Flexible Micro-Neural Interface Devices: Advances in Materials Integration and Scalable Manufacturing Technologies

Abstract

Flexible microscale neural interfaces are advancing current strategies for recording and modulating electrical activity in the brain and spinal cord. The aim of this review is to colligate recent progress in thin-film micro-electrocorticography (μECoG) systems and establish a framework for their translation toward spinal bioelectronic implants. We first outline substrate and electrode material design, ranging from polymeric and hydrogel-based materials to nanostructured conductive materials that enable high-fidelity recording on mechanically compliant platforms. We then summarize structural design rules for μECoG arrays, including electrode size, pitch, and channel scaling, and relate these to data-driven μECoG applications in brain–computer interfaces and closed-loop neuromodulation. Bidirectional μECoG architectures for simultaneous stimulation and recording are examined, with emphasis on safe charge injection, electrochemical and thermal limits, and state-of-the-art hardware and algorithmic strategies for stimulation-artifact suppression. Building upon these cortical technologies, we briefly describe adaptation to spinal interfaces, where anatomical constraints demand optimized mechanical properties. Finally, we discuss the convergence of flexible bioelectronics, wireless power and telemetry, and embedded AI decoding as a path toward autonomous, clinically translatable μECoG and spinal neuroprosthetic systems. Ultimately, by synthesizing these multidisciplinary advances, this review provides a strategic roadmap for overcoming current translational barriers and realizing the full clinical potential of soft bioelectronics.

1. Introduction

Understanding and engineering the interaction between the nervous system and external devices has emerged as one of the most innovative and transformative research domains in modern neurotechnology. Neural activity unfolds on millisecond timescales and spans multilayered spatial architectures, from individual neurons to broadly distributed networks. This intricate combination of temporal precision, anatomical heterogeneity, and multiscale organization imposes stringent engineering constraints on neural interfaces, which must reliably detect, transmit, and modulate electrical activity within the soft, dynamic, and highly interconnected environment of the brain. In this context, the constituent materials of the device play a decisive role, as their mechanical and electrochemical properties must be carefully engineered to minimize tissue reactions and ensure stable coupling with the neuronal microenvironment. By satisfying these critical requirements, microscale neural electrodes have evolved to support high-precision recording of bioelectrical activity and modulation of biological function through controlled ion-mediated electrical perturbations. The advent of semiconductor fabrication techniques gave rise to rigid silicon-based microelectrode arrays, such as the Utah array and Michigan probes, which laid the foundation for high-resolution electrophysiological recording [1,2,3]. Nevertheless, the fundamental mechanical mismatch between their rigidity and that of soft neural tissue has revealed persistent limitations, including chronic inflammation induced by micromotion, glial scar encapsulation, and progressive signal degradation phenomena, consistently observed across animal models [4,5,6,7,8,9,10].

These shortcomings have catalyzed a paradigm shift toward flexible, ultrathin polymer-based microelectrodes that better conform to the mechanical compliance and geometric scales of neural tissue [11,12]. Such devices have dramatically improved long-term biocompatibility, minimized tissue damage, and sustained stable electrophysiological access [13,14,15]. Within this technological progress, flexible micro-neural interfaces, particularly thin-film micro-electrocorticography (μECoG) arrays, have risen to prominence owing to their extensive spatial coverage, multimodal integration capability, and superior chronic stability [16,17,18]. Unlike intracortical electrodes optimized for single-neuron precision, μECoG arrays readily offer the unique advantage of stably capturing mesoscale population activity across broad cortical regions while minimizing invasiveness and mechanical burden [19]. Owing to these advantages, μECoG systems are predominantly evaluated in vivo, where their long-term stability and functional utility are validated through chronic recording of low-frequency local field potential (LFP) signals, which reflect the aggregate synaptic activity of surrounding neuronal populations within physiologically relevant cortical environments. The incorporation of advanced materials, such as graphene, PEDOT derivatives, nanostructured carbon, liquid crystal polymer, and parylene-based ultrathin substrates, has further reduced impedance, enhanced optical transparency, suppressed artifacts, and accelerated high-density multimodal signal fusion, thereby substantially expanding the potential of μECoG technology. These advances have positioned μECoG as a central platform for next-generation high-resolution brain-machine interfaces (BMIs), closed-loop neuromodulation, and emerging neuroimaging modalities (Figure 1) [18,20,21,22,23].

Despite these developments, significant barriers still remain. Robust algorithmic resilience against noise, baseline drift, and long-term variability is essential for stable decoding of population activity, while ensuring chronic bio-stability at the electrode/tissue interface continues to pose a formidable challenge. Fully implantable wireless systems demand low-power consumption, extreme miniaturization, safe power delivery, bidirectional telemetry, and seamless integration with optogenetics, ultrasound, chemical sensing, and soft electronics [24,25,26,27]. Moreover, as μECoG applications extend beyond the cortex to spinal cord interfaces and peripheral nerve prosthetics, device geometry and substrate mechanics must contend with constrained surgical corridors, persistent micromotion, high-curvature anatomy, and cerebrospinal fluid pressure gradients, which present critical new engineering challenges [28,29]. This review specifically focuses on the engineering advancements of flexible micro-neural interfaces, aiming to bridge the technological gap between cortical μECoG systems and spinal bioelectronics. In this context, we categorize current progress into three core technical domains: material innovation, structural design, and system-level integration. First, we emphasize the critical role of soft, adaptive materials, ranging from hydrogels to nanostructured conductors, in optimizing the mechanical and electrochemical parameters at the neural interface. Second, the review explores the design principles of high-density arrays for bidirectional applications, offering a comprehensive overview of stimulation safety limits and artifact suppression strategies. Expanding the scope of bioelectronic design, we transfer a concept of cortical technologies to spinal cord interfaces, addressing the unique mechanobiological constraints required for locomotor restoration. Finally, in the discussion of prospects, this review outlines a newly developed approach for autonomous neuroprosthetics by combining emerging technologies in wireless telemetry and embedded AI decoding.

Figure 1. Conceptual illustration of the technological transition from conventional macro-ECoG systems to next-generation μECoG neural interfaces. Traditional clinical macro-ECoG grids, while robust for surgical mapping, remain limited by millimeter-scale electrode spacing, bulky form factors, and low spatial resolution. Reproduced with permission from [30]. Copyright Wiley-VCH, 2017. Advances in materials integration, ultrathin substrates, and scalable microfabrication have enabled a progressive miniaturization toward micrometer-scale, high-density μECoG arrays that conform intimately to the cortical surface. Reproduced with permission from [31]. Copyright The American Association for the Advancement of Science, 2022. This transition supports precision BCI/BMI applications by offering enhanced neural fidelity, improved signal-to-noise ratios (SNR), and minimally invasive implantation profiles; these developments redefine ECoG technology as a platform for high-resolution neural decoding, closed-loop neuromodulation, and long-term biointegrated neuroprosthetic systems.

Figure 1. Conceptual illustration of the technological transition from conventional macro-ECoG systems to next-generation μECoG neural interfaces. Traditional clinical macro-ECoG grids, while robust for surgical mapping, remain limited by millimeter-scale electrode spacing, bulky form factors, and low spatial resolution. Reproduced with permission from [30]. Copyright Wiley-VCH, 2017. Advances in materials integration, ultrathin substrates, and scalable microfabrication have enabled a progressive miniaturization toward micrometer-scale, high-density μECoG arrays that conform intimately to the cortical surface. Reproduced with permission from [31]. Copyright The American Association for the Advancement of Science, 2022. This transition supports precision BCI/BMI applications by offering enhanced neural fidelity, improved signal-to-noise ratios (SNR), and minimally invasive implantation profiles; these developments redefine ECoG technology as a platform for high-resolution neural decoding, closed-loop neuromodulation, and long-term biointegrated neuroprosthetic systems.

2. Architecture and Applications of Biointerfaces for μECoG Recording

2.1. Material Components of Neural Electrodes for μECoG Recording

Neural electrode arrays designed for μECoG recording constitute the core biointerface for cortical signal acquisition. These arrays must maintain stable and conformal contact with brain tissue to provide high electrical fidelity. To achieve this, μECoG arrays generally comprise a biocompatible substrate and conductive electrode sites, each contributing distinct physicochemical properties that collectively determine device performance, signal quality, and long-term recording stability. The substrate governs mechanical stability, conformal cortical contact, and the overall electro-mechanical behavior of the array, whereas the electrode material directly influences electrical sensitivity, electrochemical performance, and impedance characteristics. Consequently, both the substrate and electrode must be engineered with precision to ensure mechanical compliance, electrochemical reliability, and biological stability within the soft cortical environment.

2.1.1. Guidelines for Substrate Material Selection

The substrate of a μECoG electrode array is a main determinant of the system’s mechanical flexibility, conformal adhesion to cortical tissue, and long-term recording stability. Importantly in this field of study, achieving mechanical compliance that matches brain tissue enables conformal adhesion, which is fundamental for high-fidelity signal acquisition. In this role, the substrate serves as a vital structural anchor for the conduction pathways (e.g., metals, conductive polymers). It ensures that these signaling materials maintain stable, intimate contact with the neural interface, thereby preventing signal degradation caused by micromotion or tissue reaction and guaranteeing long-term recording stability over extended spatiotemporal scales. Thus, selecting an appropriate substrate requires balancing two key requirements: (i) mechanical compliance closely matched to brain tissue, and (ii) chemical and structural robustness compatible with microfabrication processes. These factors involve trade-offs; therefore, substrate selection must be aligned with the intended application, implantation duration, and analytical purpose of the μECoG device. Figure 2a summarizes a representative material framework for substrate selection, comparing the elastic modulus and fabrication compatibility of common classes of soft substrates, including polymer films, elastomers, and hydrogels, used in cortical interfaces. Conventional polymeric materials typically provide high structural stability; hydrogels offer exceptional compliance, and elastomers occupy an intermediate regime with both softness and manufacturability. Historically, polyimide (PI) and parylene-C have been the most widely adopted materials as a substrate for μECoG devices, due to their outstanding chemical stability, compatibility with semiconductor microfabrication, and excellent dielectric properties in biofluidic environments [23,32,33,34]. However, their relatively high elastic modulus necessitates extreme thinning, often on the order of several to tens of micrometers to minimize mechanical mismatch with brain tissue [35,36]. A representative example is the 16-channel μECoG array fabricated on a ~12.5 μm PI film, presented in Figure 2b, in which sputtered Au electrodes maintained stable recordings even on highly curved cortical surfaces in freely moving mice [37]. Currently, PI film remains one of the most reliable materials for μECoG platforms, owing to its robust patternability, thermal stability, and process compatibility [34,38].

A similar approach has also been demonstrated using parylene-C films (Figure 2c), where parylene-C substrates (~10 μm) were used for inkjet-printed electrodes. The resulting electrode arrays exhibited partial but sufficient conformity to the cortical curvature and enabled stable μECoG recordings for over 2 weeks in murine cortex, supported by the intrinsic biocompatibility of parylene-C [23]. Because conventional polymer substrates still exhibit relatively high stiffness, recent studies have shifted toward elastomeric materials, such as polydimethylsiloxane (PDMS), Ecoflex, and styrene-ethylene-butylene-styrene (SEBS), as μECoG substrates [13,39,40,41]. These materials possess low elastic moduli in the MPa range with high elongation ability, allowing far greater mechanical matching with neural tissue. For example, Figure 2d shows a stretchable μECoG array constructed on a PDMS film (~75 μm), which significantly reduced interfacial stresses due to its inherent softness [40]. A more advanced example, shown in Figure 2e, employed a porous SBS fibrous-mat substrate (~25 μm) with an ultra-low modulus comparable to that of brain tissue, thereby minimizing mechanical mismatch at the tissue–electrode interface. Liquid-metal microelectrodes and interconnects selectively formed on the SBS fibers conformed stably to the porous scaffold and withstood extreme tensile deformation (up to ~1500% strain), far exceeding the stretchability of conventional Au/PI electrodes while maintaining reliable in vivo ECoG function for up to 32 weeks [13]. Although elastomer substrates improve mechanical conformity and signal acquisition, they introduce new challenges. Their viscoelastic nature and low structural rigidity complicate precise micro-patterning and reduce interfacial adhesion with metal electrodes, which may degrade long-term durability during repeated deformation in μECoG operation [42,43]. Therefore, developed microscale architecture often requires additional structural reinforcement, optimized encapsulation layers, or modified fabrication workflows. Indeed, new types of hydrogels represent an emerging class of substrates offering the closest mechanical match to brain tissue. Synthetic hydrogels typically exhibited elastic moduli in the tens-to-hundreds-of-kPa range and water contents of 70–90%, maintaining tissue-like hydration. These unique characteristics minimize interfacial friction and reduce inflammatory responses, improving long-term recording stability [44,45,46,47,48]. However, the same softness and high water content create fabrication challenges, such as dimensional shrinkage during drying, micro-cracking in the surface structure, limited micropatterning precision, and poor adhesion with metal electrodes. As a result, hydrogel-based μECoG substrates require careful design of interfacial bonding, modulus matching, and hybrid material integration.

Figure 2. Representative material classes and device architectures used in flexible and stretchable μECoG interfaces. (a) Comparative map of candidate substrate materials for cortical interfaces, illustrating the trade-off between mechanical property matching to soft neural tissue (vertical axis) and process/structural stability (horizontal axis). Hydrogels (E ≈ 100 kPa), elastomeric substrates (E ≈ 1 MPa), and plastic/polymer film substrates (E ≈ 1 GPa) are grouped, with squares indicating representative literature examples [23,37,40,41,47,48,49,50]. (b) PI-based multichannel μECoG electrode array with a bifurcated flap geometry for subcranial insertion and hemispheric coverage, demonstrating high pattern fidelity and robust microfabrication compatibility. Reproduced with permission from [37]. Copyright American Chemical Society, 2016. (c) Inkjet-printed flexible μECoG array on an ultrathin Parylene-C film, highlighting additive manufacturing compatibility and enhanced conformability. Reproduced with permission from [23]. Copyright MDPI, 2022. (d) Stretchable 32-channel electrode array embedded on a PDMS elastomer substrate, enabling large-strain deformation while maintaining electrical continuity. Reproduced with permission from [40]. Copyright Wiley-VCH, 2018. (e) Flexible μLME ECoG electrode array (thickness, 25 μm) fabricated on a porous SBS fibrous-mat substrate, demonstrating conformal attachment onto the soft, curved and sophisticated cerebral cortex. Reproduced with permission from [13]. Copyright American Association for the Advancement of Science, 2023. (f) Conformal adhesion of a μECoG array supported on a hydrogel-like bacterial cellulose substrate, shown on an agar gel cylinder engineered to mimic cortical curvature. Reproduced with permission from [47]. Copyright Springer Nature, 2023. (g) Hydrogel scaffold microelectrode array (MEA) connected to a flexible printed circuit (FPC) and conformally attached to human skin, demonstrating the ultrasoft mechanical compliance of the fully hydrogel-based neural interface. Reproduced with permission from [51]. Copyright Wiley-VCH, 2023.

Figure 2. Representative material classes and device architectures used in flexible and stretchable μECoG interfaces. (a) Comparative map of candidate substrate materials for cortical interfaces, illustrating the trade-off between mechanical property matching to soft neural tissue (vertical axis) and process/structural stability (horizontal axis). Hydrogels (E ≈ 100 kPa), elastomeric substrates (E ≈ 1 MPa), and plastic/polymer film substrates (E ≈ 1 GPa) are grouped, with squares indicating representative literature examples [23,37,40,41,47,48,49,50]. (b) PI-based multichannel μECoG electrode array with a bifurcated flap geometry for subcranial insertion and hemispheric coverage, demonstrating high pattern fidelity and robust microfabrication compatibility. Reproduced with permission from [37]. Copyright American Chemical Society, 2016. (c) Inkjet-printed flexible μECoG array on an ultrathin Parylene-C film, highlighting additive manufacturing compatibility and enhanced conformability. Reproduced with permission from [23]. Copyright MDPI, 2022. (d) Stretchable 32-channel electrode array embedded on a PDMS elastomer substrate, enabling large-strain deformation while maintaining electrical continuity. Reproduced with permission from [40]. Copyright Wiley-VCH, 2018. (e) Flexible μLME ECoG electrode array (thickness, 25 μm) fabricated on a porous SBS fibrous-mat substrate, demonstrating conformal attachment onto the soft, curved and sophisticated cerebral cortex. Reproduced with permission from [13]. Copyright American Association for the Advancement of Science, 2023. (f) Conformal adhesion of a μECoG array supported on a hydrogel-like bacterial cellulose substrate, shown on an agar gel cylinder engineered to mimic cortical curvature. Reproduced with permission from [47]. Copyright Springer Nature, 2023. (g) Hydrogel scaffold microelectrode array (MEA) connected to a flexible printed circuit (FPC) and conformally attached to human skin, demonstrating the ultrasoft mechanical compliance of the fully hydrogel-based neural interface. Reproduced with permission from [51]. Copyright Wiley-VCH, 2023.

As one interesting approach, a distinct biomaterial-based approach is shown in Figure 2f, where bacterial cellulose (BC) serves as a hydrated, nanofibrous substrate. The naturally formed 3D network of cellulose nanofibers (i.e., tens of nanometers in diameter) provides high water uptake, ultrathin flexibility, and hydrogel-like mechanics. Hydrated BC films (~1.4 mm thickness, modulus 80–120 kPa, swelling ratio > 7500%) sustain long-term moisture retention at the tissue interface. In vivo studies demonstrated stable acute and chronic μECoG recordings without signal-to-noise ratio (SNR) degradation, highlighting BC as a promising next-generation substrate [47]. Another emerging platform is the all-hydrogel biohybrid neural interface shown in Figure 2g, an all-hydrogel microelectrode array (MEA) that integrates an adhesive gelatin/silk hydrogel substrate crosslinked with microbial transglutaminase (GS-MTG), extracellular-matrix (ECM)-based, double-cross-linked PEDOT-doped graphene oxide electroconductive hydrogel tracks (PDGO), and an ultrathin poly(lactic acid) (PLA) passivation layer. The all-hydrogel MEA exhibits a nerve-tissue-like modulus (~4 kPa), mechanical compliance under tensile, twisting, and bending deformations, and high ionic/electronic conductivity (σ_wet ≈ 58.5 S m⁻¹) with low electrode impedance (~1.7–2.3 kΩ at 1 kHz) and a charge storage capacity of ~24 μC cm⁻². Impressively, the device maintained its electrochemical performance over 100 cyclic-voltammetry cycles and 28 days of in vitro degradation, and in vivo enabled successive peripheral nerve recording and electrical stimulation that synergistically accelerated motor-function recovery before completely disintegrating within ~1 month [51]. Overall, these studies highlight that hydrogel-based substrates with ultralow modulus, high hydration, and conformal adhesion offer a compelling solution to the mechanical mismatch and interface instability observed in traditional polymer and elastomer substrates. Nevertheless, their high water content, mechanical fragility, drying-induced shrinkage, metal/hydrogel adhesion issues, and fabrication-compatibility constraints indicate that further advancement in materials synthesis, patterning strategies, and hybrid integration is still required for widespread adoption in μECoG systems.

2.1.2. Guidelines for Selection and Application of Electrode Materials

Microscale electrodes embedded within ECoG arrays constitute electrically active sites that directly interface with cortical tissue. These electrodes mediate charge transfer at the electrode/tissue interface, thereby governing key electrochemical interactions essential for neural signal acquisition. The intrinsic material properties of the bioelectrode, particularly its stability, electrochemical impedance, and biocompatibility, play a decisive role in determining recording accuracy, long-term reliability, and stimulation efficiency. Noble metals, such as Au and Pt, are the most widely used electrode materials for ECoG systems, typically configured as planar micropatterned structures [31,34,52,53,54,55]. Under these conditions, the impedance characteristics of cortical electrodes can be approximated using the equivalent circuit model [49]:

$$\left|{\rm Z}\right| = {\displaystyle \frac{1}{A}}\left({\displaystyle \frac{R}{\sqrt{1+{\omega }^2{R}^2{C}^2}}}\right)$$

(1)

where A is the electrode area, R is the charge-transfer resistance, C is the double-layer capacitance, and ω is the angular frequency. For useful μECoG electrodes, miniaturization inevitably reduces the geometric area of electrode arrays, leading to higher impedance, so thus, selecting electrode materials with sophisticated designing architectures that maintain a microscale footprint while suppressing impedance elevation is a critical challenge. To overcome these constraints, a growing research area has incorporated advanced nanomaterials and combinatorial coating with conducting polymers, where most notably graphene and poly(3,4-ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS) have been notably adopted to enhance electrochemical performance [56,57,58,59,60].

Graphene, a two-dimensional carbon material with exceptional carrier mobility and chemical stability, simultaneously offers high electrical conductivity, flexibility, and optical transparency. Williams et al. reported a graphene-based 16-channel μECoG array (Figure 3a), demonstrating that the material’s transparency and mechanical compliance enabled stable neural recordings even under optical stimulation environments, supporting applications requiring concurrent electrophysiology and optogenetics [59]. Importantly, PEDOT:PSS has also emerged as a leading candidate for neural electrode materials due to its mixed ionic-electronic conductivity, which enhances charge transfer across the electrode/tissue interface. In one representative design (Figure 3b), patterned PEDOT:PSS electrodes (i.e., 50 μm diameter, 56-channel, 3 mm microdot sites) were integrated on Au/parylene-C substrates. The resulting electrodes exhibited relatively low impedance (~12–13 kΩ at 1 kHz), approximately an order of magnitude lower than that of conventional Pt electrodes (~300 kΩ), representing the significant electrochemical benefits of conductive polymer coatings [60]. Beyond material substitution, surface modification strategies also provide an additional route for reducing intrinsic impedance. Simple yet effective approaches such as introducing porosity, nanoscale roughness, or hierarchical topographies increase the effective surface area, thereby enhancing double-layer capacitance and lowering impedance. A compelling example is the Au nanonetwork (Au NN) electrode, as shown in Figure 3c. Here, PMMA nanofiber meshes, fabricated via an electrospinning process, served as a template, onto which Au was sputter-deposited to yield a conductive nanofibrous network. This networked structure increased the active surface area by ~1.3×, compared to planar Au electrodes, correspondingly reducing impedance at 1 kHz (e.g., ~11.8 kΩ vs. ~19.8 kΩ). The nanofiber network also imparted partial permeability and improved mechanical flexibility, characteristics advantageous for conformal cortical interfacing [61]. As presented above, advances in nanoscale materials engineering, conductive polymer integration, and hierarchical surface structuring have substantially progressed the design space for μECoG electrode optimization. These strategies effectively mitigate impedance penalties associated with miniaturization and offer new opportunities for high-density, high-fidelity neural biointerfaces.

Recent advances in electrode materials have also introduced hybrid systems that combine liquid metals with elastomeric substrates [55,62]. As illustrated in Figure 3d, liquid metal/polymer conductors (MPCs) represent a prominent class of stretchable μECoG electrodes, in which liquid metal microdroplets are embedded within an elastomer matrix, such as PDMS. When the MPC surface was subsequently coated with a ~200 nm Pt layer, nanoscale wrinkled topographies with an average roughness of ~200 nm were formed. This mechanically induced hierarchical morphology substantially increases the electroactive surface area of the electrode, enabling improved charge-transfer characteristics. Experimentally evidenced, this structural enhancement reduced the impedance from ~2.7 MΩ to ~250 kΩ at 1 kHz, marking an order-of-magnitude improvement in electrochemical performance and demonstrating the potential of MPC-based platforms for next-generation flexible neural interfaces [55]. Another nanomaterial-based strategy is exemplified by the carbon nanotube (CNT) electrodes shown in Figure 3e. In this approach, CNT dispersions were spray-coated onto PDMS substrates to form uniform ultrathin films, followed by low-temperature annealing to yield a porous CNT network. The resulting random CNT mesh maintained the intrinsic high conductivity of CNTs while forming interconnected ion-permeable pathways that facilitate deep electrolyte penetration. This porous morphology effectively increased the electrochemically active surface area for each electrode and promoted double-layer formation at the electrode/electrolyte interface. The optimized CNT electrodes achieved an impedance of ~200 kΩ at 1 kHz, which is more than 8-fold lower than that of monolayer graphene electrodes of similar dimensions, highlighting the benefits of CNT-based hierarchical porosity [63].

Beyond planar thin-film designs, three-dimensional microstructuring has emerged as a powerful means of lowering electrode impedance [64,65]. A notable example is the laser-induced graphene (LIG) electrode depicted in Figure 3f. Using UV-based laser irradiation, porous graphene architectures were directly written onto a liquid crystal polymer (LCP) substrate, which offers excellent biocompatibility and moisture stability. The laser-induced local heating produced a 3D porous graphene network with nanoscale pore sizes (i.e., ~200 nm), resulting in a markedly enlarged electroactive surface. Electrochemical activation measurements in PBS confirmed that electrolyte ions (Na⁺, K⁺, Cl⁻) readily infiltrated the porous graphene matrix, exhibiting pronounced pseudocapacitive behavior indicative of enhanced charge-storage and charge-injection capability [65]. As demonstrated recently, LIG-based electrodes can be one of the most promising materials for neural interface mainly due to a synergistic combination of low impedance, high charge-injection efficiency, and stable electrochemical performance, positioning them as a compelling candidate for high-performance brain-interfaced devices. Fundamentally, neural recording involves the transduction of ionic currents (i.e., LFPs) into electronic signals via capacitive or Faradaic mechanisms at the electrode-electrolyte interface. To maximize the efficiency of this transduction, these nanoscale features are typically implemented on flexible substrates using surface modification strategies such as electrochemical deposition, spray coating of nanomaterial dispersions, or direct laser structuring. These processes integrate nanostructured morphologies onto the compliant polymer base, thereby significantly increasing the effective electrochemical surface area. Consequently, this structural modification drastically reduces interfacial impedance and enhances the SNR without compromising intrinsic flexibility. However, each material strategy presents distinct trade-offs rather than offering a universal solution. For instance, while noble metals provide established stability, emerging organic conductors offer superior impedance characteristics but face challenges in long-term durability. Therefore, the final material selection requires a strategic balance between electrochemical performance, mechanical robustness, and the intended lifespan of the implant.

Figure 3. Representative material platforms and microscale architectures used to engineer low-impedance, high-fidelity μECoG electrode arrays. (a) Transparent graphene-based μECoG array illustrating ultrathin, flexible construction (left) and a magnified microscopic view showing patterned graphene recording sites (right). Reproduced with permission from [59]. Copyright Springer Nature, 2014. (b) PEDOT:PSS-based neural electrode array featuring microdot configurations, with an optical micrograph of a 56-microdot array and an SEM cross-section, highlighting the multilayer structure and the PEDOT:PSS interface over Ti/Au and Parylene-C layers. Reproduced with permission from [60]. Copyright Wiley-VCH, 2018. (c) Au nanonetwork μECoG array, showing a transparent electrode layout suitable for mouse cortical recordings (left), and SEM visualization of an individual nanonetwork electrode (right), revealing the electrospun-template-derived fibrous nanostructure. Reproduced with permission from [61]. Copyright Wiley-VCH, 2020. (d) Stretchable neural electrode array fabricated using a liquid-metal-polymer conductor composite (left), accompanied by SEM imaging of Pt-coated recording traces (middle) and a 3D AFM topography map (right), demonstrating nanoscale roughness that increases the electroactive surface area. Reproduced with permission from [55]. Copyright Wiley-VCH, 2021. (e) CNT-based μECoG array on a PDMS substrate (left), with an optical image of the array layout and an SEM image (right) showing the porous, web-like CNT network structure that facilitates ion accessibility and low impedance. Reproduced with permission from [63]. Copyright American Chemical Society, 2018. (f) SEM images of patterned LIG microelectrodes formed directly on an LCP substrate (left), and an enlarged SEM view (right) emphasizing the 3D nanoporous morphology that enhances charge-transfer characteristics. Reproduced with permission from [65]. Copyright Wiley-VCH, 2024.

Figure 3. Representative material platforms and microscale architectures used to engineer low-impedance, high-fidelity μECoG electrode arrays. (a) Transparent graphene-based μECoG array illustrating ultrathin, flexible construction (left) and a magnified microscopic view showing patterned graphene recording sites (right). Reproduced with permission from [59]. Copyright Springer Nature, 2014. (b) PEDOT:PSS-based neural electrode array featuring microdot configurations, with an optical micrograph of a 56-microdot array and an SEM cross-section, highlighting the multilayer structure and the PEDOT:PSS interface over Ti/Au and Parylene-C layers. Reproduced with permission from [60]. Copyright Wiley-VCH, 2018. (c) Au nanonetwork μECoG array, showing a transparent electrode layout suitable for mouse cortical recordings (left), and SEM visualization of an individual nanonetwork electrode (right), revealing the electrospun-template-derived fibrous nanostructure. Reproduced with permission from [61]. Copyright Wiley-VCH, 2020. (d) Stretchable neural electrode array fabricated using a liquid-metal-polymer conductor composite (left), accompanied by SEM imaging of Pt-coated recording traces (middle) and a 3D AFM topography map (right), demonstrating nanoscale roughness that increases the electroactive surface area. Reproduced with permission from [55]. Copyright Wiley-VCH, 2021. (e) CNT-based μECoG array on a PDMS substrate (left), with an optical image of the array layout and an SEM image (right) showing the porous, web-like CNT network structure that facilitates ion accessibility and low impedance. Reproduced with permission from [63]. Copyright American Chemical Society, 2018. (f) SEM images of patterned LIG microelectrodes formed directly on an LCP substrate (left), and an enlarged SEM view (right) emphasizing the 3D nanoporous morphology that enhances charge-transfer characteristics. Reproduced with permission from [65]. Copyright Wiley-VCH, 2024.

2.2. Structural Design Considerations for Neural Electrodes in ECoG Recording

2.2.1. Electrode Size Optimization for Signal Acquisition

In the development of high-density μECoG devices, securing low-impedance electrodes, as discussed in the previous section, is a fundamental requirement for stable neural signal recording. Typically, μECoG electrodes employ planar metal structures; however, as the geometric area of the electrode decreases, the electrochemical impedance increases sharply, often leading to degradation of the SNR. Experimentally, Lee et al. demonstrated that when the diameter of Au-based μECoG electrodes was reduced from 120 μm to 70 μm and 20 μm, the corresponding impedance at 1 kHz increased from 97.7 kΩ to 212 kΩ and 1.3 MΩ, respectively. This result provides direct evidence that reducing electrode size diminishes the effective surface area contributing to charge transfer, thereby increasing the interface impedance at the electrode/brain tissue boundary [52]. Such findings establish an important design guideline for future μECoG systems, underscoring the need to balance spatial resolution with electrochemical performance. Thus so far, to overcome these size-dependent limitations, electrode design strategies increasingly combine optimized microscale geometries with advanced materials and surface engineering. This includes (i) selecting an appropriate electrode size that satisfies both spatial resolution and impedance constraint, (ii) incorporating highly conductive and biocompatible materials, and (iii) employing surface modification techniques to increase the effective surface area without altering the natural properties of the electrode [39,56,66]. These approaches collectively aim to decouple the trade-off between electrode miniaturization and impedance by engineering the interface rather than solely relying on geometric scaling. However, implementing such strategies requires careful evaluation of the effective surface area, electrochemical properties, and mechanical stability as a function of electrode size and structure. The associated trade-offs, such as increased structural complexity, potential fabrication challenges, and long-term reliability, must be systematically characterized.

Conventional and emerging methods to expand the electroactive surface area include the use of metal nanostructures, CNT networks, Pt black, and other porous or roughened electrode coatings. These porous architectures have been shown to increase charge injection capacity and effectively reduce the interface impedance between the electrode and neural tissue compared to flat planar electrodes [31,39,65,67,68,69]. Recent important experimental performance metrics for these approaches are summarized in

Table 1. Summary of electrode dimensions, including geometric area, pitch, and channel-count of representative μECoG electrode arrays, along with the corresponding electrode materials and impedance values measured at 1 kHz.

Electrode Area	Pitch	Channel-Count	Electrode Material	Impedance @ 1 kHz	Ref.
20/70/120 μm (in diameter)	600 μm	16	Au	1.3 MΩ/212 kΩ/97.7 kΩ	[52]
100 × 100 μm	750 μm	16	Au	26.6 ± 0.2 kΩ	[37]
~560 μm (in diameter)	200 μm	202	Au/PEDOT:PSS	1.1 ± 0.2 kΩ	[70]
100 × 100/200 × 200 μm	550/1250 μm	64/256	Au/CNT	15 kΩ	[67]
60 μm (in diameter)	~700 μm	14	Au/PEDOT:PSS + MWCNT	20 kΩ	[39]
200 μm (in diameter)	700 μm	16	Au NN	11.8 kΩ	[61]
50 × 50 μm	295 μm	1152	Au/Pt black	26 ± 7 kΩ	[33]

, highlighting the impact of electrode size, surface structuring, and material choice on impedance, charge storage capacity, and overall neural recording quality. Within this criterion, adherence to conservative charge density thresholds is essential to prevent tissue damage or electrode degradation because μECoG electrodes typically possess smaller geometric areas than standard clinical recording systems. Furthermore, minimizing capacitive and Faradaic reactions through optimized waveform shaping, real-time impedance monitoring, and bi-directional safety checks remains critical for reliable long-term use.

2.2.2. Guidelines for Selecting the Electrode Pitch in Microelectrode Arrays

In microelectrode arrays designed for cortical electrophysiology, the spacing between microscale electrodes is a critical determinant of spatial resolution and signal independence. In other words, for μECoG systems, the electrode pitch directly influences the ability to discriminatively map cortical activity while minimizing interelectrode interference. Wang et al. reported that μECoG arrays configured with varying electrode diameters (200–500 μm) and pitches (500 μm–1 mm) exhibited improved discrimination of stimulus-evoked potential distributions at smaller pitches, although reduced spacing also increased the trend of interelectrode crosstalk [70]. In a more clinically oriented demonstration, Hermiz et al. showed that sub-millimeter μECoG grids with a 400 μm-pitch achieved substantially higher signal fidelity and more precise micro-domain mapping, compared to conventional ECoG arrays with 2–4 mm spacing [71]. Although the available dataset remains limited, the collective findings to date clearly indicate that electrode pitch is a central design parameter that governs the trade-off between spatial resolution and signal independence. Notably, many μECoG arrays intended for animal studies employ pitches within the 500–700 μm range [69,72,73]. This design choice is not merely a constraint imposed by fabrication limitations, such as microfabrication or thin-film deposition methods, but has emerged from empirical optimization based on the physiological characteristics of cortical signals. In μECoG recordings, the measured LFP reflects the aggregate behavior of spatially adjacent neuronal populations. Prior studies have reported the spatial correlation length of cortical LFP signals to lie within ~200–500 μm [74,75,76]. Therefore, selecting electrode spacing within this range enables each electrode to sample partially distinct neural populations while still capturing the continuous spatial distribution of the LFP without aliasing. As such, array designs incorporating electrode pitches of 200–700 μm represent a practical balance point that aligns the physiological propagation characteristics of cortical potentials with the engineering requirements for spatial resolution, signal independence, and device scalability.

2.2.3. Design Considerations for the Number of Channels in Single μECoG Electrode Arrays

The number of channels (i.e., microelectrode arrays) in a μECoG system defines the spatial coverage and the spatial resolution achievable during simultaneous cortical signal acquisition. Arrays with a greater number of channels, when matched with an optimized electrode pitch, capture broader cortical areas with finer spatial granularity. Conversely, even arrays with fewer channels may achieve superior local resolution if electrode density remains high. Thus, the number of channels must be selected in concert with targeted neuroscientific applications. However, increasing channel numbers generally introduces engineering challenges in circuit integration, power consumption, data bandwidth, and thermal management. As the number of electrodes grows, signal-routing complexity increases, raising the risk of noise coupling, parasitic interactions, and thermal artifacts. In densely packed flat-type μECoG arrays, capacitive coupling between adjacent electrodes becomes a pronounced issue, and prior analyses indicate that when electrode impedance exceeds ~1.7 MΩ, crosstalk can surpass 1% [77]. A commonly used strategy to mitigate these limitations involves implementing active electrode architectures, wherein analog front-end (AFE) amplifiers are placed proximal to each electrode site. This configuration electrically isolates the electrode impedance from the interconnect network, significantly reducing crosstalk [78,79]. Nevertheless, AFE-based designs introduce additional constraints, including localized heat generation (<1 °C) and limited available substrate area for integrating the amplifiers, pointing to the need for continued development of low-power, thermally stable AFE technologies [80,81]. To address these issues, time-division multiplexing (TDM) readout architectures have gained significant interest [19,34,82]. TDM enables multiple electrodes to share a single AFE block through sequential time-slot allocation, reducing the required on-chip footprint and thermal load without compromising sampling rate. This approach may provide a scalable pathway to multi-channel μECoG systems while maintaining power efficiency. In summary, expanding the channel count in μECoG arrays is not simply a matter of increasing the number of measurement points on the cortical surface. Rather, it requires a holistic systems-engineering approach that jointly optimizes electrode pitch, impedance, interconnect architecture, and circuit integration. Future directions in μECoG technology will be centered on achieving arrays with hundreds to thousands of channels, supported by ultra-low-power, low-noise, highly integrated readout circuits capable of sustaining long-term, high-fidelity neural monitoring.

While channel-count optimization introduces unique system-level constraints, these considerations must ultimately be integrated with the broader structural parameters discussed in Section 2.2. Overall, the structural design rules discussed in this section provide a foundation for optimizing device performance. Establishing clear correlations between structural parameters, such as electrode size, pitch, and channel count, and the resulting functional behavior is essential. For example, reducing electrode size enables high-density recording yet inevitably increases interfacial impedance, whereas decreasing electrode pitch improves spatial resolution but complicates routing and wiring density. Therefore, the structural design of μECoG electrodes must be determined in direct alignment with the intended application. For instance, resolving microscale motor intentions in BCI systems requires ultrahigh-density arrays, whereas mapping seizure propagation in epilepsy prioritizes broader spatial coverage. Collectively, these principles offer a strategic framework for tailoring μECoG architectures to diverse clinical demands.

2.3. Development of Neuroengineering Technologies for μECoG Signal Utilization

2.3.1. Data-Driven System Design for BCI/BMI Applications

The μECoG devices described in this review, incorporating microscale neural electrodes, offer substantially higher temporal and spatial resolution than conventional, larger-footprint ECoG systems. Their fine spatial granularity approaches the physical scale of neuronal organization at the cortical surface, enabling more precise decoding of neural activity, but minimizing tissue disruption due to their minimally invasive placement. As a newly developed approach, μECoG platforms have become increasingly attractive for advanced brain–computer interface (BCI) and brain–machine interface (BMI) applications [83,84,85,86,87,88]. Recent high-resolution decoding studies, exemplified in Figure 4a, demonstrate a growing interest in leveraging μECoG signals to reconstruct cognitive intent, including motor behavior and speech at unprecedented fidelity [83]. For instance, in motor BCI research, Branco et al. implanted high-density μECoG arrays over the human primary somatosensory cortex (S1) and successfully decoded distinct cortical activation patterns corresponding to finger movements and hand gestures [84]. Hochberg et al. further extended these concepts by integrating μECoG signals with intracortical activity in a tetraplegic patient, enabling real-time robotic arm control driven solely by motor cortical activity [85]. Further developments include the study by Williams et al., which used high-density μECoG arrays to distinguish between voluntary motor intention and resting states, demonstrating the feasibility of asynchronous BCI control [86]. In primate models, Rouse et al. chronically implanted μECoG grids over the sensorimotor cortex and achieved computer cursor control using gamma-band (75–105 Hz) activity [87,88]. Similarly, Zhou et al. developed a chronic μECoG interface capable of stable long-term motor decoding, enabling patients undergoing awake neurosurgery to control computer-based tasks such as “Ping-Pong” and “Snake” in real time [83].

Beyond motor control, μECoG has also demonstrated strong potential for speech and communication-related BCIs [91,92,93,94]. Kellis et al. recorded μECoG signals from language-associated cortical regions, including Wernicke’s area and orofacial motor cortex, and successfully classified cortical activation patterns, corresponding to spoken words and phonemes [91]. Kellis et al. used LFP-derived μECoG signal features to reconstruct acoustic speech output and categorize spoken words, confirming that μECoG contains sufficient temporal/spatial richness for speech decoding [92]. Building on this, Duraivel et al. utilized high-density μECoG to decode phoneme-level neural signatures with high accuracy, enabling reliable decoding of words and sentences [93]. A highly progressed demonstration by Willett et al. showed that μECoG signals recorded during imagined handwriting could be interpreted at the character level, enabling real-time “brain-to-text” communication [94]. This approach highlights the potential for μECoG-based BCI technology to restore communication capabilities in individuals with severe paralysis, using minimally invasive neural interfaces. Furthermore, these studies show that μECoG-based BCIs are poised to advance toward next-generation interfaces by integrating high-resolution neural sensing with AI-driven decoding algorithms [95,96]. Such systems promise real-time transformation of neural activity into motor commands, speech, or symbolic outputs, bringing μECoG-based BCI or BMI applications closer to clinical translation.

2.3.2. Development of μECoG Signal Acquisition for Clinical Applications

Historically, ECoG-based signal analysis has been used primarily for surgical management of epilepsy, particularly for identifying the seizure onset zone by detecting interictal spikes or locating the earliest ictal activity (Figure 4b) [67,97,98,99]. With the advent of μECoG technology, clinical electrophysiology has expanded beyond seizure localization, enabling more refined characterization of cortical function with significantly enhanced resolution. As described earlier, advanced μECoG electrode arrays possess hundreds-of-micrometers-scale spacing, exhibit high mechanical flexibility, and maintain strong biocompatibility, collectively offering superior spatial/temporal resolution, compared to traditional bulky ECoG electrodes. Owing to these properties, μECoG enables not only the identification of seizure foci but also fine-grained mapping of seizure propagation pathways and delineation of pathological/normal cortical boundaries [60,100,101]. A pioneering clinical example was demonstrated by Ganji et al.; they developed PEDOT:PSS-based μECoG electrodes for intraoperative cortical monitoring during tumor resection and epilepsy surgery. With over 10-fold lower impedance than conventional clinical electrodes, the μECoG device allowed high-resolution mapping of high-gamma activity (~400 μm spatial precision), enabling improved identification of functional borders surrounding motor and language regions while maximizing lesion resection [60]. More recent developments extend μECoG’s role from passive monitoring to closed-loop neuromodulation and therapeutic intervention. Lim et al. introduced a flexible, multilayer graphene/Au/graphene hybrid μECoG system for freely moving rodents and demonstrated seizure detection and real-time electrical stimulation to suppress ictal events (Figure 4c) [89]. The combination of low impedance, high SNR, and mechanical robustness enabled precise detection of seizure activity and effective suppression of aberrant behavior, using high-frequency stimulation, delivered through the identical sensing electrode arrays. Efforts to integrate drug delivery with μECoG have also gained traction. Minev et al. developed an elastic silicon-based “e-dura” platform that combined electrical stimulation with simultaneous drug infusion along the spinal surface, promoting functional recovery [102]. Similarly, Sung et al. created a wireless, flexible drug-delivery microdevice (f-DDM) capable of releasing antiepileptic compounds to suppress seizures [103]. Proctor et al. further advanced this concept by integrating a microfluidic ion pump (μFIP) with μECoG electrodes, providing simultaneous neural recording and targeted ionic drug delivery for real-time seizure detection and suppression (Figure 4d) [90]. Collectively, these emerging technologies demonstrate that μECoG-based neural interfaces are evolving from diagnostic tools into multifunctional therapeutic systems capable of real-time sensing, stimulation, and pharmacological intervention. As data-processing techniques and intelligent neural decoding frameworks continue to advance, μECoG is expected to play a pivotal role in precision neuromodulation and rehabilitation technologies for a broad spectrum of neurological disorders, including epilepsy, movement disorders, speech impairments, and neurodegenerative diseases.

2.4. Integration of Electrical Stimulation and Signal Acquisition in μECoG Interfaces

2.4.1. Bidirectional μECoG System Technologies

Conventional ECoG systems have historically been designed as unidirectional platforms, focusing solely on recording neural activity, which is useful for monitoring cortical signals. However, these systems often have limited capability for real-time correction of pathological activity or for closed-loop neuromodulation. For next-generation BCI systems, however, purely read-out-based architectures are insufficient, as they restrict fine control of movement and prevent natural, bidirectional interaction between the user and external devices. To address these limitations, μECoG-based bidirectional systems have been developed that sense neural activity and deliver electrical stimulation through the microelectrode arrays. Such architectures can detect abnormal patterns in real time, deliver corrective stimulation, and even return artificial sensory feedback directly to the cortex [31,41,89,104]. For example, as reported previously, Lim et al. demonstrated a graphene-based μECoG array in freely moving animal models that detected seizure activity and delivered high-frequency stimulation via the same electrodes, successfully suppressing convulsive events in real time [89]. These advances transform traditional passive monitoring systems into active neuromodulation platforms that enable causal manipulation of neural circuits. This bidirectional paradigm implies a transition from simple electrophysiological read-out to neural prosthetic systems capable of restoring function, modulating pathological network dynamics, and implementing closed-loop BCIs. As such, it expands a concept to a broad spectrum of neuroengineering applications in which information is continuously sensed, interpreted, and fed back to the brain in real time.

2.4.2. Electrode Design for Cortical Stimulation Using μECoG Systems

For μECoG electrodes to reliably perform stimulation, they must support sufficiently high charge injection capacity (CIC) while maintaining long-term electrochemical and biological stability. Cortical stimulation electrodes modulate neural activity by injecting current into the surface layers of the cortex, and the electrochemical reactions occurring at the electrode/electrolyte interface largely determine both stimulation performance and safety. Instabilities or irreversible reactions at this interface may lead to electrode corrosion, tissue damage, increased impedance, and deterioration of stimulation efficacy [105]. Because μECoG arrays typically employ microscale electrodes with constraint areas on the order of tens to hundreds of micrometers, their per-electrode CIC is inherently limited. While noble metals like platinum (Pt) and iridium oxide (IrO_x) offer excellent electrochemical stability and chemical inertness suitable for chronic interfaces, their charge injection limits restrict the miniaturization required for high-selectivity stimulation. Consequently, recent works have focused on material and surface engineering to enhance charge storage capacity (CSC) and CIC. Ganji et al. employed PEDOT:PSS, a conducting polymer with high CSC, as an electrode material and demonstrated substantially higher charge injection capabilities compared to conventional clinical electrodes made of platinum or platinum–iridium (30–100 μC cm²) [106]. PEDOT:PSS/Au electrodes exhibited ~9.5-fold higher CIC than bare Au electrodes of identical geometry, while PEDOT:PSS/Pt electrodes achieved roughly 3.2-fold higher CIC than bare Pt. In addition, complementary strategies aim to increase the effective electroactive surface area. Boehler et al. introduced nanostructured platinum (NanoPt) coatings on Pt electrodes via electrochemical reduction, forming well-defined nanoflake architectures that dramatically increased surface area [107]. This approach enhanced CSC by ~28-fold and CIC by more than an order of magnitude relative to planar Pt, illustrating the strong link between nanoscale morphology and macroscopic stimulation performance.

Stable μECoG stimulation further requires precise control of electrochemical reactions and thermal effects. During current injections, potential excursions at the electrode surface drive oxidation/reduction reactions. When these reactions become irreversible due to excessive charge density or operation outside the water window (e.g., −0.6 to +0.8 V vs. reference), undesirable phenomena such as electrode degradation, pH shifts, metal ion release, and gas evolution may occur, ultimately damaging electrode arrays and tissue [108,109,110,111]. To avoid these risks, stimulation potentials must be constrained within the safe electrochemical window, and biphasic current pulses are employed to ensure near-zero net DC offset. In these waveforms, the second phase retrieves charge delivered in the first, limiting net charge accumulation and helping to maintain electrochemical equilibrium. Even so, long-term or high-frequency stimulation can lead to cumulative electrochemical and structural changes [112,113]. Several studies have reported that repetitive stimulation promotes the generation of reactive oxygen species (ROS) and local pH fluctuations, contributing to reduced conductivity and surface oxidation [114,115,116]. To counteract such degradation, several groups have proposed introducing chemical stability interlayers or ion-limiting polymer coatings [115,117,118]. For instance, applying a PEDOT protective layer onto Pt electrodes has been shown to preserve CIC while reducing Pt corrosion and cell damage [118]. These multilayered architectures help buffer potential excursions and maintain stable charge transfer within the water window.

Thermal effects constitute another key safety consideration. Joule heating during current injection can induce local temperature elevations, and sustained temperatures above ~42 °C are associated with protein denaturation and cellular damage. This risk is exacerbated in high-frequency or long-duration stimulation protocols [119,120]. Ebrahimibasabi et al. reported that repeated stimulation can induce thermal-mechanical expansion, accelerating delamination at the electrode/insulator interface [121]. To circumvent such effects, stimulation parameters (i.e., current amplitude, pulse width, frequency) should be constrained within empirically established safety ranges such as those derived from Shannon’s criterion [122], and electrode-substrate stacks should be engineered to minimize mismatch in the coefficient of thermal expansion (CTE). Combining advanced nanoscale electrode structuring with porous or 3D architecture allows the same total charge to be distributed over a larger effective area, thereby minimizing local heating and improving electrochemical stability.

2.4.3. Artifact Suppression During Stimulation in Bidirectional μECoG Systems

Bidirectional μECoG systems, which record neural signals and deliver electrical stimulation through the same electrode array, introduce substantial technical complexity. A principal challenge is the presence of stimulation artifacts; large, non-physiological signals generated by the stimulation pulses themselves. These artifacts can exceed neural signals by orders of magnitude, leading to amplifier saturation, obscuring evoked neural responses, and complicating real-time feedback control. Stimulation artifacts arise from multiple mechanisms, including electric field spread, current diffusion through tissue, capacitive coupling between stimulation and recording paths, and nonlinear electrochemical charging and discharging at the electrode/electrolyte interface [123,124,125]. Current spread can induce large transient potentials in neighboring recording electrodes, and capacitive coupling between closely spaced conductors becomes particularly problematic in high-density arrays. Electrochemical capacitive behavior at the interface also contributes to extended decay tails, prolonging artifact duration. These artifacts exhibit characteristic temporal, spectral, and spatial signatures. Temporally, they are tightly time-locked to the stimulus pulses and often display stereotyped waveforms. In the frequency domain, they show strong energy concentration at the stimulation frequency and its harmonics. Spatially, they often form dipolar or radially symmetric potential patterns centered on the stimulation site, contrasting with the more localized and heterogeneous patterns of genuine cortical activity [126,127]. To address these issues, both hardware- and software-based artifact suppression strategies have been proposed [128,129]. Hardware-level approaches aim to prevent artifact propagation before it reaches the recording front-end. Bipolar or dipolar cancelation schemes introduce auxiliary stimulation electrodes carrying opposite-polarity currents to partially cancel the field at recording sites. Pu et al. demonstrated that setting the auxiliary stimulation amplitude to approximately 10% of the primary stimulation current prevented amplifier saturation and achieved an average reduction of ~22.9 dB in artifact magnitude [128]. Reference-tuned push-pull stimulation (RTPPS), employing a three-electrode configuration, injects complementary currents to cancel common-mode potentials at the recording location; Liu et al. reported that such schemes reduced artifact amplitudes to ~10–20% of neural spike amplitudes in vivo [129]. Electrode and circuit-level design also provide important means of a hardware-based approach. Capacitive-coupled electrodes and active microchannel interfaces have been proposed to reduce electrical coupling between stimulation and recording pathways [130,131]. Separating stimulation and recording routes at the circuit level and momentarily disconnecting the input stage during stimulation can offer additional protection, lowering the risk of front-end saturation and improving overall signal quality. On the signal-processing side, classical approaches include temporal blanking, discarding data during stimulation, and filtering strategies that attenuate artifact-dominated frequency bands [132,133]. More sophisticated methods such as null projection and pre-whitening filters selectively suppress artifact-dominated components while preserving neural activity, achieving attenuation on the order of 20–25 dB and expanding the artifact-free frequency range [128,134,135,136]. Adaptive filtering techniques, often driven by a reference of the stimulation pulse waveform, can predict and subtract artifacts in real time; least-mean-square (LMS)-based algorithms have demonstrated up to ~33 dB of artifact attenuation in bandwidths around 10 kHz, making them suitable for closed-loop applications [137,138].

Recent developments have also introduced machine learning-based techniques to further enhance artifact removal under nonstationary conditions [16,93,137,139,140]. Sayed et al. utilized the output bitstream of ΔΣ modulators to continuously learn and compensate stimulation-induced artifacts using a power-efficient stochastic signal processor, achieving ~33 dB peak-to-peak artifact suppression over a 10 kHz bandwidth [137]. In parallel, optimization of electrode materials and multilayer structures reduced artifact generation at the source; for example, Lee et al. showed that incorporating multilayer electrode architectures with high-quality dielectric interlayers weakens electrode-electrolyte coupling and limits current spread, thereby improving the fidelity of recorded signals during stimulation [141]. As noted above, while a distinct trade-off remains between the robust saturation protection of hardware designs and the implementation flexibility of software algorithms, artifact suppression in bidirectional μECoG systems has collectively evolved from simple post-processing into a core enabling technology for real-time, closed-loop neuromodulation. Future work is expected to integrate low-power mixed-signal circuitry, advanced electrode engineering, and AI-based signal restoration algorithms to achieve robust, fully bidirectional μECoG interfaces that support stable, high-fidelity neural recording even in the presence of ongoing stimulation.

2.5. Hardware for μECoG Neural Signal Acquisition: Commercial Systems

To faithfully record the low-amplitude potentials generated by μECoG electrodes, data acquisition hardware must provide high input impedance, low noise, and scalable multichannel capability. In current research practice, several commercial electrophysiology platforms are widely adopted for μECoG studies, including the RHD2000 series (Intan Technologies, Los Angeles, CA, USA), the Cerebus Neural Processing System (Blackrock Neurotech, Salt Lake City, UT, USA), the OmniPlex Neural Data Acquisition System (Plexon, Dallas, TX, USA), and the RZ2 platform (Tucker-Davis Technologies, Alachua, FL, USA) [55,59,72,99]. The RHD2000 series by Intan Technologies is a family of low-power, highly integrated microchips designed for multichannel bioelectrical signal acquisition, frequently used in headstage designs for neural recordings in small animal models such as mice and rats. Each chip integrates low-noise amplifier arrays, programmable bandwidth filters, a multiplexed 16-bit analog-to-digital converter (ADC), and on-chip electrode impedance measurement circuitry. The product line includes RHD2216 (16 channels), RHD2132 (32 channels), and RHD2164 (64 channels). Each channel exhibits an input-referred noise floor of approximately 2.4 μV_rms and a high input impedance on the order of ~1.3 GΩ at 10 Hz, enabling accurate recording even from relatively high-impedance microelectrodes. The ADC can operate at sampling rates up to 1.05 MS s⁻¹, allowing, for example, 32 amplifier channels to be simultaneously sampled at 30 kS s⁻¹. The lower and upper cutoff frequencies of the amplifiers are programmable over the range of 0.1–500 Hz and 100 Hz–20 kHz, respectively, accommodating LFP and spike-band recordings. Data transmission is handled via a serial peripheral interface (SPI) bus using low-voltage differential signaling (LVDS), which maintains signal integrity and low noise over extended cable lengths [142]. Owing to its modularity, compact form factor, and low power consumption, the RHD2000 series is widely used as a core building block in custom μECoG acquisition systems [58].

The Cerebus Neural Processing System from Blackrock Neurotech is a representative high-end commercial platform for large-scale neurophysiology. It supports real-time acquisition and processing of both action potentials (i.e., spikes) and field potentials, synchronized with behavioral or experimental events. The system provides real-time processing for up to 128 channels, and multiple systems can be synchronized to scale the channel count to several hundred. Signals are digitized with 16-bit resolution at sampling rates up to 30 kHz. The system features an exceptionally high input impedance of ~1 TΩ, which allows stable amplification of microvolt-level signals, across a wide range of electrode impedances. The input-referred noise remains below 3.0 μV_rms over the full bandwidth, and common-mode rejection ratios (CMRR) exceeding 100 dB at 50/60 Hz ensure robust performance in typical laboratory environments [143]. These characteristics make the Cerebus system particularly well suited for high-precision μECoG experiments in non-human primates, where high channel count and signal fidelity are critical [144].

The OmniPlex Neural Data Acquisition System from Plexon is another high-performance platform designed for flexible scaling of channel count and integration with behavioral paradigms. It supports configurations from 16 up to 256 neural channels, with all channels sampled at up to 40 kHz and 16-bit resolution. In addition, OmniPlex system provides up to 32 auxiliary analog input channels for non-neural signals (e.g., behavioral, sensor, or stimulation-related signals), facilitating multimodal experiment designs. The front-end amplifiers maintain a CMRR of over 100 dB in the 0–60 Hz range, and the filtering architecture enables effective separation of spike and field-potential bands from broadband neural signals. Advanced digital referencing schemes, including Common Average Referencing (CAR) and Common Median Referencing (CMR), are supported to minimize shared noise across channels. A key feature of the OmniPlex system is its online spike-sorting capability, which provides reliable real-time classification in various feature spaces (e.g., time–voltage waveforms, template matching, peak–valley metrics). PlexControl software (version 1.23.1; Plexon, Dallas, TX, USA) tightly integrates acquisition, visualization, and closed-loop experimental control, and the system interfaces with MATLAB and C⁺⁺ software development kit (SDKs) (MATLAB SDK: online v1.4.1, offline v1.2.0; C⁺⁺ SDK: online v2.4, offline v1.0.0) for online and offline analysis [145]. These capabilities make the OmniPlex system particularly attractive for μECoG experiments that require precise synchronization of neural activity with behavioral data and real-time event-driven paradigms.

The RZ2 BioAmp Processor from Tucker-Davis Technologies is a DSP-based platform optimized for multi-channel-count neurophysiology and real-time signal processing. The RZ2 employs a multiprocessor architecture consisting of 2, 4, or 8 DSP cards, enabling parallel processing of large data streams and high-speed memory access. Four dedicated data buses minimize bottlenecks in data flow, and when combined with Z-series amplifiers such as the PZ5, the system can acquire up to 256 channels at 25 kHz, or 128 channels at 50 kHz, simultaneously. The onboard I/O configuration includes 8-channel 16-bit PCM D/A outputs (for stimulus generation and experimental control), 8-channel 16-bit PCM A/D inputs (for external analog signals), and 24-bit digital I/O ports. The Synapse software (version 102; Tucker-Davis Technologies, Alachua, FL, USA) environment allows intuitive graphical configuration of acquisition and processing pipelines, automatically distributing computational tasks across DSP cards to support efficient real-time operation. The RZ2 can generate electrical stimulation signals directly via its onboard D/A outputs and can be integrated with dedicated stimulation modules such as the IZ2 Stimulator to provide precise, current-controlled stimulation. This combination enables simultaneous real-time neural recording and electrical stimulation within a single unified hardware framework, making the RZ2 particularly suitable for closed-loop μECoG neuromodulation experiments [146].

2.6. Transitioning Flexible Neural Interfaces Toward Spinal Cord Electrode Systems

Building upon the flexible material-based neural interface technologies introduced in earlier sections, such as representative μECoG electrode fabrication and associated data-acquisition system, there is growing interest in adapting similar classes of microscale flexible electrodes for spinal cord neural interfacing (Figure 5a). Spinal interfaces are gaining great attention since they can overcome the intrinsic limitations of traditional, unidirectional brain interfaces by enabling bidirectional control, direct modulation of damaged pathways, and circuit-level rerouting. The spinal cord contains intrinsic locomotor circuits, such as central pattern generators (CPGs), that can be re-engaged through appropriately tuned electrical stimulation, promoting partial recovery of coordinated gait even after injury. Consequently, spinal interfaces have emerged as a promising extension of brain-interface technologies, offering new opportunities in motor restoration, autonomic modulation, and chronic pain management domains historically difficult to access within neuroengineering.

Unlike cortical interfaces, spinal electrodes can readily be designed for a distinct anatomical and mechanical environment, necessitating tailored microelectrode layouts. The spinal cord resides within the vertebral canal, tightly confined by bony structures, which severely limits available space for electrode placement. Thus, epidurally placed spinal interfaces require extremely thin, narrow, and conformable electrodes; in mouse models, the epidural space above the dorsal columns is typically on the order of ~120 μm [147]. Targeting small anatomical structures, such as the dorsal roots or dorsal root ganglia located within narrow intervertebral foramina, further necessitates miniaturized, highly flexible designs. Moreover, the spinal cord undergoes continuous micro-movements with respiration, cardiac pulsation, and postural adjustments, and can translate axially during trunk or neck motion. This mobility introduces relative sliding, shear, and friction between the electrode and spinal tissue, increasing the risk of mechanical irritation or damage. Recent analyses of spinal mechanobiology emphasize that the electrode/tissue interface is governed by gross anatomical constraints and microscale dynamic interactions that unfold during physiological loading [102,147]. The spinal cord experiences cyclic shear, compression, and transverse strain arising from respiration, vascular pulsatility, and posture-dependent deformation; these strains typically fall within the 1–5% range but can locally exceed 10% near tethering points or during rapid trunk movement [148,149,150]. Such deformation creates spatially heterogeneous stress fields across the dura/cerebrospinal fluid/cord interface, increasing the possibility of relative sliding and frictional shear at the surface of implanted electrodes. Over time, these repetitive micromechanical events may induce low-grade inflammation, glial activation, or microvascular irritation, particularly when the mechanical modulus or bending stiffness of the implant is poorly matched to that of spinal tissue [151,152]. Thus, the mechanobiological environment of the spinal cord imposes unique demands that extend beyond simple flexibility, requiring ultralow bending stiffness, reduced frictional coefficients, optimized strain-transfer profiles, and designs that minimize stress concentrations at implanted electrode edges or anchor points. Accordingly, the design of spinal interface electrodes must explicitly account for the combined constraints of restricted epidural space and persistent, physiologically driven micromotion, ensuring mechanical conformity and functional stability under dynamic loading conditions.

Figure 5. Representative flexible and implantable spinal neuro-interface technologies enabling high-fidelity recording, neuromodulation, and closed-loop motor restoration. (a) Schematic comparison of anatomical and mechanical characteristics of spinal cord environments that constrain electrode design, including dorsal positioning, curvature, and dynamic micromotion. (b) Soft e-dura implant featuring stretchable microcracked Au interconnects and Pt-PDMS composite electrodes, designed to match spinal tissue mechanics and support chronic neuromodulation during locomotor behavior. Reproduced with permission from [153]. Copyright Wiley–VCH, 2020. (c) Bifurcated PI-based subdural bioelectronic implant with Pt microelectrodes engineered to offload pressure from the central spinal vein, enabling stable epidural or subdural spinal cord recording across thoracic and lumbar segments. Reproduced with permission from [154]. Copyright Wiley–VCH, 2022. (d) Flexible circumferential bioelectronic array (i360) fabricated on an ultrathin Parylene-C substrate, incorporating Ti/Au electrodes coated with PEDOT:PSS to improve charge transfer and conformal wrapping around the spinal cord. Reproduced with permission from [28]. Copyright American Association for the Advancement of Science, 2024. (e) Schematic of the electrochemical neuroprosthesis and hindlimb endpoint trajectories in rodent models of severe spinal cord injury, showing that increasing the frequency of epidural electrical stimulation (20–80 Hz) during treadmill locomotion produces graded increases in step height. Reproduced with permission from [155]. Copyright American Association for the Advancement of Science, 2014. (f) Conceptual and technological design of a proportional brain–spine interface in rats with severe spinal cord injury, in which multiunit activity recorded from a 32-channel array in the leg area of motor cortex is converted into cumulative firing to regulate the timing and intensity of epidural stimulation over lumbar segments that recruit flexor-related proprioceptive circuits during gravity-assisted locomotion. Reproduced with permission from [156]. Copyright Springer Nature, 2018.

Figure 5. Representative flexible and implantable spinal neuro-interface technologies enabling high-fidelity recording, neuromodulation, and closed-loop motor restoration. (a) Schematic comparison of anatomical and mechanical characteristics of spinal cord environments that constrain electrode design, including dorsal positioning, curvature, and dynamic micromotion. (b) Soft e-dura implant featuring stretchable microcracked Au interconnects and Pt-PDMS composite electrodes, designed to match spinal tissue mechanics and support chronic neuromodulation during locomotor behavior. Reproduced with permission from [153]. Copyright Wiley–VCH, 2020. (c) Bifurcated PI-based subdural bioelectronic implant with Pt microelectrodes engineered to offload pressure from the central spinal vein, enabling stable epidural or subdural spinal cord recording across thoracic and lumbar segments. Reproduced with permission from [154]. Copyright Wiley–VCH, 2022. (d) Flexible circumferential bioelectronic array (i360) fabricated on an ultrathin Parylene-C substrate, incorporating Ti/Au electrodes coated with PEDOT:PSS to improve charge transfer and conformal wrapping around the spinal cord. Reproduced with permission from [28]. Copyright American Association for the Advancement of Science, 2024. (e) Schematic of the electrochemical neuroprosthesis and hindlimb endpoint trajectories in rodent models of severe spinal cord injury, showing that increasing the frequency of epidural electrical stimulation (20–80 Hz) during treadmill locomotion produces graded increases in step height. Reproduced with permission from [155]. Copyright American Association for the Advancement of Science, 2014. (f) Conceptual and technological design of a proportional brain–spine interface in rats with severe spinal cord injury, in which multiunit activity recorded from a 32-channel array in the leg area of motor cortex is converted into cumulative firing to regulate the timing and intensity of epidural stimulation over lumbar segments that recruit flexor-related proprioceptive circuits during gravity-assisted locomotion. Reproduced with permission from [156]. Copyright Springer Nature, 2018.

As with cortical interfaces, spinal electrodes require materials with mechanical compliance comparable to soft neural tissues. Thin polymer films (e.g., PI) and elastomers (e.g., PDMS) are frequently employed as substrate materials to ensure adequate conformability [153,154]. In addition, similar to μECoG technology, efforts to enhance electrochemical stability and stimulation efficiency include surface modification, nanostructuring, and the incorporation of high-performance electrode materials. For example, Schiavone et al. fabricated a flexible spinal electrode using PDMS, a material with mechanical properties resembling the dura mater, and incorporated a Pt-PDMS composite as the electrode coating (Figure 5b) [153]. This composite enhanced the mechanical compliance of silicone and the favorable electrochemical characteristics of platinum, achieving a high charge-injection capacity (CIC; 214 μC cm⁻²), stable impedance, and excellent mechanical durability. To further reinforce long-term stability under spinal micromotion, Au thin films were deposited on the PDMS substrate and encapsulation layer, allowing the electrodes to withstand one million mechanical cycles without loss of function.

Beyond materials optimization, device-level structural design is tailored for the spinal environment. Harland et al. introduced a bifurcated electrode layout in which the central region is split to reduce pressure on the dorsal median vein and improve biocompatibility (Figure 5c) [154]. This architecture reduced inflammatory responses relative to planar electrodes and enabled stable recordings over 12 weeks in awake, freely moving mice. Given the cylindrical geometry of the spinal cord, circumferential electrode designs have also been explored. One example incorporates a 360° wrap-around interface capable of simultaneously recording and stimulating the dorsal, lateral, and ventral epidural surfaces (Figure 5d) [28]. The linear offset arrangement of 32-electrodes maximized spatial coverage while minimizing cross-talk. This allowed comprehensive topographic representation of signal amplitudes and selective activation of specific hind-limb muscle groups by targeting ventral or lateral epidural hotspots. Such architecture facilitates high-dimensional mapping of spinal function and provides a platform extensible to closed-loop neuromodulation and electronic bypass systems.

Spinal-specific electrode systems are now being applied for neural recording with circuit-level modulation, functional restoration, and mechanistic investigation. Wenger et al. demonstrated in rodent models of severe spinal cord injury (SCI) that epidural electrical stimulation (EES) of lumbosacral spinal circuits can be tuned to provide graded, reproducible control of hindlimb step height during locomotion (Figure 5e) [155]. Their work revealed that this precise control emerges from frequency-dependent recruitment of proprioceptive afferents and the coordinated engagement of mono- and polysynaptic reflex circuits within spinal sensorimotor networks. They further established EES frequency and step height as a mechanistically coupled control pair, providing a concrete framework for closed-loop neuromodulation strategies to restore refined locomotion after SCI. Similarly, Bonizzato et al. developed a proportional brain–spine interface (BSI) in rats with severe spinal cord injury (SCI), in which brain-controlled neuromodulation linked motor cortex activity to epidural electrical stimulation of lumbar spinal circuits below the lesion (Figure 5f) [156]. By implanting a 32-channel microwire array in the leg area of the motor cortex, they decoded cortical ensemble activity associated with leg flexion and step height and used these signals in real time to configure spinal stimulation protocols. When the decoded cortical activity crossed predefined thresholds, the BSI delivered appropriately timed bursts of stimulation over L2 and, in proportional mode, continuously scaled stimulation amplitude, enabling task-specific modulation of ankle flexor activity and foot clearance, including during stair climbing. Applied throughout a 5-week rehabilitation program, this brain-controlled neuromodulation immediately restored overground locomotion in otherwise paralyzed rats and significantly accelerated and enhanced long-term gait recovery compared with continuous spinal stimulation alone.

3. Outlook: Future Challenges and Opportunities

In summary, this review has explored the foundational advancements in flexible μECoG interfaces, encompassing the full spectrum from material engineering to system-level integration. By synthesizing early works and current trends in soft substrates and nanostructured electrodes, we highlight how emerging strategies have resolved the mechanical-electrochemical trade-off, enabling high-fidelity recording. We further detailed structural design rules, such as electrode pitch and channel scaling, that optimize these arrays for specific clinical applications, including high-resolution BCI and closed-loop epilepsy management. Moreover, we emphasized the expansion of these technologies to the spinal cord, addressing the unique mechanobiological constraints for locomotor rehabilitation.

The future trajectory of μECoG research in neuroengineering is increasingly oriented toward developing fully integrated, miniaturized platforms that unify microscale electrode arrays, signal-conditioning circuits, wireless communication, and long-term power solutions within a single biocompatible module. Although commercial benchtop data-acquisition systems enabled high-precision electrophysiological studies, their reliance on wired connections and bulky instrumentation constrains freely behaving animal experiments, continuous monitoring, and practical BCI applications. These limitations catalyzed a shift toward tissue-conformal and implantable μECoG microsystems, in which electrodes and front-end electronics are co-designed at the materials, circuit, and packaging levels. A central requirement for this new generation of devices is the incorporation of wireless data transmission. Recent systems using Bluetooth, Wi-Fi, and RF telemetry have demonstrated multi-channel ECoG streaming with low latency and high fidelity, enabling remote monitoring and real-time data analytics. For example, low-power Bluetooth architectures have succeeded in transmitting 32-channel ECoG signals to mobile devices and cloud platforms for automated seizure detection and intervention [157]. More advanced demonstrations, such as the W-HERBS system by Matsushita et al., integrating a 128-channel μECoG array with Wi-Fi telemetry, illustrate the feasibility of real-time, fully implantable, human-compatible BCI platforms [158]. Similarly, Parylene-C-based μECoG arrays combined with 2.4 GHz RF headstages enabled untethered data transmission in freely moving non-human primates, supporting stable high-bandwidth ECoG acquisition in naturalistic behavioral environments. Although early wireless μECoG systems relied on lithium batteries with limited lifespan and safety concerns, recent batteryless platforms using inductive or RF energy harvesting now enable truly uninterrupted chronic operation [159]. Chang et al. implemented a 6.78 MHz inductive power transfer module to realize a batteryless μECoG device capable of continuously recording cortical activity in freely moving mice using a Parylene-C flexible grid integrated with a 3D silicon probe array [160]. Additional systems using 300 MHz inductive power transfer and CMOS integration further highlight the promise of battery-free architectures for uninterrupted chronic implantation [161].

Collectively, these technological developments indicate a clear evolution toward autonomous, wirelessly powered, and chronically stable μECoG interfaces. Looking forward, the integration of on-device processing with AI-driven neural decoding is expected to reshape next-generation BCI architectures. Lightweight neural-network models, encompassing adaptive state-space decoders, compact deep learning modules, and reinforcement-driven controllers, are now capable of operating directly on implantable or near-implantable hardware. When paired with low-latency wireless telemetry and continuous power delivery, these AI-enhanced systems will enable closed-loop BMIs capable of sensing neural states, interpreting high-bandwidth activity, predicting behavioral intent, and deploying neuromodulatory feedback with minimal human intervention. Such systems lay the foundation for autonomous cortical and spinal neuroprostheses that maintain robust performance during natural behavior and long-term use.

Despite the above advances, regulatory and translational challenges remain significant. Chronic implant safety must be verified through rigorous evaluation of biostability, electrode durability, foreign-body responses, and long-term electrochemical stability under dynamic in vivo micromotion. Wireless communication and power transfer modules meet stringent standards for electromagnetic safety, cybersecurity, and fault tolerance. Demonstrating consistent performance across heterogeneous patient populations will require large-scale clinical trials that validate neural decoding accuracy, neuromodulatory efficacy, and critical reliability. These hurdles represent essential steps toward delivering intelligent μECoG systems suitable for medical-grade neuromodulation, rehabilitation, and neuroprosthetic control. Next-generation μECoG platforms will therefore emerge from deep integration across materials science, microfabrication, embedded electronics, wireless engineering, and machine intelligence. Even within this trend, structural and material design with practical fabrication processing technologies of the electrodes remain critical factors that must be prioritized. This is because these elements not only have a significant impact on the fundamental functions of the electrodes but also serve as the technological basis for key next-generation μECoG platform features such as the outward transmission of intracranial data, biocompatibility and durability, microstructured electrode materials, and energy supply. As this convergence accelerates, μECoG systems are poised to evolve from research tools into clinically deployable, adaptive neural interfaces that support long-term monitoring, personalized neuromodulation, and seamless brain-machine interaction.