Multimodal biosensing platforms for sensor-computing integration towards intelligent healthcare

Biosensors for electrochemical biosensing

As one of the most widely used and mature biosensing technologies, electrochemical biosensors play a pivotal role in fields ranging from clinical diagnostics to environmental monitoring. Their core advantage lies in the ability to translate specific biorecognition events into quantifiable electrical signals. These events include interactions between enzymes, antibodies, aptamers, or nucleic acids and their respective targets. This transformation is mediated by redox reactions occurring at a functionalized working electrode, where the transfer of electrons during biochemical reactions generates measurable currents, potentials, or impedance changes.​

To provide a comprehensive understanding of their operational mechanisms and design principles, this section first outlines the three primary electrochemical transduction modes that underpin these sensors. It then explores the common electrode materials and device architectures that enable their sensitivity, selectivity, and practical deployment in diverse applications.

(1) Amperometric biosensors

Amperometric biosensors are among the most widely utilized electrochemical sensing platforms, particularly for detecting small-molecule metabolites such as glucose, lactate, and uric acid (UA)^[25,26]. As shown in Figure 1A, these sensors operate by applying a constant potential to the working electrode (WE), where the target molecule undergoes a redox reaction, typically catalyzed by an immobilized, target-specific oxidase enzyme [e.g., glucose oxidase (GOx), lactate oxidase, uricase]^[27-31]. The generated faradaic current is directly proportional to the analyte concentration and serves as the quantifiable output.

Figure 1. Electrochemical biosensor technologies based on different sensing principles. (A) Amperometric detection of metabolites using enzyme-functionalized electrodes; (B) Voltammetric sensing of proteins or nucleic acids via electrodes modified with antibodies or oligonucleotides. Detection strategies include multistep sandwich formats, binding-induced conformational switches, or proximity-based affinity recognition; (C) Representative architectures of ion-selective electrodes for potentiometric biosensing; (D) Conceptual illustration of conductometric sensing, where changes in electrical conductivity upon target interaction are used for analyte quantification. Reproduced with permission^[44]. Copyright 2022, MDPI; (E) Impedimetric biosensing of analytes at varying concentrations, with Nyquist plots illustrating the system’s electrochemical behavior. The inset shows the randles equivalent circuit, comprising R_s, R_ct, C_dl, and W. Reproduced with permission^[45]. Copyright 2014, American Society for Microbiology. WE: Working electrode; AC: alternating current; R_s: solution resistance; R_ct: charge-transfer resistance; C_dl: double-layer capacitance; W: Warburg impedance.

The amperometric sensing mechanism can be classified into three generations based on the mode of electron transfer between the enzyme and the electrode. First-generation sensors rely on endogenous cofactors such as dissolved oxygen as electron acceptors; the enzymatic reaction produces hydrogen peroxide (H₂O₂), which is subsequently oxidized or reduced at the WE to generate a current signal. Second-generation designs introduce redox mediators (e.g., ferricyanide, ferrocene derivatives, Prussian blue), which shuttle electrons between the enzyme’s active site and the electrode surface, improving electron transfer efficiency and reducing oxygen dependence^[32]. Third-generation biosensors eliminate mediators entirely by enabling direct electron transfer (DET) between the redox center of the enzyme and the electrode, often facilitated by conductive nanomaterials or engineered enzyme-electrode interfaces^[33]. Standard amperometric biosensors typically adopt a three-electrode configuration, comprising a WE functionalized with the biorecognition element, a counter electrode (CE) to complete the circuit, and a reference electrode (RE) to maintain a stable potential. This system enables rapid, sensitive, and selective detection under low power operation, making amperometric biosensors particularly suitable for wearable and portable diagnostic devices. Their relatively simple fabrication process, compatibility with flexible substrates, and strong electrochemical response have made them a foundational component in non-invasive health monitoring systems.

(2) Voltammetric biosensors

Voltammetric biosensors convert the selective binding of a target molecule into an electrical signal by applying a varying potential to the working electrode and recording the resulting current. In a typical format for protein or nucleic acid detection [Figure 1B], the electrode surface is functionalized with capture probes, for example antibodies, aptamers or complementary DNA strands, that bind the analyte of interest^[34]. Because these biorecognition elements are not themselves electroactive, an electroactive label such as horseradish peroxidase, methylene blue or metal nanoparticle tags is introduced to generate a signal. When a potential sweep is carried out by cyclic voltammetry^[35,36], scanning forward and reverse potentials, by differential pulse voltammetry^[37,38], overlaying voltage pulses on a staircase waveform, or by square wave voltammetry^[39,40], superimposing symmetrical pulses, the tagged complexes undergo oxidation or reduction at their characteristic potentials. Quantitation is achieved by measuring the height of the resulting current peaks or shifts in their position relative to a control. Modern implementations often employ screen-printed electrodes integrated with microfluidic flow injection systems to automate sample handling and allow a single-step, multiplexed assay. These configurations, and support the miniaturization needed for point-of-care applications, enhancing throughput, minimizing reagent use, and facilitating the miniaturization required for wearable diagnostics.

(3) Potentiometric biosensors

As depicted in Figure 1C, potentiometric biosensors quantify target analytes by recording the open-circuit potential difference between a selective sensing electrode and a stable reference electrode, with no net current flow^[41]. The measured voltage adheres to the Nernst equation, shifting predictably as the activity of specific ions at the sensing membrane changes. Conventional ion-selective electrodes employ glass, crystalline or polymeric membranes to discriminate H⁺, K⁺, Na⁺, Ca²⁺ , and other biologically relevant ions, while solid-state variants replace internal electrolytes with conducting polymers or nanomaterial layers to enable miniaturization^[42]. By tailoring the membrane composition and incorporating appropriate ionophores or recognition elements, potentiometric sensors achieve high sensitivity (tens of millivolts per decade change in concentration), rapid response, and ultralow power consumption^[43]. These attributes make them well-suited for real-time monitoring of electrolytes in sweat, blood, or environmental samples, as well as for integration into portable and wearable diagnostic platforms.

(4) Conductometric biosensors

As illustrated in Figure 1D, conductometric biosensors detect analytes by measuring variations in the electrical conductance of the sensing element when a small alternating voltage is applied across two electrodes^[44]. When target molecules bind to the sensor surface or undergo a reaction, the overall resistance changes and can be monitored directly or by using impedance spectroscopy. Because this method does not require a reference electrode and operates under low excitation voltages, conductometric sensors can be fabricated as thin-film devices using standard manufacturing processes and scaled down easily.

These sensors generally exhibit lower specificity and sensitivity than faradaic or potentiometric systems because the measured signal integrates multiple conductive pathways. Nonetheless, conductometric devices are widely used in environmental applications, where shifts in water conductivity provide a straightforward indicator of ionic composition and pollutant levels.

(5) Impedimetric biosensors

Impedimetric biosensors measure the frequency-dependent impedance of the electrode-electrolyte interface to detect molecular recognition events without the need for labels^[45]. In practice, a small alternating potential is applied between the working and counter electrodes across a spectrum of frequencies while the resulting current is recorded [Figure 1E]. The system is commonly represented by the Randles equivalent circuit, which comprises the solution resistance (R_s), attributable to the bulk electrolyte; the double-layer capacitance (C_dl) at the electrode surface; the charge-transfer resistance (R_ct) associated with interfacial redox reactions; and the Warburg impedance (W), reflecting mass-transport limitations^[46]. When plotted as a Nyquist diagram, the high-frequency regime appears as a semicircular arc whose diameter directly corresponds to R_ct, whereas the low-frequency regime rises at approximately 45 degrees, illustrating the influence of diffusion processes. Binding of a target analyte to a recognition layer on the electrode modifies local charge density and introduces steric hindrance, thereby impeding access of redox probes such as ferrocyanide or ferricyanide and increasing both R_ct and overall impedance^[47]. The addition of a redox mediator may shift these impedance features by altering electron-transfer kinetics. Although this label-free approach yields detailed mechanistic information and is amenable to miniaturization for point-of-care applications, nonspecific adsorption and complex sample matrices can compromise sensitivity and selectivity, necessitating careful surface engineering and assay validation^[48].

Biosensors for microwave biosensing

Microwave biosensor technology offers promising solutions for continuous, non-invasive glucose monitoring. Microwave sensors harness the power of low-energy, safe electromagnetic waves that can penetrate through skin, fat, and muscle tissues to reach blood vessels and interstitial fluids^[49,50]. The fundamental principle is that variations in glucose concentration alter the dielectric properties of tissues, including how water molecules and other components interact with microwave energy^[51,52]. By detecting these subtle permittivity shifts, microwave sensors can generate real-time glucose readings^[53-55]. Microwave-based monitoring offers several advantages, including deep tissue penetration on the millimeter-to-centimeter scale, rapid response times of less than one second, inherent safety, and compatibility with wearable integration^[56-59]. External factors such as temperature and hydration may also affect dielectric responses, underscoring the need for robust calibration in practical applications^[60-63].

Recent developments in microwave biosensing have focused on optimizing resonant structures and transmission lines to enhance sensitivity and miniaturization. Among these, metamaterial-inspired elements such as split-ring resonator (SRR) and complementary split-ring resonator (CSRR) are particularly attractive due to their strong electromagnetic field localization and compact size^[64,65]. These resonators exhibit frequency shifts in reflection (S₁₁) or transmission (S₂₁) parameters when interacting with biological tissues, enabling detection of subtle changes in permittivity^[66-68]. The CSRR, in particular, enhances field confinement by etching resonant rings onto the ground plane of a transmission line, which intensifies the local electric field and improves sensitivity^[55,69]. Figure 2A presents a representative CSRR-based biosensor design featuring a portable radar-driven configuration capable of non-invasive fingertip glucose detection^[55]. Although SRR and CSRR devices exhibit excellent sensitivity, their performance can be affected by fabrication tolerances and environmental conditions, which necessitate precise calibration and multi-cell configurations to ensure measurement stability^[70,71].

Figure 2. Representative designs and working principles of microwave biosensors for non-invasive and wearable monitoring. (A) Conceptual illustration and fabricated prototype of a portable radar-driven sensor employing a four-cell CSRR configuration for non-invasive blood glucose measurement from the fingertip. Reproduced with permission^[55]. Copyright 2020, Springer; (B) Fabrication process and structural schematic of a gated graphene microwave CPW sensor. Reproduced with permission^[63]. Copyright 2024, Wiley; (C) Design of a tunable, non-invasive, wearable electromagnetic multi-sensing system inspired by vascular anatomy, showing slot antenna and band-reject filter prototypes adapted for hand and arm glucose monitoring. Reproduced with permission^[76]. Copyright 2020, AAAS; (D) Schematic of a wearable, flexible, body-matched electromagnetic sensing system employing multi-sensing and multi-location detection to correlate reflected EM waves with blood glucose variations. Reproduced with permission^[77]. Copyright 2022, Springer. SMA: SubMiniature version A; PCB: printed circuit board; Rx: Receiver; Tx: Transmitter; CSRR: complementary split-ring resonator; PDMS: polydimethylsiloxane; CPW: coplanar waveguide; FR4: flame retardant 4; EM: electromagnetic; VNA: vector network analyzer.

Another important class of microwave biosensors is based on graphene-integrated coplanar waveguide (CPW), which combine the superior electrical conductivity of graphene with the high sensitivity of planar microwave transmission lines^[63]. A CPW consists of a central conductor flanked by ground planes on the same substrate surface, forming a confined electromagnetic region that interacts with biological media placed above^[72-75]. As shown in Figure 2B, embedding graphene within the CPW enhances field confinement and amplifies small dielectric variations caused by glucose concentration changes^[63]. These sensors are compact and sensitive, though challenges remain in impedance matching and maintaining long-term operational stability under physiological conditions.

Building upon these principles, flexible and wearable microwave devices have emerged as a new frontier for biomedical sensing. The incorporation of soft, conformal materials allows these systems to maintain stable skin contact and continuously monitor multiple physiological parameters in real time. As illustrated in Figure 2C, vascular-inspired sensor systems replicate the geometry of veins and arteries to achieve depth-selective glucose detection. These architectures employ slot antennas and band-reject filters to optimize coupling with subsurface blood vessels^[76]. Another innovative design approach, as shown in Figure 2D, involves body-matched electromagnetic sensors that can simultaneously measure signals at different on-body locations while compensating for environmental factors such as temperature, humidity, and motion^[77].

In summary, microwave biosensors that utilize metamaterial resonators, graphene-based CPWs, and flexible vascular-inspired architectures are advancing toward practical, non-invasive glucose monitoring. Future work should emphasize improved calibration, miniaturization, and multimodal data fusion to enhance sensitivity, stability, and reliability in clinical and wearable healthcare applications.

Biosensors for optical biosensing

Optical biosensors convert light-matter interactions in biological media into measurable signals. Depending on the underlying phenomenon, they exploit reflection, refraction, scattering, absorption, or plasmonic resonance to deliver high sensitivity and selectivity^[78-80]. These sensors are generally categorized into several main types, including surface plasmon resonance (SPR) sensors, localized surface plasmon resonance (LSPR) sensors, fluorescence-based biosensors, microfluidic photonic platforms, and smartphone-integrated optical systems. Together, these diverse modalities demonstrate the versatility of optical transduction in biomedical applications^[81-83].

In the realm of plasmonic sensing, a critical distinction exists between propagating SPR on planar films and LSPR on nanostructures. While conventional SPR offers high sensitivity for interface kinetics, it typically requires bulky prism coupling. In contrast, LSPR arises from electron oscillations confined within metallic nanoparticles, eliminating the need for prisms and enabling flexible device integration. As illustrated in Figure 3A, a stretchable and highly sensitive LSPR-based optical strain sensor is utilized for human-activity monitoring^[84]. This sensor embeds high-density gold nanoparticles (AuNPs) within a soft elastomeric matrix. Unlike geometric waveguides, the sensing mechanism relies on the modulation of LSPR coupling. When the sensor is stretched or bent, the inter-particle distance between AuNPs changes, significantly altering the optical absorption or scattering intensity. This nanostructure-enhanced sensitivity enables the precise detection of minute physiological signals, ranging from respiration to pulse beats, while maintaining the mechanical compliance required for wearable skins.

Figure 3. Representative optical biosensor configurations for biochemical sensing. (A) Schematic of a stretchable optical strain sensor based on LSPR for human-activity monitoring. Reproduced with permission^[84]. Copyright 2019, American Chemical Society; (B) Plasmonic fluorescence biosensor based on PCR amplification for rapid and sensitive biomolecular analysis. Reproduced with permission^[85]. Copyright 2021, Royal Society of Chemistry; (C) Schematic illustration of a magnetic nanochain-integrated microfluidic biochip platform for multiplexed biosensing. Reproduced with permission^[87]. Copyright 2018, Springer; (D) Ultra-compact dual-channel integrated CO₂ infrared gas sensor based on on-chip NDIR technology. Reproduced with permission^[88]. Copyright 2024, Springer. GNP: Gold nanoparticle; PDMS: polydimethylsiloxane; IR: infrared; LED: light-emitting diode; PPT: plasmonic photothermal; PCR: polymerase chain reaction; PMN: plasmonic magnetic nanoparticle; SERS: surface-enhanced Raman scattering; MEMS: PCB: printed circuit board; LSPR: localized surface plasmon resonance; NDIR: non-dispersive infrared.

Complementing index-based sensing, emissive readouts exploit fluorescence intensity or lifetime to provide molecular specificity and ultralow detection limits. Figure 3B presents a fluorescence biosensor employing plasmonic enhancement in conjunction with polymerase chain reaction (PCR) amplification^[85]. Fluorescence-based biosensors operate by detecting the light emitted from fluorescent molecules following optical excitation, and the signal intensity or lifetime is directly related to analyte concentration. In this example, metallic nanostructures amplify the local electromagnetic field, which significantly increases fluorophore excitation and emission efficiency. When coupled with PCR amplification, this effect enables highly sensitive and rapid detection of nucleic acid biomarkers. Such designs demonstrate how combining photonic field enhancement with biochemical amplification mechanisms can substantially improve analytical performance in biosensing^[86].

To manage transport phenomena and enable scalable automation, microfluidic optics structure both light paths and analyte motion within confined volumes. The integration of multiple detection principles is further demonstrated in Figure 3C, which shows a magnetic nanochain-assisted microfluidic biochip^[87]. Microfluidic optical biosensors typically direct light through transparent channels containing the analyte, where variations in refractive index or absorption modulate the transmitted or scattered light signal. In this particular system, magnetic nanochains are used to align and concentrate target molecules within microchannels, thereby improving the interaction efficiency between light and biological components. The result is faster response time, higher sensitivity, and improved reproducibility. This microfluidic configuration exemplifies how optical and magnetic control strategies can be combined to achieve compact, automated, and high-throughput sensing platforms^[87].

For environmental and physiological gas analysis, miniaturization is essential for portable deployment. Figure 3D shows an ultra-compact dual-channel integrated CO₂ infrared gas sensor. As depicted in the device architecture, this platform leverages the non-dispersive infrared (NDIR) absorption principle on a photonic integrated circuit^[88]. The innovative “dual-channel” design incorporates a reference waveguide alongside the sensing arm to allow for differential measurement, effectively compensating for environmental drifts and light source fluctuations. By confining light within micro-scale structures to interact with the target gas, this on-chip sensor achieves high selectivity and stability in a minimal footprint, exemplifying the potential of integrated photonics for robust environmental and breath gas diagnostics.

Overall, optical biosensors represent a rapidly evolving class of analytical tools that translate light-matter interactions into biochemical information. Different approaches provide complementary advantages by aligning sensing physics with deployment needs, where SPR sensors enable real-time label-free detection, LSPR sensors facilitate flexible motion tracking, fluorescence biosensors offer high sensitivity and selectivity, microfluidic optical systems deliver scalability and automation, and smartphone-based devices increase accessibility for field applications^{[84,85,88-90]}. Ongoing advances in materials, optical design, and signal processing are expected to drive the integration of these modalities, paving the way for hybrid, intelligent, and personalized biosensing platforms suitable for both clinical and on-site use^[91-94].

Biosensors for gas detection

Gas biosensors are emerging analytical platforms that integrate biological recognition elements (e.g., enzymes, antibodies, aptamers/peptides/DNA, living cells, and olfactory receptors) with physical transducers to detect volatile organic compounds (VOCs) and gaseous biomarkers from breath, skin emissions, indoor air, and environmental samples. By operating within a thin, hydrated microenvironment that sustains biorecognition activity under humid, complex matrices, these systems enable continuous or near-real-time measurements suited to breathomics, environmental surveillance, food safety, and occupational exposure assessment. Compared with conventional chemoresistive gas sensors, they offer lock-and-key selectivity, biological signal amplification via catalytic or cellular pathways, and robust performance in high-humidity conditions, while supporting point-of-care deployment through miniaturization, rapid response, and electronic readouts^[95,96].

The choice of biorecognition play a critical role for sensing performance. Enzymatic systems continue to be the backbone of continuous gas sensing, where immobilized oxidases and dehydrogenases quantify targets through processes such as oxygen consumption, hydrogen peroxide generation, or the conversion between NADH and NAD⁺. These systems provide selective and humidity-resilient detection for analytes like ethanol, formaldehyde, and acetone across ppb to sub-ppm ranges^[97-99]. Advanced implementations amplify signals through multi-enzyme cascades, oxygen/NAD-linked amperometry, or triphasic gas-liquid-solid reactors that enhance mass transfer^[100]. Optical “bio-sniffers” extend enzyme assays into fluorometric formats, using O₂-quenched probes or NADH autofluorescence, to deliver linear, real-time readouts within clinically relevant breath windows with low coefficients of variation^[101,102]. As depicted in Figure 4A, whole-cell biosensors offer integrative, systems-level responses: recombinant lux reporters quantify gas toxicity (for example, benzene) with dose dependence and long shelf-life, and immobilized promoter arrays resolve VOC fingerprints from contaminated foods; coupled with machine learning, such arrays can achieve near-perfect discrimination of pre-mold states and commodity identity^[103,104]. Olfactory-receptor (OR) biosensors, built by co-expressing insect or mammalian ORs with olfactory receptor co-receptor (Orco) and calcium ion (Ca²⁺) indicators, emulate biological olfaction to reach ppb sensitivity and quantify mixtures using relative-comparison protocols that mitigate drift and nonlinearity; co-expressing multiple ORs in a single cell line broadens dynamic range under cross-sensitivity^[105]. As illustrated in Figure 4B and C, biofunctional nanocomposites enhance selectivity by integrating bioligands with two-dimensional conductors, such as peptide- or DNA-modified MXenes or graphene^[106,107]. These composites achieve ppb-level detection limits, provide enhanced responses compared to pristine materials, and demonstrate robust breath classification for conditions like cancer and other pathophysiologies when analyzed with neural networks^[108].

Figure 4. Representative designs and working principles of gas biosensors. (A) Schematic illustration of metal-based nanosystems with improved physico-chemical attributes through different strategies for enhanced detection of LC breath biomarkers in humans with smart and point-of-care modules. Reproduced with permission. Reproduced with permission^[103]. Copyright 2024, American Chemical Society; (B) Schematic diagram of the human olfactory system and the sensor array system for gas identification; (C) Schematic diagram of the instant testing platform for cancer recognition with the assistance of machine learning (ML). (B and C) Reproduced with permission^[107]. Copyright 2025, American Chemical Society.

Transduction modalities and device architectures determine how signals are captured and delivered. Electrochemical formats (amperometric, potentiometric, impedimetric) convert enzymatic redox or binding events into current, potential, or impedance changes; NAD-dependent dehydrogenase schemes expand target scope, and gas-liquid-solid triphasic photoelectrochemical reactors markedly increase oxygen utilization and mass transfer, enabling ultralow limits with rapid recovery in complex matrices^[32,100]. Optical biosensors implement enzyme-coupled fluorescence or chemiluminescence with fiber optics for remote readouts and microdomain mapping of VOC release on skin and in breath^[102]. Acoustic biosensors (quartz crystal microbalance; shear-horizontal surface acoustic wave, SH-SAW) track mass loading with millisecond kinetics; dual-channel SH-SAW chips with in-situ references and inkjet-printed capture coatings have demonstrated combined 100% sensitivity and 100% specificity for dual HIV antibody biomarkers in a pilot clinical study using microliter samples and sub-minute electronic outputs, highlighting compact, connected diagnostics for point-of-care screening^[109]. Field-effect transistor (FET) biosensors couple catalytic or affinity layers (e.g., metal-phthalocyanine-graphene composites) to high-mobility channels for label-free, real-time gas measurements with picomolar sensitivity to species such as H₂S and strong anti-interference characteristics, enabling in situ monitoring of cellular fluxes and inhibitor pharmacology^[110]. Chemiresistive nanoarrays functionalized with peptides or DNA on 2D materials deliver low-power, room-temperature operation, programmable selectivity, and scalable array integration, supporting nose-on-chip platforms that, when paired with machine learning, resolve multi-class odor patterns and disease cohorts with high accuracy^[107].

Biosensors for piezoelectric biosensing

Piezoelectric biosensors convert mechanical stress into an electrical signal by exploiting the piezoelectric effect of certain crystals. When an alternating voltage causes the crystal to vibrate at its resonance frequency, any additional mass on the surface slows the oscillation and lowers the measured frequency. This frequency shift is proportional to the mass of biomolecules bound to the surface.

Expanding the scope from surface mass sensing to deep-tissue imaging, Figure 5A illustrates a wearable cardiac ultrasound imager. This platform utilizes a matrix of piezoelectric transducers embedded in a stretchable substrate to conform to the curvature of the human chest^[111]. The transducers operate by transmitting ultrasonic waves into the body and receiving the echoes reflected from internal organs. By converting these mechanical echoes back into electrical signals, the device constructs real-time images of the cardiac structure. This application highlights the capability of piezoelectric sensors to perform continuous, non-invasive hemodynamic monitoring, capturing critical parameters such as stroke volume and cardiac output without the need for bulky hospital equipment.

Figure 5. Representative architectures and working principles of piezoelectric biosensors. (A) Schematic of a wearable cardiac ultrasound imager based on stretchable piezoelectric transducer arrays for real-time heart monitoring. Reproduced with permission^[111]. Copyright 2023, Springer; (B) Configuration of a SAW biosensor for E. coil detection, illustrating antibody immobilization on the delay line and bacterial capture. Reproduced with permission^[112]. Copyright 2024, Royal Society of Chemistry; (C) Real-time monitoring of S. aureus lysis on a QCM sensor, showing the experimental setup for tracking mass and dissipation changes during phage/enzyme treatment. Reproduced with permission^[114]. Copyright 2025, Springer. IDT: Interdigital transducer; E. coli: Escherichia coli; S. aureus: Staphylococcus aureus; QCM: quartz crystal microbalance.

For molecular-level detection, surface acoustic wave (SAW) devices offer high sensitivity for pathogen screening. Figure 5B details the configuration of a SAW biosensor designed for foodborne pathogen detection^[112]. In this architecture, an input interdigital transducer (IDT) generates acoustic waves that travel across the piezoelectric substrate’s delay line. A gold film deposited on this delay line is functionalized with antibodies specific to the target pathogen (Escherichia coli, E. coli) via self-assembled monolayers (SAMs). Upon exposure to a contaminated sample, the bacteria bind specifically to the antibodies. This binding event introduces a significant mass load and viscoelastic change to the sensor surface, which perturbs the propagating acoustic wave. The resulting shift in wave velocity or amplitude allows for the rapid, label-free quantification of bacterial contamination directly from complex matrices.

Complementing SAW technology, the quartz crystal microbalance (QCM) remains a powerful tool for monitoring dynamic biological processes. In a typical QCM sensor, a thin quartz wafer with metal electrodes on each side is driven by an oscillator circuit to vibrate in a shear mode. Surface immobilization of antibodies or other recognition elements does not significantly alter the baseline frequency^[113]. Once the target analyte binds, the added mass produces a measurable decrease in resonance frequency according to the Sauerbrey equation. Figure 5C demonstrates the real-time monitoring of bacterial lysis on a QCM platform^[114]. The approach simulates phage antibiotic synergy by examining bacterial degradation induced by either lysostaphin or phage P68. Staphylococcus aureus (S. aureus) cells were immobilized on a poly-L-lysine-coated QCM surface. Changes in resonance frequency and energy dissipation were recorded as indicators of bacterial membrane disruption and biofilm breakdown^[114]. The system enables tracking of the mass variations caused by cell lysis and release of intracellular contents. Additionally, a flow-through configuration permitted simultaneous nutrient delivery and metabolite removal, allowing comprehensive assessment of the dynamic interactions between bacteriophages and their bacterial hosts.

Biosensors for thermal sensing

Thermal and thermo-responsive interfaces play a dual role in multimodal biosensing platforms. They serve as diagnostic tools for monitoring physiological temperature variations and as energy harvesting units to power wearable electronics. While traditional calorimetric biosensors focus on detecting heat from enzymatic reactions^[115,116], recent advances have pivoted toward thermoelectric generators (TEGs) that exploit the Seebeck effect^[117,118]. These devices convert the temperature gradient between the human body and the environment into electrical energy. This mechanism enables self-powered operation while simultaneously providing precise temperature mapping, which is a critical capability for calibrating temperature-sensitive biochemical sensors in a multimodal system.

To address the power constraints of wearable systems, Figure 6A illustrates a self-powered electronic skin with multisensory functions based on thermoelectric conversion^[119]. In this architecture, flexible thermoelectric materials are integrated into a skin-conformable patch. The device operates by capturing body heat to generate a voltage potential. This potential serves as both a sensing signal for continuous skin temperature monitoring and a renewable power source for the integrated circuit. This dual-functionality allows the platform to detect subtle physiological thermal fluctuations, including those indicative of inflammation or metabolic changes, without the need for external batteries. Consequently, this design significantly reduces the device footprint and maintenance burden for long-term health monitoring.

Figure 6. Thermoelectric platforms for self-powered thermal sensing and energy harvesting. (A) Schematic of a self-powered electronic skin based on thermoelectric conversion, enabling dual functions of physiological temperature monitoring and energy harvesting from body heat. Reproduced with permission^[119]. Copyright 2020, Wiley; (B) Structural design and optical characterization of a stretchable temperature electronic skin, featuring high-performance Bi-Te micro-thermoelectric generators connected via an island-bridge architecture for distributed mini-region sensing. Reproduced with permission^[120]. Copyright 2023, Wiley. F-TEG: Flexible thermoelectric generator; PDMS: polydimethylsiloxane; STES: self-powered temperature electronic skin; Bi-Te: Bismuth Telluride.

However, integrating rigid high-performance thermoelectric materials into soft wearables presents mechanical challenges. Addressing the conflict between flexibility and energy efficiency, Figure 6B presents a self-powered temperature electronic skin based on an “island-bridge” structure and Bismuth Telluride (Bi-Te) micro-thermoelectric generators^[120]. As shown in the detailed optical images, rigid Bi-Te pillars act as the islands and are interconnected by stretchable conductive wiring that functions as the bridges. This design mechanically isolates the active thermoelectric elements from strain, ensuring that the device maintains stable performance even under stretching or bending. The array configuration enables distributed mini-region sensing, allowing for high-resolution spatiotemporal mapping of skin temperature. Such precise thermal profiling is essential not only for fever detection but also for decoupling temperature interference from concurrent electrochemical or mechanical signal recordings in a multimodal setup.

While each sensing modality discussed above possesses unique transduction characteristics, their practical implementation in “sensor-computing” systems requires a careful trade-off between performance and integration constraints. Table 1 provides a comparative analysis of these key biosensing modalities, contrasting their mechanisms, sensitivity, response times, and power consumption. Crucially, the table evaluates their suitability for sensor-computing integration, highlighting how specific limitations can be addressed through hierarchical computing architectures (Level 2/3 processing).

Blood biomarkers

Blood serves as a rich reservoir of biomarkers that play a crucial role in assessing systemic health. Glucose, one of the most extensively monitored biomarkers, is a core metabolic indicator. With a normal fasting range of 3.9 - 6.1 mmol/L, glucose is of utmost importance in diabetes management. Deviations from this range, such as hyperglycemia or hypoglycemia, are indicative of metabolic disorders^[123]. Biosensors for glucose commonly utilize glucose oxidase to catalyze redox reactions, effectively converting glucose concentration into electrical signals. Advanced designs incorporate nanomaterials, enabling micromolar level sensitivity for continuous blood glucose monitoring^[124].

In addition to glucose, blood contains a plethora of other vital biomarkers. Lipids, including cholesterol [total, low-density lipoprotein (LDL), high-density lipoprotein (HDL)] and triglycerides, are key indicators. Enzymatic or immunological biosensors are employed to monitor these lipids, providing valuable insights into cardiovascular risk. Hormones like insulin, which regulates glucose metabolism, and thyroid hormones such as triiodothyronine (T3) and thyroxine (T4), are essential for diagnosing endocrine disorders^[125]. Proteins such as C-reactive protein (CRP), an inflammatory marker, and troponin, which indicates myocardial infarction, are detected through highly specific immunosensors^[126]. Electrolytes like K⁺, Na⁺, and Ca²⁺ are also significant. Imbalances in their levels can signal renal or metabolic problems, and they are measured using potentiometric ion-selective electrodes^[127].

Sweat biomarkers

Sweat, secreted by eccrine glands on the skin surface, serves as a rich source of non-invasively accessible physiological information, making it an attractive medium for wearable monitoring devices. The key analytes present in sweat offer valuable insights into multiple physiological processes.

Lactate, a byproduct of anaerobic metabolism, is a crucial biomarker. During physical activity, muscle cells switch to anaerobic respiration when oxygen supply is insufficient, leading to lactate production. Sweat lactate concentrations typically range from 1 to 10 mM, and these levels are directly proportional to muscle exertion^[128]. Biosensors that utilize lactate oxidase can catalyze the oxidation of lactate, generating current signals. These sensors have been shown to accurately reflect changes in exercise intensity, providing athletes and trainers with immediate feedback on energy expenditure and muscle fatigue^[129].

Electrolytes, including sodium ions (Na⁺, with a concentration range of 5-40 mM) and potassium ions (K⁺, 3-15 mM) in sweat, are vital for maintaining proper physiological function. The levels of these electrolytes mirror the body’s hydration status and electrolyte balance. Potentiometric sensors equipped with ion-selective membranes, such as sodium-selective glass membranes for Na⁺ detection, play a fundamental role in preventing dehydration^[130]. By precisely measuring electrolyte concentrations in sweat, these sensors can alert individuals to the need for fluid and electrolyte replacement^[131].

Metabolites in sweat, such as urea and creatinine, also contribute to the assessment of overall metabolic health. Urea is an end product of protein metabolism and serves as an indicator of renal function. Creatinine, produced from muscle metabolism, provides additional information about metabolic activity^[132]. Enzymatic reactions are commonly employed to detect these metabolites. For instance, urease can be used to break down urea, and the resulting products can be measured to determine urea concentration^[133]. The detection of these metabolites in sweat, offers a non-invasive means to monitor renal function and overall metabolic status over time^[134].

Tumor biomarkers

Tumor markers, a class of biomolecules aberrantly overexpressed in cancer cells or tissues, play a pivotal role in facilitating early cancer detection, evaluating disease prognosis, and monitoring treatment responses. These biomarkers, which include proteins, glycoproteins, and nucleic acids, exhibit altered expression profiles in pathological states, making them valuable indicators for clinical oncology. Key tumor markers and their clinical associations are as follows:

Carcinoembryonic antigen (CEA): A glycoprotein exhibits elevated serum levels in various malignancies such as colorectal, lung, and breast cancers. In healthy populations, normal serum CEA concentrations are generally maintained below 5 ng/mL^[135], with elevations beyond this range potentially indicating the need for further diagnostic evaluation. This biomarker’s association with multiple cancer types underscores its utility in clinical settings for aiding in the identification of pathological processes related to malignant growths^[136,137].

Prostate-specific antigen (PSA): A serine protease predominantly secreted by prostate epithelial cells, serving as a specific marker for prostate cancer. Clinically, serum PSA levels exceeding 4 ng/mL are considered indicative of potential malignancy, though this threshold may vary based on age and individual risk factors^[138].

Alpha-fetoprotein (AFP): A fetal glycoprotein re-expressed in hepatocellular carcinoma (HCC). Elevated serum AFP levels (> 20 ng/mL) are recognized as a critical diagnostic indicator for HCC, particularly when combined with imaging modalities^[139].

CA125: A mucin-like glycoprotein widely used as a biomarker for ovarian cancer. Beyond aiding in initial diagnosis, CA125 levels are clinically valuable for monitoring treatment efficacy and detecting disease recurrence^[140].

In summary, tumor markers, as biomolecules aberrantly overexpressed in cancer cells or tissues, play a crucial role in early cancer detection, prognostic evaluation, and monitoring of treatment responses. Key markers such as CEA, PSA, AFP, and CA125 are each associated with specific types of cancer, providing valuable indicative information for clinical oncology. Biosensors, relying on high-specificity recognition mechanisms like antibody-antigen interactions and nucleic acid hybridization, have emerged as powerful tools for detecting these tumor markers^[141,142]. Their significance in improving patient survival rates through timely intervention is paramount.

Food safety-related analytes

Biosensors designed for food safety applications primarily target harmful substances, serving as a critical barrier to prevent contamination-induced illnesses. Their utility spans the detection of diverse hazardous agents, with key focus areas as follows:

Pesticides: Organophosphates and carbamates, two major classes of pesticides, are commonly monitored using enzyme inhibition assays. A typical example is the inhibition of acetylcholinesterase, whose activity is quantitatively suppressed in the presence of these compounds. This mechanism enables sensitive quantification of pesticide residues in matrices such as fruits, vegetables, and aqueous samples, with detection limits ranging from nanograms per milliliter to micrograms per milliliter^[143,144].

Pathogens: Microbial contaminants, including pathogenic bacteria (e.g., E. coli O157:H7, Salmonella spp.) and viruses, are detected through two primary strategies. Nucleic acid-based biosensors, often integrated with polymerase chain reaction (PCR) amplification, enable highly specific identification of genetic signatures. Alternatively, antibody-based biosensors leverage specific recognition of surface antigens, facilitating rapid screening of intact pathogens^[145].

Toxins: Mycotoxins (e.g., aflatoxin B1) and bacterial toxins (e.g., botulinum toxin) represent a class of low-molecular-weight hazards requiring ultra-sensitive detection. Immunosensors, which rely on high-affinity antibody-toxin interactions, and aptasensors, utilizing target-specific nucleic acid aptamers, are particularly effective in this context. These platforms achieve trace-level quantification with exceptional specificity, ensuring the accurate identification of even minute toxin concentrations^[146-148].

Environmental pollutants and toxins

Environmental biosensors serve as pivotal tools for monitoring pollutants, playing a crucial role in evaluating risks to ecosystems and human health through the specific detection of hazardous substances. Their application encompasses the monitoring of diverse environmental pollutants, with key focuses outlined as follows:

Heavy metals: Toxic heavy metal ions such as lead (Pb²⁺), mercury (Hg²⁺), and cadmium (Cd²⁺) are primary targets of environmental biosensors. These sensors typically employ recognition elements like metal-binding proteins (e.g., metallothioneins) or DNA enzymes, which exhibit high specificity toward heavy metal ions^[149,150]. The remarkable sensitivity of such systems allows for detection limits as low as parts per billion (ppb) to parts per trillion (ppt), enabling the identification of even trace-level heavy metal contamination in environmental matrices such as water, soil, and sediment.

Organic pollutants: A wide range of organic pollutants, including polycyclic aromatic hydrocarbons (PAHs), pharmaceuticals, and industrial chemicals, are monitored using specialized biosensors. Microbial biosensors, which utilize pollutant-degrading bacteria engineered to produce detectable signals upon exposure to specific organic compounds, are widely applied for this purpose. Alternatively, molecularly imprinted polymers (MIPs) serve as effective recognition elements in biosensors for organic pollutant detection, leveraging their high selectivity for target analytes. These approaches facilitate accurate quantification of organic pollutants in complex environmental samples^[151,152].

Neurotransmitters and neuromodulators

Neurotransmitters, as key mediators of neural signaling, play a pivotal role in maintaining normal nervous system function. Imbalances in their concentrations are closely associated with the pathogenesis of various neurological and neuropsychiatric disorders, making their accurate detection critical for both disease mechanism research and clinical diagnosis.

Dopamine: This neurotransmitter is primarily involved in the regulation of reward pathways and motor control. A deficiency in dopamine levels is a hallmark of Parkinson’s disease, characterized by motor dysfunction, while excessive dopamine activity has been implicated in the pathophysiology of schizophrenia. The biosensors capitalize on the electrochemical oxidation of dopamine, generating measurable currents that correlate with its concentration. Such systems achieve sub-micromolar sensitivity, enabling precise quantification in complex biological matrices such as brain tissue extracts or cerebrospinal fluid^[153,154].

Serotonin (5-hydroxytryptamine, 5-HT): Functionally, serotonin modulates mood, sleep-wake cycles, and emotional regulation, with reduced levels strongly linked to depressive disorders. Optical biosensors have been developed for serotonin detection, leveraging either fluorescent probes that undergo specific spectral changes upon binding to 5-HT or aptamers with high affinity for serotonin^[155,156]. The application of specialized biosensors in detecting these neurotransmitters not only enables precise monitoring of their dynamics in neural interfaces but also contributes to advancing our understanding of neurological disorders, thereby laying the groundwork for the development of novel diagnostic and therapeutic interventions.

To systematically guide the design of multimodal platforms, it is essential to match the biochemical nature of these analytes with the most appropriate transduction interfaces. Table 2 classifies representative biomarkers into distinct categories spanning from simple ions to complex macromolecules, mapping them to their preferred sensing modalities and recognition elements. This classification underscores the rationale for multimodal integration: by combining sensors that target distinct biochemical classes, integrated platforms can achieve a more holistic and accurate assessment of physiological states.

Advantages of multimodal biosensors

Multimodal biosensors offer a range of significant advantages over single-mode systems by leveraging the complementary strengths of different sensing technologies. A key benefit is their ability to capture orthogonal biochemical and physiological signals, creating a multidimensional data layer that enriches diagnostic insights. For example, Al-Rawhani et al. reported a multimodal integrated sensor platform which combines optical and electrochemical sensing modalities to achieve comprehensive biomarker analysis^[157]. This dual-mode system integrates optical detection (e.g., surface-enhanced spectroscopy) with electrochemical transduction, enabling simultaneous quantification of multiple analytes in complex biological samples. Such integration allows for cross-validation of results. Optical signals provide high specificity for molecular recognition events while electrochemical readouts offer quantitative precision and real-time response. This addresses the limitations of single-mode systems that would otherwise capture only partial biochemical information. Similarly, Zhao et al. developed a multimodal sensing platform integrating surface-enhanced Raman scattering (SERS) and colorimetric sensing for sweat biomarker detection^[158]. The SERS modality provided high specificity for molecular recognition events, while the colorimetric readout offered rapid qualitative validation. This integration effectively reduced false positives caused by non-specific adsorption in complex sweat matrices, achieving lower detection limits (down to micromolar levels for glucose and lactate) compared to either single modality alone. Such results demonstrate how combining optical sensing techniques can synergistically enhance overall detection performance, highlighting the advantage of multimodal designs in addressing the limitations of individual sensing methods. These examples demonstrate how multimodal architectures mitigate the inherent limitations of individual techniques. Such limitations include background interference in optical sensing and limited specificity in single-mode detection. By synergizing complementary strengths, multimodal architectures enhance overall detection performance.

Another advantage is the enhanced robustness in dynamic biological environments, where temperature fluctuations, pH shifts, and matrix effects can perturb single-mode sensor readings. Multimodal platforms address this by incorporating dedicated reference channels. A typical configuration might include a thermoresistive sensor to normalize electrochemical signals against temperature drifts, combined with a potentiometric pH probe to correct for environmental acidity^[159]. This self-calibrating capability is particularly valuable for wearable devices, where skin temperature variations (32 to 36 °C) can alter enzyme activity in electrochemical sensors by up to 20% per °C. By contrast, single-mode devices often require frequent recalibration or suffer from cumulative drift in real-world use^[160,161].

Moreover, multimodal sensing platforms overcome the limitations of single-analyte detection by enabling concurrent measurement of physiologically correlated biomarkers across multiple scales. For instance, the biosensing system develop by Wang et al. employs genetically encoded fluorescent probes to achieve spatiotemporally resolved lactate monitoring from subcellular compartments to whole-body dynamics in diabetic models, integrating responses to metabolic perturbations (glucose deprivation) and pharmacological interventions with ≤ 5-s temporal resolution^[162]. This contrasts sharply with conventional single-mode setups that require separate devices for each analyte. In clinical diagnostics, such multimodal synergy enhances early disease detection through cross-validated biomarker networks. A representative implementation couples an ultrasensitive electrochemical aptasensor for cardiac troponin I with microwave Doppler-based heart rate variability (HRV) sensing using vector analytic demodulation (VAD) algorithms^[163]. This dual-modality platform detects molecular-level myocardial injury (cTnI elevation) concurrently with autonomic dysfunction (HRV anomalies), achieving 98% accuracy in QRS complex identification under physiological noise^[164]. Such integration delivers essential diagnostic context unattainable with single-mode devices, specifically by correlating nanoscale biomolecular events (e.g., cardiac troponin I fluctuations) with organ-level functional dynamics (e.g., autonomic regulation via heart rate variability) to discriminate benign biomarker variations from pathological states such as acute coronary syndromes^[163,164].

While multimodal biosensing offers richer datasets, it is crucial to delineate when this complexity translates into tangible clinical benefits versus when single-mode approaches suffice. Multimodal strategies clearly outperform single-mode sensors in two specific scenarios: (1) Decoupling cross-sensitivity: In complex bio-fluids, a single sensor often responds to interfering species (e.g., ascorbic acid affecting glucose readings)^[165]. A multimodal system can use a secondary modality (e.g., a selective polymer coating or an orthogonal transduction method) to calibrate and subtract these interferences in real-time^[166]. (2) Correlated physiological profiling: For conditions like cardiovascular disease, measuring a single marker (e.g., ECG) provides incomplete risk assessment. Integrating biochemical markers (e.g., Troponin) with physical signals (e.g., ECG/blood pressure) creates a closed-loop diagnostic model that single-mode sensors cannot achieve^[167].

However, the added complexity of multimodality is not justified when targeting highly specific, high-concentration analytes in controlled environments. For instance, in basic fitness tracking where only heart rate is required, adding biochemical sensing capabilities may introduce unnecessary power drain, data processing latency, and fabrication costs without significantly improving the user experience. Therefore, the design of “sensor-computing” platforms must be driven by the specific clinical question rather than a desire for technological novelty.

Limitations for practical applications

Increased system complexity arises from integrating heterogeneous transduction mechanisms, leading to material compatibility conflicts and signal crosstalk. For instance, conductive electrode materials (e.g., gold, carbon nanotubes) essential for electrochemical sensors can absorb or scatter light, degrading optical sensing performance^[168]. Similarly, microwave antennas coupled with enzymatic electrodes may induce localized heating that denatures biomolecules^[169,170]. Such interference necessitates advanced packaging strategies like microfluidic channel isolation or frequency-domain filtering^[171,172], which can decrease manufacturing complexity and failure risks. Empirical studies on wearable sensors confirm that inter-modality interference is a primary contributor to reliability degradation, though specific failure rate comparisons require further clinical validation^[173,174].

Enhanced power consumption and form factor constraints further limit practicality. Each modality imposes distinct power demands: optical systems require light sources, electrochemical modules need potentiostats, and impedance analyzers rely on high-frequency signal processors. Cumulatively, these increase energy consumption by 3 to 5 times compared to single-mode devices, straining battery life in wearables^[175,176].

Data processing and analysis burdens scale with multimodal data heterogeneity. Platforms combining voltammetric, potentiometric, and optical outputs generate time-series, frequency-domain, and intensity datasets requiring synchronization and joint analysis. This demands computationally intensive machine learning models (e.g., convolutional neural networks (CNNs) for spatial patterns, long short-term memory networks (LSTMs) for temporal correlations)^[177-179], yet unoptimized fusion can degrade accuracy. Effective analysis thus hinges on adaptive algorithms like physiologically inspired spatiotemporal fusion models to resolve signal conflicts^[180].

Printing techniques for biosensor fabrication

Printing technologies have emerged as versatile approaches for manufacturing flexible and integrated biosensor platforms, offering high spatial resolution, compatibility with diverse functional materials, and scalability for large-area production. Among these, inkjet printing, extrusion-based printing, and laser-assisted printing represent three complementary modalities, each with unique operational principles and advantages.

Inkjet printing is a key technique for creating flexible biosensor components because it enables non-contact, mask-free patterning with resolutions down to 20 to 50 µm and accommodates a variety of bioink formulations^[182]. The process proceeds in three stages: the generation of droplets by thermal or piezoelectric actuation, the interaction of those droplets with the substrate surface, and the evaporation or curing of solvent to fix the patterned material^[183]. Two main operating modes are used in biosensor fabrication. In continuous inkjet printers a high-pressure pump and electric field create a steady stream of droplets that may be recirculated if not deposited, which can raise contamination risks. In drop on demand systems a voltage pulse applied to a piezoelectric actuator induces a pressure wave that ejects a droplet whose volume matches the nozzle diameter. Drop on demand avoids ink recirculation, supports lower-viscosity inks, and reduces thermal stress on biomolecules^[184]. This mode is particularly advantageous for integrating multiple sensitive biological or chemical components on flexible substrates, and has been widely employed in the development of next-generation wearable and photonic devices. For instance, a fully inkjet-printed multiplexed sweat sensor on flexible polyethylene terephthalate (PET) has been reported [Figure 7A-C], in which silver ink forms current collectors and sequentially printed functional inks construct enzymatic layers, mediators, pH sensing elements, and temperature sensing traces^[172]. Integration with flexible circuit modules enables continuous sweat analysis with high sensitivity, low drift, and mechanical robustness.

Figure 7. Illustration of different printing-based fabrication strategies. (A-C) Diagram and photograph of a fully printed wearable sweat sensor array on PET, featuring integrated electrodes for pH, glucose, alcohol, temperature, and reference measurements. Reproduced with permission^[172]. Copyright 2024, Wiley; (D) Schematic of a multi-material extrusion bioprinter, showing motorized sled for automated syringe selection and nozzle rinsing; (E) Build plate placement into bioink droplet forms a controlled filament layer; (D and E) Reproduced with permission^[186]. Copyright 2021, Springer; (F) Workflow for the laser-printed paper-based analytical device: digital pattern design, laser printing and thermal baking to form fluidic barriers, surface functionalization with silver and gold nanoparticles for atrazine recognition, sample application, and smartphone-based colorimetric readout. Reproduced with permission^[196]. Copyright 2023, American Chemical Society. PET: Polyethylene terephthalate; UV: ultraviolent; μPAD: microfluidic paper-based analytical device; NP: nanoparticle.

Extrusion printing is one of the most versatile approaches for fabricating 3D biosensor arrays and tissue scaffolds because it supports the deposition of bioinks with a wide range of viscosities. These inks can include low-viscosity aqueous cell suspensions as well as highly concentrated, cell-laden hydrogels, all while maintaining fine spatial resolution at the micrometer scale^[185]. In a typical setup [Figure 7D and E], bioink is loaded into a syringe or cartridge and extruded through a nozzle onto a substrate in a layer-by-layer manner^[186]. This process is driven by pneumatic pressure, a piston mechanism, or a screw-based system. The resolution and fidelity of the printed structures are determined by several factors, including nozzle diameter, applied extrusion force or pressure, the rheological properties of the bioink, and the movement speed of the printhead or substrate. Compared with droplet-based or light-assisted printing technologies, extrusion systems are capable of handling inks with viscosities ranging from approximately 30 millipascal-seconds to 10⁷ millipascal-seconds, making them especially well suited for printing robust, cell-containing constructs^[187].

A key advantage of extrusion printing is its compatibility with a wide variety of biomaterials and cell types, allowing for the fabrication of multimaterial and multicellular systems. However, the mechanical shear generated during the extrusion process can negatively affect cell viability. To address this, bioinks are often formulated to exhibit strong shear-thinning behavior, meaning that their viscosity decreases under shear stress, which helps reduce damage during printing. Additionally, these inks are designed to rapidly solidify or crosslink after deposition to maintain structural integrity and promote biological activity^[188]. Commonly used formulations include natural and synthetic polymers such as alginate, gelatin methacrylate, and nanocellulose-based materials, which provide the necessary mechanical support and biocompatibility for various biomedical applications^[189,190].

Laser-based bioprinting uses short, high-energy light pulses to deposit live-cell inks onto a target surface without any physical nozzle. A typical system comprises a pulsed laser, focusing optics, a donor ribbon (glass support, thin metal absorber, bioink layer) and a receiving substrate^[191]. When the laser irradiates the metal layer, rapid vaporization creates a localized pressure bubble that ejects a picoliter droplet of bioink toward the substrate^[192]. Because there is no nozzle, this method tolerates viscosities from 1 to 300 mPa·s without clogging and subjects cells to minimal mechanical stress, routinely achieving over 95% viability even at high cell densities^[193,194]. Key factors controlling droplet size and placement include laser energy density, bioink film thickness and viscosity, and the spacing and surface chemistry of the receiver^[195]. Figure 7F illustrates a dual-chromatic laser-printing strategy for fabricating a microfluidic paper-based analytical device^[196]. One wavelength patterns hydrophobic barriers that define channel geometries, while a second wavelength deposits atrazine-sensing reagents into channel interiors. This example demonstrates how laser-based deposition can create complex, multimaterial sensor architectures in a seamless, mask-free process.

Overall, these three printing modalities offer complementary capabilities for biosensor fabrication: inkjet printing excels in high-resolution, mask-free deposition; extrusion printing enables multi-material and 3D structuring; and laser-assisted printing provides precision patterning without nozzle-related limitations.

Photopolymerization techniques for biosensor fabrication

Photopolymerization-based additive manufacturing technologies offer high-resolution, rapid prototyping capabilities that are well suited for creating complex, 3D biosensor architectures. Among these, volumetric printing, stereolithography, and digital light projection printing represent three powerful approaches that differ in curing principles, achievable geometries, and processing speeds.

Volumetric printing utilizes tomographic projection to polymerize the entire volume of photosensitive resin in one step. A series of two-dimensional light patterns, which are computed as the inverse Radon transform of the target 3D object, are projected simultaneously from multiple angles through a rotating resin bath^[197]. The system typically employs a laser source, precision motion stages, and a real-time control module to coordinate light irradiation and resin rotation^[198]. When the cumulative light dose exceeds the threshold, the resin cross links to form the final structure within seconds, with sub millimeter precision and a fully continuous internal architecture^[199]. Since the resin remains fluid until the final exposure, there is no layer by layer “stair stepping” effect, resulting in smooth surfaces and isotropic mechanical properties. The key to volumetric printing lies in the generation of material growth functions (MGFs). These are Gaussian-shaped polymerized footprints induced by laser irradiation that accumulate spatially to form complex geometries. However, volumetric printing currently only supports single material resins and requires photosensitive formulations with narrowly tuned absorption and gelation kinetics^[200].

Stereolithography is an additive manufacturing method in which a focused beam of ultraviolet or visible light induces polymerization in a photosensitive resin, creating 3D parts layer by layer. In a typical setup, a laser source is directed by galvanometer mirrors through a transparent window into a vat of liquid prepolymer. Wherever the light contacts the resin, it triggers rapid crosslinking; after one layer is cured, the build platform lowers or raises to allow fresh resin to flow beneath, and the process repeats until the final structure is complete^[201]. An alternative approach uses a micromirror array to project an entire layer pattern simultaneously, curing a whole plane of material in a single exposure and thereby reducing overall build time^[202]. The lateral resolution is determined by the optics of the projection system and mirror pixel size, while the vertical resolution depends on the absorption and scattering properties of the uncured resin^[203]. This technology achieves sub-10-micrometer precision and allows uncured material to be washed away cleanly. Building on the high precision and versatile material properties afforded by stereolithographic fabrication, Antonelli et al. integrated photopolymerized microstructures into a Lab-on-Chip platform featuring multispectral readout^[204]. The system uses a patterned visible light projector to locally polymerize vertical micropillars within a transparent resin, while a secondary red-filtered illumination path enables real-time monitoring of pillar growth. This combined curing and imaging approach supports in situ assessment of feature formation and facilitates subsequent multispectral biochemical signal analysis through machine learning.

Digital light projection printing uses a dynamic micromirror array (a digital micromirror device, DMD) to project entire cross-sectional images of the 3D model onto the resin surface. This system relies on a programmable light source, a DMD chip for pattern generation, and a printing platform to control layer stacking^[205]. Each exposure cures an entire layer (tens to hundreds of micrometers thick) in fractions of a second, after which the build platform moves to refresh the resin. Lateral resolution is determined by the projected pixel size of the DMD optics, while vertical resolution is governed by the optical penetration depth of the resin and the layer step size^[206]. This approach combines true layer wise control, which enables easy switching between multiple materials, with high throughput, achieving feature sizes ranging from 5 to 300 µm and build speeds of up to tens of millimeters per hour^[207].

For all three methods, material formulation plays a decisive role in performance. Resin viscosity, photoinitiator concentration, and the presence of photoabsorbers or dyes must be balanced to achieve rapid curing, adequate resolution, and minimal scattering. The development of biocompatible and cell-friendly resins, particularly those curable under visible light, is expanding the applicability of these technologies to integrated biosensors and tissue engineering scaffolds^[208,209]. By combining the volumetric speed of tomographic projection, the fine resolution of stereolithography, and the material versatility of digital light projection, researchers can create sophisticated, fully functional biosensor platforms within minutes while maintaining microscale precision.

Layer-by-layer assembly for biosensor integration

Layer-by-layer (LbL) assembly provides a highly versatile and modular fabrication strategy, particularly suitable for constructing flexible and functional sensor interfaces. This technique is based on the sequential adsorption of complementary materials, most commonly oppositely charged polyelectrolytes, onto a substrate. After each adsorption step, a rinsing process is performed to remove unbound species and ensure the precision of the deposited layer^[210]. LbL assembly can be implemented through three main approaches^[211]. In the immersion (or dipping) method, the substrate is alternately immersed into solutions containing the desired materials, with rinsing between immersions to maintain purity. The spraying method employs atomized droplets of each solution to coat the surface sequentially, allowing rapid and uniform deposition over large or non-planar substrates. The microfluidic method channels the solutions over the substrate within confined flow paths, enabling localized patterning, precise control of adsorption time, and efficient reagent use.

The primary advantage of LbL assembly lies in its ability to precisely control film thickness, composition, and biofunctionalization at the nanometer scale, thereby enabling the incorporation of polymers, proteins, nanoparticles, or enzymes into tailored sensing layers. Furthermore, because it typically employs aqueous processing conditions, the method is compatible with preserving biomolecular activity, a critical consideration in biosensor integration. These attributes make LbL assembly a powerful complement to printing technologies, especially in applications requiring fine-tuned surface chemistry, hierarchical architectures, or controlled release of recognition elements in wearable or implantable biosensors^[212].

Roll-to-roll manufacturing for scalable biosensor production

R2R fabrication is a continuous manufacturing technique in which flexible substrates are transferred between supply and take-up reels while passing through a sequence of processing stations. This method supports large-scale and cost-efficient production of flexible biosensors by integrating multiple fabrication steps such as patterning, functionalization, lamination, and cutting into one automated workflow^[213]. The substrate, typically a polymer film, moves through a series of rollers where each station performs a specific process in-line. Patterning can be accomplished by hot embossing, photolithography, or nanoimprinting. Functionalization steps, including reagent deposition or thin-film coating, can be implemented before subsequent lamination or encapsulation. Because the substrate remains in continuous motion, R2R systems can reach production speeds of tens to hundreds of meters per hour, making the approach suitable for high-volume manufacturing of disposable biosensors.

In biosensing applications, R2R technology has been successfully applied to both electrochemical and microfluidic device fabrication^[214]. As shown in Figure 8A and B, R2R gravure printing enables the scalable production of electrochemical biosensors for wearable and medical devices^[215]. This process utilizes rotating cylinders to continuously deposit functional inks onto the flexible substrate. By printing multiple layers sequentially, including silver tracks, carbon working electrodes, and dielectric insulation layers, the system achieves high-throughput fabrication of uniform electrode arrays. This layer-by-layer printing strategy demonstrates the feasibility of manufacturing large-area and low-cost sensors with the precision required for medical diagnostics. Furthermore, as illustrated in Figure 8C, R2R systems can also incorporate microfluidic structures, sensing electrodes, and optical components directly onto flexible polymer substrates^[216]. Hot embossing can define microchannels, die-cutting can create fluidic inlets and outlets, and automated dispensing can deposit biorecognition elements. After these steps, lamination seals the microfluidic network, and auxiliary components such as membranes or reservoirs can be added during the same production run. The integration of these processes reduces manual assembly, minimizes variability, and enables scalable manufacturing suitable for clinical diagnostics and environmental monitoring.

Figure 8. Representative fabrication strategies and device architectures for R2R printed and integrated microfluidic biosensors. (A) Schematic of R2R gravure printing for electrochemical sensors, showing the layer-by-layer deposition of functional inks to form electrode arrays on a flexible substrate; (B) Image of the R2R gravure printing system. (A and B) Reproduced with permission^[215]. Copyright 2018, American Chemical Society; (C) Illustration of the R2R hot embossing process for the continuous patterning of microfluidic channels on polymer films. Reproduced with permission^[216]. Copyright 2020, American Chemical Society. PET: Polyethylene terephthalate; R2R: roll-to-roll.

Multimodal data fusion strategies: glucose monitoring as a paradigm

To illustrate the practical implementation of AI architectures, we focus on non-invasive glucose monitoring as a representative paradigm^[220-223]. Glucose detection exemplifies the need for multimodal data fusion to overcome the limitations of single-analyte sensing. Effective fusion strategies are pivotal for handling heterogeneous data and are generally categorized into three levels: Data-Level (Early) Fusion for raw signal concatenation; Feature-Level (Intermediate) Fusion for merging extracted attributes; and Decision-Level (Late) Fusion for aggregating independent model predictions.

In this context, signal processing techniques are essential to extract meaningful data from raw sensor signals before fusion. For Level 2 on-device processing, techniques like wavelet packet decomposition are used to decompose signals into different frequency sub-bands^[224], while empirical mode decomposition to capture nonlinear and non-stationary components^[225,226]. Additionally, spectral filtering to suppress noise and physiological artifacts so that weak analyte signatures become measurable. The general process of sensing and classification involves several key stages, including data collection, acquisition and preprocessing, feature engineering, fusion and modeling, training, and finally evaluation analysis. During feature engineering (Intermediate Fusion), researchers extract spectral and temporal features such as peaks, trends, and learned embeddings with ML models including autoencoders and CNNs. Principal component analysis (PCA) is often employed to reduce redundancy and maintain interpretability. In the final stage (Level 3 Prediction), multimodal fusion integrates heterogeneous sources and ML models such as XGBoost, LSTM, and graph neural networks are used for prediction or classification, with cross-validation and bootstrapping ensuring generalizable performance.

In addition to classification, regression algorithms are used to predict continuous glucose concentrations. Representative examples include gaussian process regression for aqueous benchmarks^[227], and random forest or compact CNNs for estimating fasting blood glucose from real samples^[228]. These models, integrated with microstrip microwave sensors, allow for accurate predictions of glucose concentrations by capturing spectral features and repeatability behaviors. Furthermore, deep learning models such as LSTM networks are increasingly used to forecast future values based on past sensor data. Bayesian optimization fine-tunes hyperparameters like lookback windows and learning rates to optimize model performance^[229].

Critical challenges and validation analysis

While AI-enabled biosensing shows immense promise, rigorous validation and critical analysis of current limitations are essential for clinical translation. Evaluation metrics for glucose prediction and classification in diabetes care using various AI methods include root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) for regression tasks, with accuracy, precision, and AUC-ROC used for classification tasks^{[228,227,230,231]}. Supplementary Table 1 [Supplementary Materials] summarizes various AI-powered glucose sensors utilizing microwave, optical, and electrochemical technologies to enhance diabetes monitoring. As shown in Supplementary Table 1, most mentioned studies formulated glucose estimation as a regression task because blood glucose is inherently a continuous variable. Results showed high accuracy in controlled settings, with some models achieving over 99% classification accuracy or strong regression performance.

However, a critical analysis reveals significant gaps remaining for effective implementation. As noted in Supplementary Table 1, most experiments were limited by small sample sizes or synthetic datasets, leading to a severe domain shift problem where models trained on controlled lab data often fail in real-world scenarios due to user variability and environmental noise (e.g., temperature and humidity)^[232,233]. Consequently, robustness to motion artifacts remains a primary hurdle that necessitates hardware-aware algorithms at Level 2. Beyond signal fidelity, the “black-box” nature of deep learning models (e.g., CNNs, LSTMs) raises concerns regarding clinical trust. Although techniques like shapley additive explanations help reduce redundancy, developing fully explainable AI is required to visualize feature importance for clinicians to verify autonomous predictions. Finally, from a deployment perspective, transmitting high-resolution physiological data to the cloud exposes sensitive user information. Future systems must therefore adopt federated learning to train models locally on devices without sharing raw data, balancing personalized healthcare with rigorous privacy and security preservation.