1. Introduction
Neural interfaces establish a bidirectional communication link between the nervous system and external devices to capture, convert, and transmit neural signals, as well as provide feedback to the nervous system. Non-invasive neural interfaces, such as electroencephalography (EEG), provide a convenient, low-risk method of neural signal acquisition, but at the cost of frequency richness and spatial resolution
[1]. Intracortical interfaces make direct contact with neurons to capture high-frequency activities such as action potentials, which are filtered out by the dura mater, skull, and scalp
[2]. Furthermore, direct interfaces offer the potential for multimodal neural information detection, which is the ability to record both electrical and chemical signals
[3]. In addition, invasive interfaces enable precise spatial localization of neuronal activity and targeted neural modulation
[4]. These advantages provide new impetus for the ongoing innovation and application of invasive neural interface technologies.
Intracortical neural interfaces in freely moving animals have increased technical requirements, as shown in
Fig. 1. First, the development of high spatial density integration, which involves placing more electrodes within a limited area, aims to achieve a higher signal throughput without additional burden to the animal’s head. Second, stable long-term detection requires a reduction in the immune response of the brain tissue and relative displacement of the electrodes to minimize data variability during subject movement. Third, the multimodal recording function of neural electrical and chemical activities necessitates further integration of microfabrication and sensing technologies to construct interface-sensitive layers for different modal information while avoiding crosstalk between different signal pathways. Finally, bidirectional integration involves the fusion of advanced neural detection (spikes, local field potentials (LFPs), and neurotransmitter concentrations) with neural modulation (electrical, optical, and chemical methods), which requires integrated probe fabrication techniques for electrical stimulation (ES) electrodes, optical waveguides, or microfluidic channels.
The application of the aforementioned neural interface technology in behavioral studies allows researchers to directly observe animals in their natural states. In particular, high-density detection of multimodal neural information across neural circuits in multiple brain regions offers significant insight into the mechanism of how neuronal ensembles coordinate behavior. Compared with acute studies with head-fixed animals, long-term stable monitoring is crucial to elucidate the development, adaptation, and plasticity of animal behavior and brain function
[5]. This approach enables tracking of the evolution of cognitive functions over time, such as learning and memory, or clinical brain disorders. Furthermore, advancement in neurotransmitter detection technologies has enabled researchers to explore how chemical communication in the brain affects the physiologic functions, behaviors, and emotions of freely moving animals. Neural modulation can provide new strategies for restoring normal functions and treating neurological disorders.
In this article, we focus on the research directions in the development of intracortical neural interface technology, highlighting recent technological advancements in these areas and discussing their current and potential applications in brain cognition and diseases.
2. High spatial density microelectrode array (MEA)
Early single- and multi-channel-electrode structures typically consisted of a fine metal wire coated with an insulating material, leaving only the tip exposed for electrical contact with the tissue
[6]. These structures were mostly handmade, making large-scale production difficult. Advances in microfabrication have resulted in a significant increase in the spatial density and consistency of electrode arrays, leading to a preference for materials such as Si and its compounds as substrates and isolation layers. In 2011, Stevenson and Kording
[7] described Moore’s law of neural recording: The number of simultaneously recorded neurons doubles approximately every seven years since 1950. This trend is reinforced by the ongoing development of new high-density electrode architectures and microfabrication processes. Additionally, recent advancements in complementary metal–oxide–semiconductor (CMOS) technology have enabled a more compact integration of neural probes and multiplexing circuits, thereby providing unprecedented spatial resolution.
2.1. High-density structural design
Since the late 1960s, two classic MEA structures based on microelectromechanical systems (MEMS) technology have emerged, namely the Utah array
[8] and Michigan array
[9],
[10]. In the Utah array, the electrode sites are positioned at the tip of a shank, forming a sensing plane parallel to the brain surface, which enables large-area horizontal sampling. In the Michigan array, multiple detection sites are concentrated on a flat shank, allowing sampling perpendicular to the cortical surface, which enables the recording of neural activities from different layers of the brain. These two architectures continue to be reconfigured using new processes and materials to achieve a higher spatial density for neural electrodes.
The earliest silicon-based Utah array featured a channel count of 10 × 10 needle-shaped electrodes with a spacing of 400 μm, resulting in a density of 6.25 mm
−2. The high spatial density limitation of this architecture means that a more densely packed array of planar needles may result in greater stress damage to the brain tissue. Tilting the electrode needles allows contact with neurons in different layers, thereby overcoming the limitations of planar spacing. As shown in
Fig. 2(a), a Utah graded electrode array with a 200 μm spacing achieved a channel density of 25 mm
−2 [11]. Further development involves longitudinal multisite integration on a needle, where the sidewalls can be patterned using three-dimensional (3D) shadow mask techniques. In 2020, Shandhi and Negi
[12] designed a multi-point Utah array that deposited eight recording points on the sidewall of a needle, resulting in 900 active sites and a channel density of 56.25 mm
−2. In addition, reducing the electrode diameter can reduce tissue damage, leading to the expansion of higher-density electrodes
[13],
[14].
The Michigan array integrates multiple sites on one or more flat shanks and has the advantage of a smaller cross-sectional area than the Utah array. The limitation of the conventional Michigan array design is that the connecting leads for all the independent electrode sites occupy most of the available shank width. Increasing the number of sites would lead to an increase in the shank width. The use of conventional photolithography techniques often results in Michigan arrays with wire widths and spacings of ≥ 1 μm
[15]. Electron beam lithography can reduce these feature sizes to submicrometer dimensions
[16], enabling the creation of fine structures on the probe. When combined with classic MEMS techniques for manufacturing larger structures (e.g., non-fine wiring, shank shape, and lead pads), this approach can achieve local densely packed sites (2 × 100 sites within a 50 μm × 1100 μm area)
[17]. However, the total channel count remains limited, and the high cost of electron beam lithography makes mass production difficult. Moreover, dual-layer wiring processes can double the channel density, as demonstrated in a study using a parylene-based electrode array with two layers of patterning, releasing 512 recording sites on both the front and back of the array
[18]. External expansion of the planar array is another option. Rios et al.
[19] combined 16 64-channel Michigan probes to create a 4 × 4 electrode array (
Fig. 2(b)), expanding the sampling channels four-fold compared with those of conventional flat four-shank probes. In addition, an assembly method for a 3D unique Michigan array was developed using an anisotropic conductive film to increase the spatial density and achieve multiple brain regions recording
[20].
2.2. Compact fabrication process
Neural electrodes must be paired with a signal processing circuit. As the number of channels increases, the size of the backend circuit increases disproportionately. In animal behavioral studies, the weight of the on-head system (including the MEA, backend electronics, and supporting infrastructure) can significantly affect movement if it exceeds 10%–15% of the animal’s bodyweight, thereby hindering natural behavior and increasing the risk of tissue damage
[21],
[22]. To address this issue, miniaturized on-chip time-division multiplexing systems have become standard practice, as they reduce the overall size
[23]. Additionally, flexible connectors between on-chip circuits and probes are necessary to reduce the shear motion of the probes within the brain, further minimizing their impact on animal behavior.
Advances in CMOS technology have made it possible to achieve the integration of neural electrodes with an amplifier circuit in high-density packages
[24]. In 2020, Obaid et al.
[25] proposed a packaging strategy for a 3D microwire electrode array (
Fig. 2(c)) in which the microwires are vertically mechanically crimped to a commercial CMOS amplifier array, forming a structure similar to that of the Utah electrode array. This approach achieved over 90% conduction when the microwires were spaced 18 μm apart (linewidth: 15 μm). However, the system used in mice was limited to a 100 μm spacing to reduce tissue damage, capping the total channel count at 251 within a 3.5 mm diameter. For Michigan arrays, CMOS technology enables the embedding of amplifiers, time-division multiplexers, and addressing logic circuits into the probe shank.
In situ amplification of neural signals can reduce capacitive coupling of metal wires, further overcoming the limitations of wiring spacing. Both Neuropixels 1.0
[24] and 2.0
[26] were manufactured using 130 nm CMOS technology (
Fig. 2(d)). Neuropixels 2.0 features a four-shank structure, with each shank containing 1280 recording sites. It integrates sophisticated electronics for multiplexing, addressing, and local amplification into a silicon shank with dimensions of 70 μm, 20 μm, and 10 mm (width, thickness, and length, respectively). This allows the simultaneous recording of up to 768 channels. However, Neuropixels 1.0 and 2.0 employ 10-bit and 14-bit analog-to-digital converters (ADCs), respectively, which are inferior to the 16–24 bit quantization
[27],
[28] used in advanced systems, thereby limiting the dynamic range and affecting the signal quality.
The integration of probe electronics has been explored for decades, with early studies involving the placement of an ADC circuit at the probe base
[29],
[30]. De Dorigo et al.
[31] further enhanced this integration strategy by embedding the ADC directly into an implantable probe shank, thereby eliminating the need for extensive analog circuitry in the probe base and reducing the length of the external routing for sensitive neuronal signals, which helps minimize the overall footprint and mitigate signal interference. However, integrating additional circuits into implantable components introduces problems associated with increased heat dissipation. Compared with the base, the probe shank, which directly contacts the brain tissue, has stricter power consumption limits. Brain tissue can tolerate a temperature increase of ~2 °C over several hours without significant damage
[32]. Raducanu et al.
[33] conducted finite element simulations of the heat dissipation of a single-shank NeuroSeeker probe with 1356 integrated electrodes. They estimated that to keep the temperature rise below the 1 °C threshold, the power dissipation limit for the implanted shank is 4.5 mW, whereas the base can handle up to 45 mW. When designing high-density neural probes and circuits, it is crucial to assess the power consumption limit to prevent overheating.
2.3. Applications of high-density MEA in neural decoding
The development of high-throughput neural interface technology has significantly increased data acquisition efficiency. This technology reduces the number of animals required for MEA implantation experiments, thereby enhancing statistical efficiency, while simultaneously augmenting the likelihood of capturing target neural activities. By combining high-density electrophysiological responses with behavioral recordings, researchers can better understand the functional mapping of neuronal ensembles during tasks related to cognition, language, memory, and several more. This offers a new perspective on the relationship between the brain and behavior. For instance, in the spatial navigation behavior of animals, grid cells are typical spatially selective neurons that constitute approximately 18% of the cells in the medial entorhinal cortex
[34]. Neuropixels probes enable the simultaneous recording of a large number of grid cells across multiple modules, thereby elucidating the dynamic coordination mechanisms of spatial firing representations among grid cells in different modules
[35]. Additionally, in human language experiments, Neuropixels probes have been implanted in the superior temporal gyrus of awake patients to achieve large-scale single-neuron recordings, revealing the diverse neural population encoding of various speech features across different cortical layers
[36].
Tightly packed MEAs enable high-density neuronal recordings and precise anatomical mappings, providing high spatiotemporal resolution dynamics of brain activity across multiple regions. The potential applications include: ①High-density data collection from several brain regions within a neural circuit. By analyzing the correlated firing activities between neurons, researchers can gain insight into interregional communication
[37]. For instance, neurons that fire in synchrony often exhibit functional or structural connections, which could be as a result of shared upstream inputs, local circuit connectivity, columnar organization, or joint participation in similar sensory or cognitive tasks. Detecting paired neuronal activations with higher spatial resolution helps clarify the transmission models of neural signals within circuits. ②Assisting in stereotactic localization during electrode implantation. The insertion of electrodes can cause strain and displacement in brain tissue
[38], making brain atlas
[39] or magnetic resonance imaging (MRI)-based localization
[40] inaccurate. Continuous spatial recording of neural firing characteristics across different brain regions helps to precisely locate electrode positions in clinical applications, providing a more accurate alternative to stereotactic surgery and deep brain stimulation (DBS).
3. Long-term stable MEA
The implantation of MEA inevitably causes tissue damage and a progressive inflammatory response, leading to a decline in the recording signal quality. Physically, because of the mechanical mismatch between the tissue and implants, natural body motions such as respiration and vascular pulsatility can lead to relative shear movements
[41]. The shear forces inevitably increase as the animal moves freely, resulting in signal instability and tissue damage. MEA movements can exacerbate signal instability and increase shearing damage to tissues. Biologically, the implantation of MEA triggers immune responses involving microglial and astrocytic activation, leading to inflammation and glial scar formation
[42], which isolates the device from neurons. Various factors contribute to the progressive decline in the signal-to-noise ratio (SNR) until the neuronal signal eventually disappears. To address these challenges, improvement in flexible, biocompatible materials, optimized structural designs, and surface treatments have been explored to minimize shear forces and immune reactions, thereby enhancing long-term stability.
3.1. Flexible substrate
The mechanical mismatch between the tissue and implants is reflected in the Young’s modulus, which describes the ease of axial stretching and compression of a material within its linear elastic region. The Young’s modulus of semiconductor materials such as silicon is approximately six orders of magnitude higher than that of brain tissue
[2], whereas flexible materials have a much lower modulus, making them more comparable to brain tissues. Fluorescence imaging of microglia
[43] and astrocytes
[44] has shown that flexible materials elicit a significantly reduced immune response compared with silicon. MEAs based on flexible substrates, such as polyimide
[45],
[46], parylene
[47],
[48], and polydimethylsiloxane (PDMS)
[49],
[50],
[51], are increasingly emerging as viable options.
Flexible MEAs feature versatile geometries, such as planar slabs
[18], cylindrical wires
[52], helices
[53], and 3D meshes
[54]. Planar flexible MEAs are typically produced using silicon-based photolithographic processes. Flexible MEAs can be made thinner owing to the excellent flexibility of the substrates. However, because the CMOS processes for flexible materials are less mature than those of semiconductors, there are currently no intracortical flexible MEAs that match the spatial density of Neuropixels 2.0, as presented in
Table 1 [26],
[49],
[55],
[56],
[57],
[58]. In 2019, a 64-channel, 14 μm-thick, flexible, multi-shank MEA with a polyimide substrate was reported, as shown in
Fig. 3(a), which can be modularly expanded into a 1024-channel recording system
[58]. However, the relatively large size of the 1024-channel system limits its applicability to freely moving rodents. The study demonstrated successful long-term recordings in rats for up to five months with the 382-channel version. Wire structures fabricated through microfabrication and self-assembly exhibit minimal footprints to achieve a lower bending stiffness. For instance, Guan et al.
[52] developed self-assembled neural tassel electrodes (
Fig. 3(b)) with a cross-sectional footprint of 3 μm × 1.5 μm, which exhibited remarkable flexibility. However, highly flexible electrodes are difficult to implant into deep brain regions. Common strategies for flexible MEA implantation include rigid guides
[56] and reinforcement materials such as polyethylene glycol (PEG)
[59] and silk proteins
[60],
[61]. Guo et al.
[46] reported a silicon-based shuttle for implanting flexible adhesive electrodes into the brain tissue. In 2024, Liu et al.
[55] developed a method for rolling flexible electrodes onto linear tungsten wires for implantation and achieved
in vivo recordings for 105 weeks. However, these methods mostly increase the size of the implants during surgery, thereby exacerbating tissue damage. One approach to address this limitation is the use of shape memory or adaptive materials to enable controlled flexibility adjustments. In 2023, Yi et al.
[62] developed a water-responsive ultracontractive polymer neural electrode that remained rigid at room temperature in a dry environment, which is suitable for electronic manufacturing and implantation, but became soft in the moist environment of the brain, ensuring minimal tissue damage.
The 3D mesh electrodes provide an open structure that is optimal for neuronal interaction, thereby reducing probe drift and enabling long-term stability in neural recordings
[63],
[64], as shown in
Fig. 3(c). In 2019, Yang et al.
[65] proposed a biomimetic design for neural probes based on photolithographic patterns in which the electrode shape mimics the subcellular structural features of neurons. After injection into the brain, the electrodes diffuse to form a 3D structure, achieving interpenetration with neurons and improving glial cell isolation on the electrode surface. Mesh-structured electrodes have been reported to achieve stable neural recordings in behaving mice for over one year
[54]. Mesh electrodes can also be delivered to the brain via injection through the neck vessels
[66], enabling electrophysiological recordings of LFPs and single-unit spikes across the blood vessel wall. This technique eliminates damage to the brain tissue caused by surgical procedures and syringe insertion. However, it also limits the electrode density and capability for multimodal recording and stimulation.
3.2. Surface preparation
Surface preparation is used as a post-processing method and can be easily integrated into existing MEAs. This approach improves the physicochemical stability of the neural tissue interface and enhances the quality and longevity of the recorded signals. The modification layers can be categorized into electrode coatings that cover the entire probe and electroplated layers localized at the electrode sites. Electrode coatings utilize various stretchable, biocompatible materials, such as polymers, hydrogels
[67],
[68], and proteins
[69],
[70], which modify the biocompatibility of the electrode surface. Probes with more pliable surfaces exert less stress on brain tissue than non-compliant probes
[71],
[72]. Lee et al.
[73] developed a lubricated probe surface coating, which was demonstrated to significantly reduce insertion damage and decrease long-term immune responses, thereby improving the quality and longevity of neuronal signals.
Furthermore, coatings with anti-inflammatory agents or growth factors can reduce bioreactivity
[74] and tissue repair
[75]. As shown in
Fig. 3(d), calcium alginate/chitosan hydrogel loaded with dexamethasone sodium phosphate has been reported to reduce post-implantation tissue inflammation
[76]. However, a significant challenge in this strategy is the limited effective period, as biomolecules quickly diffuse away from the implant. An effective solution is to use responsive coatings that enable on-demand chronic drug release, for instance, through the application of cyclic voltammetry
[77]. Nonetheless, the benefits of reducing inflammation through coatings must be weighed against the increased acute injury caused by larger cross-sectional dimensions.
Electrode site surface deposition methods are effective in reducing impedance and thermal noise. During long-term recordings, consistent, high-quality acquisition of action potentials for single neurons requires small electrode sites to match the neurons. However, the performance of small sites deteriorates owing to increased impedance and thermal noise
[78]. Therefore, the impedance should be reduced without increasing the size of the electrode sites. A common strategy to reduce electrical impedance is to increase the surface area morphology by depositing materials with roughness and porosity. Frequently used surface-deposited materials include nanometallic materials
[79],
[80], conductive polymers (CPs)
[81],
[82], and carbon-based materials
[83], as listed in
Table 2 [80],
[81],
[83],
[84],
[85],
[86]. Nanometallic materials such as platinum nanoparticles (PtNPs) and gold nanoparticles (AuNPs) offer excellent conductivity, whereas CPs such as polyethylenedioxythiophene (PEDOT) provide superior biocompatibility. Furthermore, combining these materials can improve the overall performance. For instance, Wang et al.
[87] proposed a co-deposition method of graphene with PEDOT on microelectrode sites, which is more effective than the use of any single material in terms of enhancing the stability and decreasing the impedance. In 2024, Yang et al.
[70] developed a composite modification scheme for PEDOT-Citrate/SIKVAV. The excellent conductivity and surface roughness of PEDOT significantly reduced the impedance, whereas SIKVAV, a laminin peptide, promoted neuron adhesion and axon growth. This modification scheme ensures that the implanted electrodes maintain an SNR greater than 10 even after seven weeks. Furthermore, the deposition techniques can change the structure and morphology of the electrode site surface. For instance, depositing gold to form mushroom-shaped 3D vertical microelectrodes
[88],
[89] can reduce the distance between neurons and microelectrodes, thereby facilitating the capture of single neuron action potentials.
3.3. Applications of long-term MEA in neural mechanism and brain disease
Subject movement can exacerbate probe drift, resulting in significant variability in the data. Flexible neural probes can potentially track neural activity from the same neurons over extended periods owing to their minimal relative movement with tissue
[54], enabling the identification of stable behavior–neuron matching patterns. For instance, in maze or social interaction tests
[90], stable tracking of individual neurons allows researchers to identify functional neurons involved in spatial perception, navigation, and social interactions, thereby clarifying how neural circuits encode external information.
Stable recordings of neural activity can enable long-term monitoring of clinical brain diseases such as epilepsy, depression, Alzheimer’s disease, and Parkinson’s disease. Consistent long-term data collection from patients is crucial in understanding pathogenic mechanisms, evaluating the efficacy of drugs and other treatments, and developing new therapeutic strategies. Moreover, high-throughput and long-term stable neural data are essential for the construction of accurate and reliable neural decoders that can aid in predicting and preventing disease episodes and controlling neuroprostheses.
4. Multimodal recording MEA
Existing multimodal recordings integrated into brain–computer interfaces include substance concentration, pressure
[91],
[92], oxygen content
[93], pondus hydrogenii (pH)
[88], magnetic
[94],
[95], and light
[22],
[96]. However, we will focus on the release of neurotransmitters, as there is a clear complementarity with conventional neurophysiological signals in understanding neural activity. Neurotransmitters include cholinergic (acetylcholine (ACh)), biogenic amines (norepinephrine (NE), dopamine (DA), and 5-hydroxytryptamine (5-HT)), amino acids (excitatory transmitters such as glutamate (Glu) and aspartate (Asp); inhibitory transmitters such as gamma-aminobutyric acid (GABA); glycine (Gly); and taurine (Tau)), and soluble gases (nitric oxide (NO) and hydrogen sulfide (H
2S)). The measurement methods for neurotransmitters include chromatography, spectrophotometry, fluorescence, laser irradiation, liquid chromatography, and chemical assays. Electrochemical methods assess the concentration by measuring the current fluctuation generated by the redox reaction of neurotransmitters or their metabolites on a working electrode. They have the advantages of high sensitivity and fast response
[95]. Moreover, the electrochemical detection of neurotransmitters and electrophysiology can be easily integrated into the probe. The method of constructing a sensitive layer through surface modification is compatible with previously introduced electrode technology without significantly increasing the probe size
[97]. This approach has considerable potential for simultaneous detection of electrophysiological and electrochemical bimodal neural information in behavioral animals.
4.1. Electrophysiological/electrochemical detection methods
Since Adams
[98] first detected neurotransmitters in rat brains using electrochemical detection techniques, various methods, such as amperometry, cyclic voltammetry, and potential pulse methods, have been developed
[99]. As shown in
Fig. 4(a) [100], voltammetry involves scanning the potential to measure current changes, and hence, the identification of redox potentials, whereas amperometry involves applying a constant potential to measure the current resulting from the redox reaction of analytes. However, unlike in the > 10 kHz time resolution of electrophysiological techniques, achieving a time resolution similar to that of electrochemical methods remains a challenge. Fast-scan cyclic voltammetry (FSCV) and constant-potential amperometry are currently the fastest techniques in terms of time resolution and have been used for dual-modal detection in the brain
[76],
[101]. Constant potential amperometry analyzes the concentration of a target according to Faraday’s law by applying a constant potential between the working and auxiliary electrodes and detecting the electrolytic current of the redox reaction. Because the detection rate is limited only by the circuit acquisition rate, it has been used to capture vesicular release from neurons
in vitro [89]. However, amperometry detects all molecules that undergo redox reactions at a given potential, resulting in a lack of selectivity. For instance, DA, serotonin, and ascorbic acid in the brain can produce overlapping current signals at the same potential
[102]. In FSCV, a high-rate triangular waveform (> 100 V·s
−1) is applied to the working electrode, generating cyclic voltammograms that provide specific identification features for different chemical neurotransmitters
[103]. The improved selectivity of FSCV allows for simultaneous detection of DA and serotonin
[104]. However, for multimodal detection in freely moving animals, the integration of FSCV circuits is difficult owing to size and power consumption constraints in wearable systems. In contrast, amperometry circuits are relatively simple, making it worthwhile to explore methods such as surface modification to enhance the selectivity of amperometry. For instance, Keighron et al.
[105] designed ultrafast biosensors based on planar carbon fiber microelectrodes (CFMEs) coated with Au nanoparticles and enzyme monolayers, which can detect ACh and Glu on a sub-millisecond timescale.
The different principles of electrophysiological and electrochemical detection require distinct backend circuits, leading to the development of two synchronous detection strategies. ① Parallel measurement: In this approach, a multichannel electrode array is connected to independent circuit systems to achieve simultaneous parallel measurements of electrophysiological and electrochemical signals. Johnson et al.
[106] reported a viable strategy for parallel measurement of DA amperometric currents and electrophysiological signals using the MEA. ② Serial measurement: This strategy involves the use of time-division multiplexing circuits for
in situ serial measurements of electrophysiological and electrochemical signals at the same electrode site. Cheer et al.
[107] explored a serial strategy in which voltammetric scanning and potential recording circuits alternately used a carbon fiber electrode. Both strategies enable integrated and synchronous monitoring of neural activities. The parallel approach has the advantage of simultaneous detection; however, it cannot achieve
in situ detection of multimodal signals. The serial approach benefits from detection at the same location but limits the number of neurotransmitter channels that can be detected simultaneously. Although the integration of these methods has not yet been reported, we believe that it is a promising technical direction. Integrating high-density electrode sites with flexibly addressable electrophysiological/electrochemical signal measurement circuits could enable tracking of multiple neurotransmitter releases and neuronal firing with a higher spatiotemporal resolution. However, the application of multimodal detection backend circuits in freely moving animals poses significant challenges in terms of miniaturization and low power consumption.
MEAs used for dual-modal electrophysiological/electrochemical detection offer another advantage; both amperometric and voltammetric biosensors employ a three-electrode system. This system consists of a working electrode for target recognition, counter electrode to complete the current circuit, and reference electrode to apply a stable potential. Therefore, the stability of the reference electrode potential is crucial. Multichannel electrode arrays can integrate a reference electrode with sensing elements into a single micromachined probe, thereby enhancing spatiotemporal resolution and reducing tissue damage. Various
in vivo reference electrode materials have been explored, such as iridium oxide
[108], boron-doped diamond
[109], and polypyrrole (PPy)
[110]. Wang et al.
[111] developed a Michigan probe that integrates an
in situ Ag/AgCl reference electrode for simultaneous pH sensing and electrophysiological recordings. Mo et al.
[112] also reported an implantable dual-modal probe that integrates a PtNPs/PEDOT-modified working electrode, PtNPs-modified counter electrode, and PtNPs/PPy-modified reference electrode (
Fig. 4(b)). This probe is used for simultaneous electrophysiology and DA detection in freely moving mice.
4.2. Sensitive layer construction
A specific sensitive layer interface is fundamental for neurotransmitter detection. Conventional carbon electrode materials can capture the redox reactions of electroactive substances such as DA, NE, and serotonin. With the introduction of neurotransmitter detection in silicon-based MEAs, carbon-based materials are now preferred for microelectrode surface modification
[113]. For instance, Li et al.
[114] reported a stretchable graphene-based neural interface that can monitor monoamine neurotransmitters, including DA and 5-HT. These materials exhibit excellent chemical stability and electronic transmission capabilities, which significantly enhance the sensitivity of electrochemical detection. CPs, metal nanoparticles, and nanocomposites are also widely used for electrochemical electrode modification. Cation exchange membranes such as nafion (analytical reagent (AR) grade, Sigma-Aldrich, USA) are highly selective for cationic DA in the brain, effectively blocking interference from ascorbic acid and serotonin
[115]. Sun et al.
[116] modified electrode surfaces with NiO/CNT/PEDOT composites (
Fig. 4(c)), enabling simultaneous detection of DA, serotonin, and tryptophan in human serum.
For non-electroactive neurotransmitters, such as Glu and GABA, the sensitive interface requires biorecognition elements, such as enzymes
[117],
[118], aptamers
[119], and antibodies
[120] to achieve specific capture of the target. For instance, Glu oxidase (GluOx) catalyzes the reaction of Glu to produce hydrogen peroxide, which undergoes a redox reaction on the electrode surface. Xiao et al.
[121] reported enzyme-based
in vivo detection of Glu and DA in the hippocampus of epileptic mice for up to 10 days. GABA detection is more complex because no electroactive products are produced during GABA aminotransferase (GABAase) catalysis. The reaction product, Glu, must be further catalyzed by GluOx. Therefore, dual-enzyme modification is required using a two-step reaction to obtain mixed signals for Glu and GABA while subtracting endogenous Glu signals to determine the target concentration (
Fig. 4(d))
[118].
Various strategies exist for enzyme immobilization, including adsorption, covalent bonding, crosslinking, and electrodeposition
[122]. Electrodeposition involves the deposition of enzyme-containing polymers or nanomaterials onto an electrode surface. This approach offers the advantage of precise unit point addressing for multichannel electrodes, thereby improving spatial resolution for neurotransmitter detection. As shown in
Fig. 4(e), Xie et al.
[117] used electrodeposition to construct PtNPs/GO-GluOx-modified MEAs for simultaneous measurement of Glu and electrophysiological signals in anesthetized rats. However, the long-term stability of enzymes
in vivo remains an issue as they gradually lose activity over time, resulting in few reports of long-term
in vivo recordings of Glu.
Aptamers, which are synthetic oligonucleotides or peptides capable of specifically binding to target molecules, represent an emerging surface modification method for neurotransmitter detection
[123]. As shown in
Fig. 4(f), Zhao et al.
[119] developed an aptamer field-effect transistor neuroprobes system for serotonin detection, which exhibited exceptional sensitivity. Similar to enzyme-based sensors, aptamer- and antibody-based sensors face challenges related to their long-term
in vivo stability. Nonbiological materials for neurotransmitter detection offer advantages in terms of stability. Transition metals such as nickel
[124], cobalt
[125], and copper
[126] can oxidize neurotransmitters in alkaline environments, enabling the detection of non-electroactive substances such as Glu and ACh without oxidases
[127],
[128]. Although Ni-based sensors are far less selective and sensitive than enzyme-based sensors, they offer superior stability
in vivo, making them acceptable alternatives for long-term neurotransmitter monitoring.
Table 3 [76],
[103],
[117],
[118],
[119],
[127] presents a summary of the sensitive layers reported for various neurotransmitters.
4.3. Applications and new requirements of multimodal MEA
In the field of neuroscience, decoding neural mechanisms by synchronously detecting neurotransmitters and electrophysiology is crucial for tasks such as identifying cell types and mapping detailed chemical dynamics in the brain. This approach helps establish a correlation between brain signals and emotions/behaviors. For instance, DA has been reported to play a significant role in learning, memory
[129], reward
[130], addiction, and stress responses
[131].
Another development direction in multimodal detection is compatibility with other detection systems, such as optical imaging and MRI. Optical imaging methods based on molecular probes for freely moving animals complement large-scale extracellular recording techniques. Genetically encoded molecular probes are widely used to measure intracellular voltages
[132] and neurotransmitters
[133], providing information on cell types
[134] and biochemical states. While optical methods are limited to recordings from superficial structures, with the latest multiphoton microscopes penetrating less than 2 mm
[135],
[136], intracortical MEA can densely detect brain regions that optical imaging cannot reach. This complementarity enables a more comprehensive recording of the brain. Research on transparent MEA
[137],
[138] is a promising direction for integrating optical imaging with extracellular recording techniques to overcome the issue of blocking of the optical field of view.
MRI compatibility with neural probes is important for clinical applications. Achieving MRI compatibility involves preventing metallic components from causing artifacts and preventing radio frequency (RF) energy from inducing metal heating and electromagnetic interference. This requires the use of nonmagnetic materials, such as nonmagnetic metals
[139], carbon materials
[140], or polymers
[141], along with proper insulation.
5. Bidirectional neural probe
Bidirectional neural interfaces enable simultaneous recording and modulation of neural activity, facilitating understanding of brain function and development of therapeutic interventions. Electrical, optical, and chemical stimulation methods can help researchers decode the dynamics of neural processes. Furthermore, their ability to modulate abnormal neural activity makes them a valuable tool for treating neurological disorders and restoring lost functions.
Recent advancements in integrated bidirectional neural probes have incorporated components such as stimulation electrodes
[142], micro-light-emitting diodes (μLEDs)
[143], optical waveguides, and microfluidic channels
[144] into the neural recording probes, as presented in
Table 4 [142],
[143],
[144],
[145],
[146],
[147]. These integrations allow for simultaneous stimulation and modulation functionalities, with ongoing efforts to enhance the spatiotemporal resolution, specificity, and efficiency of neural modulation.
5.1. Electrical stimulation
ES is one of the most conventional and widely used neural modulation techniques, and DBS is extensively applied in the clinical treatment of movement disorders and psychiatric diseases
[148]. However, a significant disadvantage of ES is its lack of specificity, as it often elicits a complex spatiotemporal pattern of excitation and inhibition rings
[149]. The effectiveness of ES is affected by stimulation parameters
[150], such as amplitude, frequency, and pulse width, as well as the anatomical structure and functional state of the neural network
[151]. Consequently, ES parameters often require extensive closed-loop and personalized adjustments for optimization
[152]. Furthermore, the miniaturization of ES electrodes enables precise targeting of small groups of neurons, thereby reducing the activation of surrounding non-target neurons. Integrating ES capabilities into MEA presents challenges in terms of ensuring efficient current delivery while maintaining signal fidelity, which requires reducing the electrode impedance and enhancing the charge injection capacity (CIC) and charge storage capacity (CSC) of ES sites. Strategies for reducing impedance through surface modifications have been discussed in Section 3. Among these strategies, nanostructured coatings are often capable of simultaneously enhancing the CIC and CSC
[153]. As shown in
Fig. 5(a), Jia et al.
[142] developed an MEA that integrates both recording and ES electrodes, along with a surface modification strategy utilizing PtNPs/IrO
x nanocomposites. This modification significantly reduced the impedance and enhanced the CSC, demonstrating stable performance
in vivo for over four months.
5.2. Optical modulation
The integration of optical waveguides and μLEDs with recording electrodes for optical stimulation has enhanced the capability for precise control of neural activities. This technology can serve as a tool for optogenetics, allowing targeted neurons to be activated or inhibited by exposing light-sensitive proteins to specific wavelengths of light
[154]. Unlike ES, optical bidirectional probes offer superior cellular specificity, making them an ideal tool for unraveling neural circuit mechanisms. Optical waveguides deliver light to neurons from external light sources such as lasers. Miniaturized waveguides integrated into MEA can be fabricated using microfabrication techniques
[155]; however, additional techniques are needed to ensure precise alignment with external light sources. In contrast, μLEDs can be directly embedded within neural interfaces, providing localized light emission but potentially increasing thermal effects on surrounding tissue, as shown in
Fig. 5(b) [156].
Multicolor optical stimulation enhances the ability to independently modulate diverse cell types, enabling more precise and versatile neural regulation. Optical waveguides can achieve multicolor stimulation through external light-emitting diode (LED) or laser arrays requiring minimal modifications to the probe, as illustrated in
Fig. 5(c) [146]. However, the integration of multicolor μLEDs is more challenging. Ko et al.
[146] reported a dual-color μLED-integrated neural optoelectrode in which a flexible red μLED probe was stacked on a blue μLED silicon probe, enabling independent modulation of multiple cell types. However, the device size and fabrication complexity increase exponentially with the addition of light colors. Fluorescent coatings offer a low-cost method to alter the light frequency of μLEDs for multicolor emission
[157], albeit at the cost of reduced luminous efficiency.
5.3. Microfluidic delivery
Microfluidic channels for the delivery of drugs or chemical molecules
[158] can eliminate the limitations of the blood–brain barrier, allowing targeted administration to specific brain regions. The integration of microfluidic channels and MEA has been reported previously
[144]. These channels enable precise and localized delivery of multiple fluidic substances to target specific receptors or neurotransmitter systems, paving the way for personalized therapeutic strategies. Implanted devices combining iontophoresis and hollow silicon MEAs have been developed to enhance spatial resolution in drug modulation, enabling the integration of neuronal recordings with pathway tracing at the microcircuit (single neurons)
[153].
The combination of drug delivery methods with optogenetics and chemogenetics is a promising approach to further enhance specificity
[159]. Shin et al.
[144] combined flexible microfluidic probes with optical waveguides, enabling precise delivery of protein receptors or ligands to specific locations for targeted expression, along with simultaneous optical stimulation in the same cellular region. Another intriguing approach involves the incorporation of responsive drug carriers that react to specific triggers, such as light, temperature, or electrical signals, to achieve controllable drug release. An example is the photocaged drugs shown in
Fig. 5(d), which typically consist of a bioactive molecule and a photosensitive protective group that dissociates under specific light wavelengths, thereby releasing the original active molecule
[160]. This approach combines the high spatial and temporal resolution of optical techniques with the biological specificity of drug stimulation, thereby offering significant advantages.
5.4. Applications and new requirements of bidirectional neural interface
Stimulation artifacts are a common issue when integrating electrical and optical stimulations with recording electrodes
[161],
[162], which significantly affect accurate separation of neural signals. Further optimization is required in shielding designs for electrodes
[156], artifact suppression circuits
[163], and signal processing algorithms
[164].
Bidirectional neural probes also bridge the gap between neuroscience research and clinical applications. The bidirectional electrode arrays discussed in this section hold promise as high spatiotemporal resolution alternatives to conventional clinical electrodes. Flexible implants can enhance long-term in vivo stability, and high-density multimodal detection offers richer neuroinformatic biomarkers for these diseases. Bidirectional neural probes further enable detailed exploration of the pathogenesis of clinical disorders, paving the way for more stable, sensitive, and personalized closed-loop therapeutic systems. In the future, these technologies will be extended to applications aimed at cognitive enhancement, including improvement of memory, attention, and other cognitive functions.
6. Conclusions and future directions
This paper reviews recent advancements in intracortical neural interface technologies, with a focus on four key requirements: high spatial integration density, long-term stability, multimodal recording, and neural modulation. Progress in these four areas exhibits a trend of mutual development and convergence. Therefore, while improving individual performance metrics remains crucial, enhancing compatibility between technologies to form a complete process is the main trend in the future development of brain–computer interface technologies. We envision that combining CMOS high-density active integration with flexible substrates will increase the probe detection throughput while reducing implantation damage. Furthermore, interface modifications can further reduce immune responses and enable electrochemical neurotransmitter detection. However, several challenges remain, including the maturity of flexible CMOS fabrication technologies, complexity of multifunctional surface modification for high-throughput electrodes, and management of thermal and electrical noise in multifunctional devices, all of which need to be addressed.
Advancement in intracortical neural interface technology is expected to drive significant progress across various fields of application (
Fig. 6)
[80],
[165],
[166],
[167]. In basic research, they can provide deeper insight into neural circuit functions and the mechanisms of neural encoding and decoding. In the clinical setting, they can improve the prevention, diagnosis, and treatment of neurological diseases by providing more effective and personalized therapies. In addition, high-throughput, long-term, and stable neural interfaces can help restore motor function in patients with paralysis. Finally, neural interfaces can restore or enhance sensory perception (vision, hearing, or touch), and enable more immersive experiences in applications like virtual and augmented reality. As neural interfaces evolve, mutual growth and success of the technology and applications of implantable brain–computer interfaces are foreseeable.
Acknowledgments
This work was sponsored by the National Natural Science Foundation of China (62121003, T2293730, T2293731, 61960206012, 62333020, and 62171434), the National Key Research and Development Program of China (2022YFC2402501 and 2022YFB3205602), and the Major Program of Scientific and Technical Innovation 2030 (2021ZD02016030).
Compliance with ethics guidelines
Xinxia Cai, Zhaojie Xu, Jingquan Liu, Robert Wang, and Yirong Wu declare that they have no conflict of interest or financial conflicts to disclose.
PDF
(2677KB)
