This knowledge often involves

metabolites in biofluids not traditionally studied.

In each case, the contents of the fluids

will need to be compared to current gold standards (typically blood) to determine whether

the fluid is representative of physiological state and what conversion factors are appropriate

to analyze the data obtained. Then, the advantages of more frequent and, in some cases,

continuous monitoring can be realized.

Holistic Approach to Accuracy Assurance.—

Reporting accurate values of the

parameters of interest is essential to sensors. To ensure sensor accuracy, it is important

to take a holistic approach spanning the entire life cycle from the development to the

deployment of a sensor technology (Scheme 1). First, during the design stage, fundamental

materials research is required to understand the materials properties, transduction

mechanisms, and device physics. This knowledge leads to optimized materials and device

structures. Going back and forth between scientific inquiry and engineering optimization

would lead to improved sensor accuracy, while the 3S’s and other factors should be

considered. Moving from design to deployment, well-controlled fabrication to produce

consistent devices is critical. Moreover, large-scale validation with standardized procedures

and benchmarking against gold-standard measurements are necessary to obtain reliable

and trustworthy calibration curves. For biomedical sensors, validation experiments can

be designed in accordance with the guidelines of the Clinical and Laboratory Standards

Institute.

In real-world deployment, calibration can be a complex issue. There are two levels of

consideration: the frequency of calibration during the entire sensor lifetime (manufacture,

shipment, and usage) and the number of calibration points for each calibration. Calibration

frequency usually concerns whether calibration can be performed by the manufacturer prior

to shipment of the product. Most commercial sensors fall in this category. In this case, the

cost of calibration is typically one third of the total cost of most commercial sensors today.

However, the exact cost depends on the number of calibration points that are needed. If the

sensor has a linear response in the needed dynamic range and if it has the same sensitivity

in that range for all manufactured components but different offset values, then a single-point

calibration is all that is needed. If the manufactured sensors do not have any offset in their

Luo et al.

Page 14

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

base value with the same sensitivity, then no calibration is needed (which is rare but would

significantly reduce costs). If a sensor has a linear response in the needed dynamic range

but there is variability in the sensitivity from sensor to sensor, then two-point calibration is

needed. If the sensor response is not linear, then multi-point calibration is needed (which

is commercially unattractive). Furthermore, calibration against temperature, humidity and

other environmental factors may be needed. For some emerging sensor technologies, the

calibration can shift over time, for example, from the time of manufacturing to the time of

usage. In that case, additional one-point or two-point calibrations may be needed prior to

use, which complicate use. Therefore, it is critically important for the community to report

linear response ranges, sensor-to-sensor variability, stability against environmental factors,

calibration method, and calibration drift over time.

Reliable Correlations between Sensor Signals and Object Status.—

Data without

interpretation is of little use. Making sense of data collected by sensors is equally, if

not more important with high-quality data acquisition. As flexible sensors enable many

parameters to be acquired in unconventional situations or from previously inaccessible

locations, the correlations between these parameters and the status of the monitored objects,

environments,

etc

. should be carefully examined.

250

Even for a single physical parameter,

the underlying meaning can be complicated to unravel. For instance, facial strain was

recently verified as an indicator of language commands through theoretical analysis and

simulation,

250

permitting the use of conformal strain sensors on face to deliver language

commands silently.

The correlation issue is especially concerning for biomedical applications, such as

biomarker measurement for disease diagnosis

251

and physiological monitoring for health

assessment.

A recent report found close correlations between tear glucose levels and

blood glucose levels with a lag time of 10 min,

252

indicating promising noninvasive

glucose monitoring by contact lenses. The study was conducted on three rabbits in the

experiment group and the control group respectively, which may not be sufficient as

biologically conclusive or generalizable to humans. Sweat is another biofluid in which

glucose monitoring is extensively conducted.

253

Nevertheless, the correlation between sweat

glucose and blood glucose can be easily altered by sample collection methods as well

as skin and environmental conditions.

254

The large uncertainty renders wearable sweat

glucose sensors

255

only sufficient for range estimation but not currently qualified for guiding

medical interventions. Large-scale tests with standardized protocols are needed to reach

robust conclusions. In addition, equal gender representation in clinical trials is also curial

for flexible sensor development and their practical usage in public. Recent results on a

conformable multimodal sensory face mask performed on an equal number of male and

female subjects indicate that current face masks are not suitable for women subjects in

general.

256

This result suggests a comprehensive mandate to be inclusive in human subject

studies to have technologies be beneficial for all.

As flexible sensor technology is collecting signal types some of which are traditionally

inaccessible, problems emerge in terms of the implications of sensor data. This issue calls

for extensive fundamental and biomedical research, where investigations involving gold-

Luo et al.

Page 15

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript

standard tests on adequate sample sizes

257

258

are needed to test the existence of correlations

and to generate reliable reference databases.

Tolerance to Mechanical Deformation and Damage.

A major advantage of flexible sensors is the ability to withstand significant deformation

without physical failure or performance degradation. This feature permits many use cases

with which conventional rigid sensors struggle, such as conformal skin patches/tattoos and

smart clothing. Nonetheless, this flexibility also poses great challenges in maintaining sensor

integrity and performance under often-unpredictable mechanical interactions between the

sensor and the environment.

Mechanical Robustness in Long-Term Use and at Large Deformation.—

Mechanical robustness describes the sensor’s ability to withstand different forms of

deformation without mechanical failure. Some extreme cases include exceptionally

large strain and high impact,

102

prolonged cycling strain, and constant friction. While

conventional rigid sensors can be protected from mechanical damage using high-

performance ceramics, metals, and thermosets, deformability of flexible sensors does not

permit the use of these mechanical protective materials in conventional ways. Furthermore,

due to the wide variety of materials used in flexible sensors, each having distinct mechanical

properties (

e.g.

, elastic modulus, Poisson’s ratio, viscoelasticity) and surface properties

(

e.g.

, surface energy, chemical composition), interfacial mismatch contributes a major

factor to mechanical instability. The exact deformation a flexible sensor experiences varies

greatly according to application, and hence exceptional mechanical robustness is not always

required. Nevertheless, we highlight the most significant issues, and the principles should

benefit the development of a number of flexible sensors.

Robust Soft-Hard Interfaces.—

One of the most prominent mechanical challenges in

flexible sensors is the interfacial instability between dissimilar materials. Stress and/or strain

concentration occurs at soft-hard interfaces, leading to a major source of failure through

delamination/detachment. Soft-hard interfaces exist in many forms: nanocomposites, layer-

by-layer laminates, interconnects, etc. The general principles in tackling soft-hard instability

are (1) improving interfacial adhesion and (2) avoiding abrupt softness/hardness difference.

Specific methods vary in different scenarios, but the principles hold.

For example, using a single materials system (or at least reducing the number of

materials) can eliminate many interfacial issues.

118

Recently, a capacitive pressure sensor

made entirely of CNT-doped polydimethylsiloxane (PDMS) was fabricated.

259

Tuning

the dopant concentration around the percolation threshold realized either electrode or

dielectric properties with little change in mechanical softness. The interlayers were

bonded together due to the similar chemistry of the layers. The resulting sensor could

maintain stable performance under 100,000 cycles of rubbing and other harsh deformation

conditions. Substrate mechanical engineering is effective in mitigating stress concentration

in heterogeneous stretchable electronics.

260

264

By synthesizing elastomers of different

stiffness or embedding rigid islands, the area under rigid components is made harder than

the surrounding area. Consequently, abrupt soft-hard transition between rigid functional

Luo et al.

Page 16

ACS Nano

. Author manuscript; available in PMC 2024 July 04.

Author Manuscript