For a long time, we lacked a clear definition of HDMEAs. Among the published studies, MEAs that labeled high density have electrode densities ranging from less than one channel per square millimeter to thousands of channels per square millimeter. The variability in HDMEA definitions can be attributed to application-specific density requirements—high for brain neural spaces under 100 μm and lower for muscular targets over several thousand square microns (Wiedemann and McDaid, 2017; Choi et al., 2020)—and the diversity in materials and technologies, which results in differing channel densities, ranging from a few to thousands of channels per square millimeter. In a recent paper, the high density was defined as the spacing among electrode sites being no more than 100 μm (Wang et al., 2023). Considering our discussion on flexible HDMEAs is focused on BMIs that targets neurons in brain tissue in this article, we mainly select flexible HDMEAs that have a pitch within 200 μm.

The fabrication of flexible HDMEAs requires a comprehensive strategy that seamlessly integrates device geometries, material selection, fabrication techniques and approaches, surface modification and packaging methods, etc. Key considerations in fabrication of flexible HDMEAs are discussed below.

This section highlights the latest advancements in flexible HDMEAs for closed loop BMIs over the past five years. The summary of the most recent flexible HDMEAs integrated in closed-loop BMIs is shown in Table 1.

Neuralink introduced a scalable high-bandwidth BMI system. This system includes arrays of 96 flexible polymer threads, each with 32 electrodes, yielding a total of 3,072 electrodes (Musk, 2019; Shown in Figure 2A). The electrodes are made of gold and treated with surface modifications by using the electrically conductive polymer poly-ethylenedioxythiophene doped with polystyrene sulfonate (PEDOT: PSS) and IrOx, in order to lower the impedance for electrophysiology and increase the effective charge-carrying capacity of the interface. The electrode array is packaged with polyimide into a compact implantable device with custom chips for low-power amplification and digitization. The entire package for 3,072 channels is less than 23 × 18.5 × 2 mm³. The thin-film HDMEAs provide superior brain tissue compatibility and precise, robot-assisted placement for targeted brain mapping. Their compact design integrates a high-channel-count application specific integrated circuit (ASIC), optimizing size and power efficiency.

Uguz and Shepard (2022) presented a high-density, mechanically flexible subdural surface microelectrode array for spatially controlled, bipolar, cortical stimulation (Shown in Figure 2B). The electrodes are made of gold, coated with PEDOT: PSS, and packaged by an ultrathin film (4 μm) parylene. These electrodes have a diameter of 15 μm and pitch of 40 μm. with high-capacitance (>1 nF) and single neuronal resolution, the electrodes can be programmed to shape the charge injection front selectively at depths approaching 300 μm with a lateral resolution better than 100 μm. To combine surface stimulation with electrical depth recording, the flexible HDMEA incorporates an implantable shank on the same substrate. The shank extension has 10 electrodes reaching down to 500 μm below the pial surface.

Zhou et al. (2023) developed a neuro-probe system, mimicking mosquito mouthparts, with four shanks, each housing 32-channel electrodes, totaling 128 channels (Figure 3A). Polyimide was used as its insulation layer, and metal stack of 5 nm chromium/150 nm nickel/50 nm gold was used as conductive layer. The flexible HDMEAs are 2.5 μm thick and 105 μm wide. The electrodes measure 25 × 15 μm² in size and are spaced 70 μm apart. These arrays were mounted on a sharpen tungsten wire to facilitate the insertion process. Each of the shanks could be inserted into the brain independently to target different regions. The system also has sensitive sensors to monitor insertion pressure without visual aid. It offers early warning detection for delicate intracranial tissues like vessels, minimizing the risk of damage during implantation. The system demonstrates outstanding postoperative performance, with a unit yield of up to 57% recorded 12 h after surgery, and 77% of unit activities being tracked from 4 to 12 weeks.

Except the general mechanical failures for MEAs, flexible HDMEAs present unique mechanical challenges rooted in their respective material and structural properties. The main failures of flexible HDMEAs in mechanical aspect could be classified into three types: (1) intrinsic failure of stable function, (2) insertion failure, and (3) failure to long time use in physiological environments.

The intrinsic failure of stability function emerges from the very nature of the flexible HDMEA, regardless of its operational environment. Typical problems include physical fracture or cracking of the electrodes, damage to the metal layer via corrosion, delamination due to poor adhesion between layers, the failure of the insulation layer led by pinholes or water penetration (Oldroyd and Malliaras, 2022), thereby preventing the development of flexible HDMEAs.

The insertion of flexible HDMEAs into biological tissues introduces a substantial challenge (Scholten et al., 2020); maintaining vital flexibility, and/or minimizing cross-sectional dimensions might undermine stiffness, potentially jeopardizing the requisite rigidity for accurate insertion and inadvertently causing microelectrode buckling, tissue damage, or electrode misplacement (Sharafkhani et al., 2022). The challenge arises if the stiffness of flexible devices is insufficient for neural tissue penetration, which often resulting in a more complex and less preferable insertion procedure (Shi et al., 2020; Xu et al., 2022).

The harsh biological setting where electrodes are placed can compromise their mechanical stability, especially long-term post-implantation (Oldroyd and Malliaras, 2022; Ramesh et al., 2022). This environment can induce electrode fracture, corrosion, delamination, swelling, dissolution, and mechanical stress, notably in the presence of continuous electrical stimulation (Oldroyd and Malliaras, 2022).

Developing HDMEAs suffers from high initial impedance because of the small-feature geometry, so that smaller electrodes are inherently noisier, exhibit worse recording quality, and are less functional because of decreased maximum possible stimulating current (Szostak et al., 2017). The rational impedance magnitude of electrodes at 1 kHz will be needed for quality recording. At least 40–150 kΩ is necessary in order to achieve the selective detection of the action potential from a single unit, but electrode’s impedance greater than 5 MΩ recording of neural signals is overpowered (Buzsáki, 2004; Prasad and Sanchez, 2012).

Even though the electrodes implanted into tissues for recording performed well in the acute phase, the performance was inconsistent and even lost the recording ability over a longer period (Merrill and Tresco, 2005). Impedance will increase as diffusional barrier is formed in the body (Merrill and Tresco, 2005; Abidian et al., 2010). Encapsulation failure and electrode breakage are another issue that might happen over time. A long-term in vivo study indicated that failure of the insulation material is the most significant factor in the reduction of both signal quality and impedance of implanted electrodes (Barrese et al., 2013).

Crosstalk is another limitation of high-density recording devices, which may occur between electrodes if the space is too close, resulting in low SNR, signal transmission error, or signal loss (McNamara et al., 2021; Qiang et al., 2021). For neural applications, the crosstalk level has to be below 1% of the recorded signal level to make it negligible compared with the background noise (Najafi et al., 1990). The electrode spacing should also be well designed, in which HDMEAs work in an excellent spatial solution with minimum crosstalk between electrodes (Cruz et al., 2019).

In neural stimulation, the perfect electrode requires a robust storage capacity to seamlessly transmit current pulses, simultaneously curbing potential variations at the electrode/tissue nexus, thereby mitigating electrode polarization and thermal accumulation (Suresh Vajrala et al., 2021), which is a task that’s notably arduous for flexible HDMEAs.

Apart from the neural interface device itself, there’s an inconspicuous yet critically important technical challenge: the integration of flexible sensor and backend stage (Behfar et al., 2021). This unique challenge restricts system integration and directly affects the stability and performance of HDMEAs (Luo et al., 2023). SNR and low latency can potentially be improved through advancements in algorithms and backend technologies (Boran et al., 2019; Siddiqui et al., 2019; RaviPrakash et al., 2020), but the wiring issue is the true critical problem that deeply troubles HDMEA designers. Traditional silicon-based probes, benefiting from the development of modern electronics, have various mature and stable connection methods. However, the flexibility of the HDMEAs itself becomes a challenge for stable connections with subsequent systems (Jang et al., 2022; Miyamoto et al., 2022). Setting stability aside, the hundreds or even thousands of connection points on HDMEAs make manual connections extremely difficult without specific tools or methods. This also restricts the possibility of expanding more electrodes on flexible HDMEAs (Tchoe et al., 2022). The reliability issue of the soft-hard interface presents an additional challenge for making high density interconnection between flexible HDMEAs and backend electronics. The instability caused by inconsistent stress/strain concentration at the interface of two mechanically dissimilar materials can lead to delamination failure.

In the existing literature, comprehensive reviews (Oldroyd and Malliaras, 2022; Zeng and Huang, 2023; Ziai et al., 2023) discussed major chemical challenges of developing and applying MEAs in biological circumstances. Here we provide a concise overview. The challenges posed by the chemical environment to flexible HDMEAs are multifaceted, which include (1) the effects of the physical chemicals on electrodes, (2) the effects of the chemicals from the HDMEAs to the host tissues, and (3) stability for the long-term use. The physical intricate mix of organic and inorganic chemicals can impact the performance and longevity of the flexible HDMEAs. Despite the diverse materials and methods used in the fabrication of MEAs, many are not adequately prepared to handle the demanding physiological conditions. The physical chemical environment can threaten the stability of MEAs over both short and long durations, owing to a range of organic and inorganic reactions.

Additionally, the electrode’s inherent chemical instability can be potentially harmful to the organism. Pt, for instance, has long been regarded as highly biocompatible and non-toxic. However, a recent study by Shepherd et al. (2021) revealed that prolonged stimulation (lasting 6 months with a charge density of 267 μC/cm²/phase) of Pt electrode arrays led to a notable production of Pt particulates. This, in turn, resulted in an escalated fibrous tissue response and weakened the quality of long-term records.

Biocompatibility is a crucial criterion for materials utilized in biosensors, ensuring they integrate harmoniously within biological environments. Yet, even when employing these biocompatible materials, inherent biological challenges continue to arise and need addressing (Otte et al., 2022). Risks that can cause damage to the tissue and induce an acute inflammatory reaction include direct damage during surgery, mismatches between the electrode and tissue, micromovements of the electrodes, damage by released chemical from the electrodes, etc. (Kozai et al., 2015). The body’s responses to implanted HDMEAs include acute (Polikov et al., 2005; Kolaya and Firestein, 2021; Wu et al., 2021) and chronic (Golabchi et al., 2019) immune responses and damage repair processes, involving several distinct cell populations, such as erythrocytes, glial cells, neurons, etc. The processes include:

Recent advancements in HDMEAs involve utilizing new materials for substrate and/or electroactive components or innovating with traditional materials to endow them with flexible properties. Compared to traditional materials, the newly developed materials possess unique properties specifically designed to address the demands of creating flexible HDMEAs. These properties include enhanced stretchability, better mechanical control, greater flexibility, and superior conductivity, among other advantageous characteristics. As a result, they are better equipped to tackle challenges in the mechanical, electrical, chemical, or biological domains. Some reviews by Gablech and Glowacki (2023) and Tringides and Mooney (2022) have been discussed about the material advances in MEAs development, thus we will just give an overview in this part.

To enhance biocompatibility, flexibility, and overall performance in neural interfaces, the advancement of substrate materials for flexible HDMEAs has incorporated innovative options such as hydrogels, silk, and polymers. Hydrogels, networks of hydrophilic, cross-linked polymer chains, have emerged as promising candidates for the next generation of bioelectronic interfaces due to their mechanical property similarity to biological tissue and versatility across electrical, mechanical, and bioengineering fields (Zhang and Khademhosseini, 2017; Park et al., 2021). Nevertheless, hydrogels also present challenges as substrate materials for flexible HDMEAs. Hydrogels’ low Young’s modulus can cause mechanical instability and deformation, their susceptibility to swelling and dehydration may alter interface stability, and their fabrication complexity restricts shaping for varied applications. By using the natural silk as an optical waveguide material, Zhou et al. (2022) presented and flexible opto-electro neural probe, in which electrode arrays of 128 recording channels were integrated on a single probe (Figure 4A). Silk has high transparency, excellent biocompatibility, and mechanical controllability. The Silk-Optrode probe, upon hydration of the silk optical fiber, autonomously adapts to its post-implantation environment, minimizing its mechanical stiffness to facilitate high-fidelity brain insertion while preserving mechanical compliance with adjacent tissue.

To improve the flexibility of the conductive parts, graphene (Sun et al., 2021), carbon nanotubes (CNTs) (Shih et al., 2020; He et al., 2021), and conducting polymers (CPs), such as PEDOT: PSS (Kshirsagar et al., 2019) have been developed for neural interfaces. Graphene, renowned for its notable flexibility and electrical conductivity, has become a pivotal material in neuronal interface studies and has been utilized to fabricate a range of flexible and stretchable electronic devices (Park et al., 2018). Still, it is vital to concurrently acknowledge and investigate its potential drawbacks, including long-term in vivo toxicity. Pure PEDOT: PSS hydrogels is brittle yet presents a high conductivity on the level of ≈ 40 S/cm. However, Li et al. (2022) created a highly conductive and stretchable double-network (DN) conducting polymer hydrogel from PEDOT: PSS and poly (vinyl alcohol; PVA) and achieved through in situ aggregation and densification, offering promising characteristics like high PEDOT: PSS content, electrochemical properties, and biocompatibility for potential use in bioelectronic applications (Figure 4B). Lee et al. (2022) developed flexible and transparent ultrathin (<10 nm) gold MEAs by using a biocompatible polyelectrolyte multilayer (PEM) metallic film nucleation-inducing seed layer (Figure 2C). With the polymer seed layer, the ultra-thin gold film created through thermal evaporation exhibits effective conductivity, along with high optical transparency and superior mechanical flexibility. Furthermore, liquid metals (notably eutectic gallium–indium, EGaIn) have garnered significant interest in the realm of stretchable biodevices (Dong et al., 2021), due to their exceptional mechanical attributes, electrical conductivity, and biocompatibility.

The utilization of dual-side design has emerged as a vital strategy to increase the electrode density in flexible neural interface applications across several research papers. By employing a dual-side design, it is possible to enhance electrode density without enlarging the probe’s dimensions or reducing electrode size, thereby circumventing associated electrical challenges. Scholten et al. (2020) introduced a novel polymer-based microelectrode array with an impressive 512 platinum recording electrodes, optimized for chronic recordings in the brains of behaving rats, and showcase advancements in polymer microfabrication and back-side electrode patterning. In the work of Kim et al. (2023), a dual-side fabricated multimodal polymer neural probe was developed, featuring gold and platinum microelectrodes. Although a strategy of using multilayer with a sacrificial layer on the bottom could fabricate dual-sided flexible HDMEAs, it introduces challenges, such as misalignment during the multiple lithography processes (Liu et al., n.d.). A more straightforward and efficient method is still needed. Soft and stretchable electronics have been developed to create HDMEAs that can better conform to the brain’s surface and accommodate its natural movements. With some ingenious designs, these devices not only exhibit greater flexibility and stretchability but also reduce mechanical issues common in traditional designs. Moreover, due to improved contact between the electrode and tissue, they can ensure enhanced electrical performance. The stretchable opto-electric integrated neural interface (SOENI) is designed to align with the mouse brain’s somatomotor and somatosensory cortices (Ji et al., 2020a). It features a 3 × 3 microelectrode grid, 2 × 2 micro-LEDs, and a large reference electrode. Embedded within two silicone elastomer layers, the SOENI combines photostimulation and recording electrodes, with microscale light emitting diodes (LEDs) connected using a flip-on-chip method, aligning their illuminated surfaces with the microelectrode sites. Serpentine-shaped metal interconnecting wires were designed to improve the stretchability. Similar serpentine structures were also used in some other latest microsystems (Ji et al., 2020b; Xiang et al., 2022).

Mesh-like electrode designs have also been explored in order to improve the conformability of HDMEAs to the brain’s surface and minimize mechanical stress on cell or tissue. For example, Le Floch et al. (2022) reported “tissue-like” stretchable mesh nanoelectronics designed to align with brain organoids’ mechanical properties, which can be folded into 3D structures by progenitor or stem cells, facilitating three months electrophysiological measurements. Inspired by implantable mesh electronics and organoid polymer scaffold growth, McDonald et al. (2023) developed suspended mesh microelectrode arrays for neural organoids, incorporating four wells, each containing a mesh and 61 microelectrodes, which are adept for low-noise recordings and electrical stimulation with their sub-100 kΩ impedance at 1 kHz. These HDMEAs emphasized their ability to improve conformability to the brain’s surface and minimize mechanical stress. These designs represent a significant shift toward structures that are more adaptable and less invasive than traditional electrodes.

Surface modification strategies not only address the challenges of electrical, mechanical, and biological compatibility but also significantly enhance the overall performance of HDMEAs in closed-loop BMIs. A common method of surface modification is adding a layer of special materials on the electrode surface to change the electrode’s properties. These materials include metals (Hempel et al., 2017), conductive polymers (Bianchi et al., 2021), nanostructures (Fattahi et al., 2014; Barai et al., 2018), and bioactive substances (Vitale et al., 2018; Golabchi et al., 2020). PEDOT and its derivative are used to improve the electrical properties (Hempel et al., 2017; Liu et al., n.d.) and modify the physical properties, such as transmittance (Yang et al., 2021b). The application of CNTs significantly increases the surface area, thereby increasing the charge storage capacity and the injection limit, with other benefits such as good adhesion, non-toxicity, and stable properties (Fattahi et al., 2014; Barai et al., 2018). Yang et al. (2023) used polypyrrole/carboxylated single-walled carbon nanotubes (PPy/SWCNTs) nanocomposites for electrode modification. The nanocomposites not only improved the performance of microelectrodes with low impedance (60.3 ± 28.8 kappa O) and small phase delay (−32.8 ± 4.4 degrees), but also presented stability for in vivo stimulation and recording for 21 days, as shown in Figure 3B. Neuroadhesive protein coating improves the chronic performance of neuroelectronics in the mouse brain (Golabchi et al., 2020).

Innovations in surface topology, with or without additional coating, have demonstrated improved recording quality compared to traditional planar electrodes. Mushroom-shaped microelectrodes improved the recording quality compared to planar MEA (Shmoel et al., 2016). The enhanced microscale wrinkles on microelectrodes were obtained by oil extraction from the elastic substrate and electroplating modified materials PEDOT: PSS and platinum black (Pt-black) on the wrinkled microelectrode sites improve the total device performance in electrocorticography (ECoG) signal recording without causing any cracks, delamination, or exfoliation (Shi et al., 2021).

Anisotropic conductive film (ACF) bonding process has become a feasible approach to address the high-density interconnection challenges of flexible devices (Dagdeviren et al., 2014; Luo et al., 2023), yet, the high-temperature, high-pressure bonding process limits the use of some flexible materials. However, Hwang et al. (2021) introduced a stretchable anisotropic conductive film (S-ACF) capable of connecting high-resolution stretchable circuit lines to various electrodes, addressing the challenge of high-resolution stretchable interfacing at low temperatures through conductive microparticles in a thermoplastic film (Figure 3D).

Diverging from ACF, manufacturing a standard flat flexible cable (FFC) directly on the sensor to accommodate different connectors is also a viable approach. The zero insertion force (ZIF) connector can address the connection issue to some extent (Song et al., 2019; Kang et al., 2022), especially with its mechanical connection features, allowing sensors to be easily disconnected. This provides many conveniences for wearable, disposable HDMEAs. The easy-to-replace feature also makes it convenient for researchers to replace damaged sensors, which is especially valuable in some destructive experiments. However, this method seems a bit cumbersome when facing thousands of connection points. Especially the weight, size, and cost of the ZIF connector further restrict its application scenarios. The simplest method is to directly flip the sensor and affix the designed FFC directly onto a printed circuit board (PCB) or flexible PCB with conductive adhesive (Behfar et al., 2021). This method is straightforward, low-cost, and very suitable for HDMEAs with a smaller number of electrodes or in the prototype stage.

The instability of soft-hard interface can be addressed by either enhancing the adhesion between the materials or avoiding sudden changes from soft to hard, that is, by adding an intermediate layer/buffer layer to mitigate the transition. Ideally speaking, making the entire backend into flexible devices to form a fully flexible system would be best. By eliminating the differences between materials, this problem can be completely addressed. Zhang et al. report a capacitive pressure sensor (Zhang et al., 2022). Impressively, this sensor used CNT-doped PDMS, achieving both electrode and insulating materials using CNT-doped PDMS through tuning different doping concentrations. To some extent, this realized part of the full-flexibility goal. Such partial implementation alone resulted in surprisingly stable performance; the device remained stable even after 100,000 deformation cycles. This further confirms that aiming for a fully flexible design is undoubtedly a solution to the soft-hard interface issue.

Various recent fabricating strategies, including multi-layer strategies (Pimenta et al., 2021) and dual-sided micropatterning (Tooker et al., 2012; Pimenta et al., 2021; Kim et al., 2023), have been proposed to attain high-density probe capabilities without enlarging the polymer neural probes’ dimensions. Meanwhile, the advances in fabrication techniques also facilitate the development of flexible HDMEAs. Traditional fabrication techniques, such as photolithography, have significantly advanced the development of flexible HDMEAs, yet they come with limitations, including intricate processes, challenges in fabricating 3D structures, and precision constraints. Recently, emerging technologies, such as precise laser ablation and laser lithography, have introduced more streamlined, direct, and even more accurate methods for making complex electrode structures.

Laser ablation, or laser patterning, facilitates the streamlined fabrication of planar and 3D microelectrodes without the need for photolithography and etching by eschewing multi-step procedures for a more straightforward implementation (Tanwar et al., 2022). For example, soft and flexible gold microelectrodes with widths down to 3 μm were fabricated on a pliable PDMS substrate through a combination of supersonic cluster beam deposition and femtosecond laser processing (Dotan et al., 2021). Inkjet printing, garnering popularity for creating MEAs by directly depositing microelectrodes onto the substrate through single-step fabrication, facilitates the use of softer substrates compared to photolithography and permits the use of alternative conductive inks like PEDOT: PSS for printing conductive lines on soft substrate directly (Mandelli et al., 2021; Kim et al., 2022). However, it does come with the drawback of being notably expensive.

To effectively reduce crosstalk between interconnects, it is necessary to increase the distance or add a shield layer (Qiang et al., 2021; Naughton et al., 2022). Spacing them too far apart will inevitably increase the cable width and enlarge the overall system size. The core contradiction lies between the excessive number of wires and limited space. Some designs compensate for width by adjusting thickness, that is, by using multiple layers to reduce the number of wires that each single layer needs to accommodate (Shin et al., 2021; Saleh et al., 2022). Passive matrix and active matrix have become the mainstream architectures to address the signal readout of the array (Miccoli et al., 2019; Luo et al., 2023). The passive matrix design is relatively simple, consisting of intersecting rows and columns, with sensors set at the intersections (Sundaram et al., 2019; Kang et al., 2022). However, the crosstalk between sensors limits the quality of the signal. The active matrix addresses the crosstalk issue by adding switching units, such as diodes, to the electrodes (Park et al., 2019; Gwon et al., 2022). With the ideal performance comes complex circuit design requirements and internal structural issues. Nowadays, the flexible active matrix has become a mature technology used in the commercial display industry, which is worthy of designers’ study.