Brain–computer interfaces—which allow direct communication between the brain and external computers—have potential applications in neuroscience, medicine and virtual reality. Current approaches are, however, based on conventional rigid electronics and are limited by their intrinsic mechanical and geometrical mismatch with brain tissue. Flexible electronics, which can have mechanical properties compatible with the brain, could address these limitations and be used to create the next generation of brain–computer interfaces. Here we explore the use of flexible electronics in the development of brain–computer interfaces. We examine the unique advantages of flexible, stretchable and soft electronics in such interfaces and consider the potential impact of the technology on neuroscience, neuroprosthetic control, bioelectronic medicine, and brain and machine intelligence integration. We also explore the challenges in materials, device fabrication and system integration that need to be addressed to develop flexible brain–computer interfaces of general applicability.
Brain–computer interfaces (BCIs) allow direct communication between the brain and external computers and controllers. Current systems can interface with neurons in the brain, record neural signals, extract brain internal states for external control and deliver feedback signals for neuromodulation1. The potential applications of BCIs—which extend across the fields of neuroscience, neuroprosthetics, bioelectronic medicine, and virtual and augmented reality—include decoding neural computational codes2, restoring the lost functions of patients3, treating neurological disorders4 and augmenting information5.
The key technology of BCIs is the neural interface for signal collection. Non-invasive neural interfaces, such as electroencephalography, offer a preliminary way of extracting brain signals and are advantageous because they perform large-scale monitoring with little to no risk. However, the dura mater, skull and scalp function as low-pass filters, only permitting signals of low-frequency bands such as alpha and beta waves to be detected6. This leads to the loss of critical information in BCI applications. Multiplexed neural recording of millisecond-timescale action potentials from individual neurons—the basic computational units of the brain—is thus critical for high-performance BCIs.
The emergence of microfabrication techniques has allowed miniaturized penetrating multi-electrode arrays (MEAs) to be developed for cellular-level neural interfaces. For example, the Michigan probe—a silicon-based depth MEA for neural recording—has demonstrated a sufficient signal-to-noise ratio for single-neuron-resolved action potential discrimination7. To improve the throughput of recording, the Utah array was developed, which gathered up to 100 electrodes in a 4 × 4 mm2 region to record a relatively large number of neurons8. This was the first invasive neural interface approved by the United States Food and Drug Administration for human testing. Clinical BCI studies have used Utah arrays in patients to, for example, control robotic limbs3, restore sensations with microstimulation9 and decode dexterous handwriting movement for text translation10.
In recent decades, lithographic fabrication techniques have advanced drastically in the semiconductor industry and are now starting to influence neural interfaces. On the one hand, the number of transistors in an individual chip has increased to match or even surpass the number of neurons in a human brain, offering the potential to monitor a large scale of neurons or even all the neurons in the human brain11. On the other hand, sensors and stimulators are now miniaturized to the subcellular size12, making it possible to perform long-term stable recording and stimulation of individual neurons. Neuropixels13,14, for example, is a millimetre-wide, centimetre-long silicon depth probe that is based on a 130 nm fabrication technique and contains on-chip complementary metal–oxide–semiconductor (CMOS) circuits for multiplexing and amplification. It can simultaneously drive 5,120 individually addressed electrodes and record hundreds of neurons from animal13,14 and human brains15. Alternatively, densely bundled microwire arrays have been created that are based on CMOS technology and are capable of large-scale neural recording in deep rodent brain regions16.
Despite this progress, further advances in fabrication capabilities, which allow more channels to be integrated into conventional rigid probes, will not address the fundamental challenges that exist in building next-generation BCI technology (Fig. 1). In particular, in human brains, more than 85 billion neurons are distributed across a three-dimensional (3D) volume at a spatial scale from centimetres to metres, whereas the size of individual neurons is only 10 µm to 100 µm. Single-neuron action potentials also operate on millisecond scales, whereas brain activity takes days, months or even years to build up emotions, cognition and learning. Finally, there are different types of neuron in the brain17, each showing distinct gene expressions, morphologies and connectivity that collectively give rise to unique electrical behaviours.
Next-generation BCIs therefore need to interface with many neurons at single-cell resolution across a large volume of the brain; stably track and modulate the single-cell- and millisecond-resolved neural activities from the same cells across a large temporal scale; and address the genetically targeted and cell-type-specific components in the brain. Next-generation BCIs also need to be scalable to address a statistically significant number of neurons and ultimately be able to interface neurons across the entire 3D brain over an extended period.
Current state-of-the-art rigid electronics cannot address these issues because single-neuron action potential recordings are limited by the low temporal stability, and therefore rigid brain probes require computational methods to realign the signals14. Rigid electronics also cannot adapt to the volume change of the brain during development, ageing and disease, limiting their applications in children, elderly people and patients. Finally, bulky and rigid electronic structures introduce acute and chronic mechanical damage to the brain, limiting its scalability in terms of spatial coverage and channel numbers and its ability to integrate multifunctional electronics to address cell-type-specific cellular components. Flexible electronics could be used to address these limitations18,19,20,21,22,23,24,25.
In this Perspective, we explore the potential of flexible electronics in the development of next-generation BCIs. We categorize flexible electronics into three areas—flexible, stretchable and soft (low modulus) electronics—and discuss their unique advantages in the creation of BCIs. We then examine the potential impact of building flexible BCIs on neuroscience, neuroprosthetic control, bioelectronic medicine, and brain and machine intelligence integration. Finally, we consider the engineering challenges that need to be addressed to realize the full potential of the technology.
A limitation of current BCIs is the inability to chronically and stably track neural activities from the same neurons in the brain. Therefore, the recorded neural signals are unstable, requiring daily recalibration to stabilize the signals for brain decoding26. Fundamentally, the signal instability is due to the large mechanical and geometrical mismatch between conventional rigid electronics and soft brain tissues27. Implanting conventional rigid electronics into the brain inevitably introduces micromotion-induced probe drift during brain movement (Fig. 2a). The probe drift not only introduces chronic mechanical damage to the brain but also causes a breach of the blood–brain barrier. This triggers immune responses, gliosis and neuronal degeneration during long-term implantation. For example, rigid implants trigger the mechanical activation of astrocytes and microglia. The proliferation, migration and accumulation of these immune cells form a 50–200-μm-thick scar sheath around the implanted probe, separating the recording site from neurons. Moreover, blood–brain-barrier breach and the immune response will increase the chance of chemical corrosion of the brain probe and neurodegeneration. Over the course of implantation, the signal-to-noise ratio of recording with conventional rigid electronics gradually decays until single-neuron action potentials completely vanish.
Flexible electronics that have mechanical properties compatible with the brain are therefore needed to achieve stable recording at the single-cell level. Young’s moduli of electronic materials used for standard lithographic fabrication are at least six orders of magnitude higher than that of the brain. Thin-film electronics that match the flexibility of the neuron have thus been explored. Reducing thickness is crucial as the flexibility (that is, bending stiffness) is proportional to Eawh3, where Ea, w and h are Young’s modulus, width and thickness of the device, respectively. The active layers of electronics—such as metal interconnects, transistors and electrodes—are typically nanometre thick. Instead of the bulky silicon/silicon dioxide substrate, the thin-film electronics fabrication process allows the transfer of nanoelectronic components to micrometre-thick polymer substrates such as polyimide18, SU-824 and parylene28.
This form of flexible electronics can readily conform to multiple 3D structures of the nervous system and interface the less accessible tissue regions such as brain raphe, spinal cord and peripheral nerves29,30. For example, owing to the intimate and conformal coating to the brain tissue, NeuroGrid, a thin-film electrode array with low-impedance poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) electrodes, can detect the single-cell action potential from the surface of the brain28. In addition, implantation of flexible electronics with thicknesses of 10–100 µm into brain tissue has been shown to substantially reduce chronic immune response25.
Going further and eliminating the chronic immune response requires completely matching the mechanical and geometrical properties of electronics to brain tissues, creating tissue-like electronics24,31. This has been achieved by fabricating ultra-flexible nanoelectronics with feature sizes of less than 10 µm and thicknesses of 1 µm. The bending stiffness of such electronics has been reduced to be compatible with that of individual neurons or even axons22. After brain implantation, tissue-like electronics can provide seamless integration with the neural tissue (Fig. 2b). Post hoc immunostaining and imaging of brain tissues have, in particular, shown normal neuron density and little to no immune response at the brain–electronics interface after months to one year of implantation.
Leveraging this stable interface, tissue-like electronics can record single-neuron-resolved action potentials in behaving animals over an extended period21,22. The incorporation of the mesh-like design into the electronics enables interwoven structures between the electronics and neural tissue after recovery from the initial implantation. These structures further strengthen the stability of the electronics-to-neuron interface, allowing mesh electronics to stably track the action potentials from the same neurons over the entire adulthood (>1 year) of behaving animals32.
Brains undergo large volume changes during early development, growth and ageing. Sometimes, diseased brains can also shrink and swell. To achieve stable tracking of neural activities from the same neurons throughout the entire lifespan of animals and humans, flexible electronics need to be made stretchable to accommodate the tissue volume change while maintaining a high level of performance (Fig. 2c). Stretchable electronics are also critical to interface with the spinal cord and peripheral nervous system that undergo recurrent movement. There are two general methods to achieve stretchable electronics: stretchable structure design and stretchable electronic materials33.
With stretchable structure design, functional units (sensors and computing units, for example) can be placed in the rigid-island structure while the interconnects can be designed into serpentine, wrinkle, buckling or origami structures to enable stretchability33. For example, implementing a serpentine structure in mesh nanoelectronics has achieved stretchable mesh nanoelectronics that have been implanted into human-stem-cell-derived brain organoids34. The embedded nanoelectronics can fold and expand together with 3D brain tissue in early development and accommodate the volume change, capturing the emergence and dynamics of neural activities.
Alternatively, stretchable, elastomeric and viscoelastic electronic materials such as stretchable conductors, semiconductors and dielectrics have been developed for the fabrication of functional electronics. For example, electronic dura mater made with silicone and platinum–silicone-coated composite can sustain millions of stretch cycles after being implanted in the spinal cord30, providing a biocompatible interface for stably recording and modulating spinal cord functions over weeks. In addition, viscoelastic materials have been used in conductive and encapsulating layers, achieving flexible and stretchable surface electrode arrays conformable to the cortex35. It has also been shown that these stretchable electronics can grow with and functionally stimulate the peripheral nerve in young rats36.
To extend the spatial coverage of the flexible electronics in the brain while maintaining single-cell-resolved recording, the number of integrated sensors and stimulators can be scaled up using lithographic nanofabrication. In 2019, 1,024-channel, 14-µm-thick flexible electrode arrays connected with field-programmable-gate-array-enabled multiplexing circuits were demonstrated to record high-quality action potentials and local field potentials for more than 5 months18. In the same year, Neurotassel, a 1,024-channel 1.5-µm-thick flexible electrode array, performed stable recording over two months37. Connecting this flexible electrode array with the integrated multiplexing circuits can further increase the number of simultaneously addressable electrodes. Later, polyimide-encapsulated ‘threads’ were reported that contain 3,072 gold electrodes connected with application-specific integrated circuits (ASICs) for on-chip amplification, multiplexing and wireless transmission, in different animal species19. A more recent study developed a modularized system of ‘thread’ electrode arrays that integrate multiple 128-channel devices to reach altogether more than 1,000 channels and demonstrated its capability to stably record thousands of neurons in the rodent brain over an extended period of time38.
The density of the electrodes also needs to be boosted to match that of the Neuropixels, which reaches more than 1,000 electrodes per square millimetre. To achieve this without changing the mechanical and structural properties of the flexible brain probe, one way is to decrease the dimensions of interconnects by using electron-beam lithography39. However, substantially reducing the dimensions of interconnects without introducing crosstalk among channels and signal attenuations during transmission is challenging. Therefore, studies have used thin-film transistors for on-site multiplexing or 3D vertical stacking of the flexible electrode array, which increased the density of electrodes without compromising their performance.
Specifically, Neural Matrix integrated the flexible silicon membrane transistor-based active matrix, driving a 28 × 36 electrode array in 9 × 9 mm2 for non-human primate brain mapping40. An SU-8 probe with a multilayer electrode array has also been demonstrated to increase the local electrode density41. However, both methods introduced an increase in the thickness of the electronics. In particular, multiple high-performance thick dielectric encapsulation layers are needed in the transistor circuits to reduce the crosstalk between channels and prevent leakage during long-term exposure to the physiological solution.
Based on Eawh3, substantially increasing thickness (h) triggers the development of softer materials with lower modulus to maintain tissue-level flexibility in electronics. A variety of soft and intrinsically stretchable electronics made with elastomers42,43,44 have thus been developed. However, their long-term stability needs to be carefully evaluated owing to their intrinsic limitations such as susceptibility to the progressive ion penetration45 and incompatibility with lithographic fabrication.
The introduction of soft—that is, low-modulus and intrinsically stretchable—electronics into BCIs allows the targeting of cell-type specificity. First, based on existing research25,30, multifunctional electronics can potentially be incorporated and stacked for multimodal cell-type-specific neural recording and modulation (Fig. 2d). Example modalities include optical stimulation, microfluidic channel-based chemical delivery, thermal sensing, and electrical sensing and stimulation. Second, the soft electronic materials can be formed in situ on a specific group of cells by genetically targeted chemical synthesis. The early-stage work demonstrated that conductive pathways can be built from these genetically targeted electrical components46. The approach could thus potentially be used to develop a future BCI technology that can link specific types of neuron to the electrodes for recording and modulation. Soft electronics could also potentially be embedded in brain tissue for post hoc staining and imaging22,32. This could allow integration of cell-type-specific in situ imaging and sequencing information with electrical recording in a scalable manner, which could be used to build cross-modal prediction models for identifying specific cell types based on their electrical activities47. This would be of use in future applications that integrate multimodal information to decode neural signals.
Modifications of flexible electronics
The biocompatibility of BCIs based on the flexible, stretchable and intrinsically soft electronics (subsequently generalized as flexible electronics) may be further improved by incorporating biomaterials and biological components. First, using bioactive molecules, anti-inflammatory molecules and adhesion proteins to modify the surface of flexible electronics may reduce the immune response and improve device biocompatibility48 and the long-term stability of recording from the same neurons. For example, surface modifications such as bioactive coating with poly(p-xylylene) and poly(ethylene glycol) methacrylate have been demonstrated to have the capability of controlling undesirable biological processes such as biofouling caused by implantation49. More recently, robust and transparent hydrogel electrodes based on PEGylated silk fibroin were reported to have compliant mechanical properties such as moderate Young’s modulus (1–10 MPa) and stretchability50. ‘Living electrodes’, in which living neurons are encapsulated within a cylindrical implantable device51, have also been developed. Axons of these neurons then formed synaptic connections with the host neurons, which could be recorded and stimulated externally via the encapsulated neurons. This approach can potentially be integrated with flexible electronics to further enhance biocompatibility.The ideal BCIs should offer tissue-level flexibility and softness, subcellular feature size and an interlocked system at the electronics-to-neuron interface. By decreasing the thickness and ultimately using soft materials, electronics with thin-film structures can achieve tissue-level flexibility and softness41. Elastomeric neural probes with fibre structure and micrometre-scale thickness have also been created that can achieve a mechanically compliant and long-term stable interface at the deeper neural tissues52. The dimension of these electronics with thin-film or fibre-like structures can be further decreased to achieve subcellular feature size and thus reduce chronic tissue damage.
To improve the stability of flexible BCIs, the current approach involves the formation of an interwoven neural network at the interface through mesh-like flexible electronics. For example, a biomimetic design has been pursued to fabricate flexible electronics with the structural and mechanical properties of individual neurons22. Specifically, such electronics have metal electrodes and interconnects closely mimicking the geometries and dimensions of neuronal soma and neurites, respectively, providing an open mesh-like environment most suitable for the habituation of neurons while maintaining tissue-like flexibility. The various modification strategies available could also potentially be integrated into flexible electronics to further enhance the stability of flexible BCIs.
Impact on neuroscience
Flexible BCIs could be used to record and modulate large-scale neural activities at single-cell and millisecond resolution over months and years with cell-type specificity and integrate with multimodal and multifunctional interrogation and intervention methods. These properties are important to understand the complex and dynamic neural population activities, which are the collective activities of a group of neurons, during development, learning, memory and ageing, and accurately decode the brain states (Fig. 3a). For example, large-scale in vivo recording15 is fundamental to mapping functional connectomics of the brain, which can illustrate how the different brain regions connect and coordinate, and, through behaviour-dependent measurements, identify critical brain regions and specific groups of neurons that correspond to the targeted behaviours for BCI applications53.
Furthermore, chronically stable neural recording across multiple brain regions with flexible BCIs is critical to study important biological phenomena such as neural representational drift54, in which individual neurons’ activities mapped to the same sensory stimuli or behavioural outputs change over time, thus further elucidating fundamental neural coding mechanisms. Moreover, multifunctional flexible electronics with multimodal interrogation and intervention capabilities can integrate cell gene expression and connectivity information with single-cell electrophysiology47. Leveraging these data in multimodal machine learning algorithms55, we could identify neuronal subtypes for the targeted recording and modulation, integrate behaviour-defined cell types with molecularly defined cell types to identify biomarkers for neurological disorder treatment, and infer the cell-type identity based on brain recording for more accurate brain-state decoding.
Impact on neuroprosthetic control
Recent advances in BCI have enabled functional restorations such as movement and speech in individuals9,56. However, the recording of single-neuron activities by conventional rigid electronics changes over time due to probe drift, brain immune response and neuron loss, limiting the longitudinal accuracy of brain decoders for neuroprosthetics. As a result, decoder recalibration or stabilizer needs to be implemented to retain the decoding accuracy26. We argue that flexible BCIs could bypass these compromises and improve the performance of the decoder and thus the neuroprosthetic control by stably tracking neural activities from the same neurons over months and years (Fig. 3b). The initial stable recording could then be used as training data in machine-learning-based algorithms for patient-specific, self-programmable and energy-efficient long-term stable decoding.
Moreover, large-scale, long-term stable recording enables us to identify the low-dimensional, intrinsic variables of neural population activities such as neural representation57. It is known that the brain’s response to the external environment and stimuli can often be represented by low-dimensional variables58. For example, the neural activities of mouse head-direction neurons during foraging can be organized around a ring structure59 characterized by a one-dimensional intrinsic variable, which is substantially smaller than the number of recorded head-direction neurons. Extracting the low-dimensional representation can identify the most critical information, thus eliminating unnecessary complexity for successful neural decoding. As it is challenging, if not impossible, to extract correct intrinsic low-dimension variables from unstable recording due to the inevitably biased estimation, we envision that the chronically stable in vivo recording with flexible BCIs offers an opportunity to accomplish this task and enable a consistently accurate, efficient, robust and low-power neural decoder for neuroprosthetics.
Impact on bioelectronic medicine
Flexible BCIs can improve precision bioelectronic medicine4 by providing stable modulation of the individual neurons in neural circuits over time (Fig. 3c). Current rigid electronics can typically modulate only neural activities at the brain-region level comprising thousands of neurons and still suffer from immune-response-induced efficiency loss. Alternatively, flexible BCIs can provide closed-loop feedback and control based on the stable tracking of neural activities with neuron-subtype specificity. This potentially enables flexible BCIs to selectively restore single-neuron and neural circuit activities for ameliorating abnormal neural activities in neurological disorders, facilitate neural progenitor cell migration and integration through chronic neuromodulation, and modulate a large number of neurons based on their spatiotemporally orchestrated activity patterns for perceptual restoration.
Impact on brain and machine intelligence integration
Flexible BCIs create new opportunities for the integration of natural intelligence and artificial intelligence, with potential applications such as self-programmable and long-term stable real-time brain decoding with bio-inspired machine learning algorithms. Moreover, using flexible BCI-enabled large-scale and chronically stable electrical recording from the same neurons when animals perform various behaviours as training data, artificial neural networks can potentially learn and integrate neural representations from various behaviour-dependent neural population activities for brain simulation (Fig. 3d)—a task unachievable by rigid electronics due to the signal instability-induced trial-to-trial variations60,61.
When combined with programmable circuits (such as field programmable gate arrays), and even neuromorphic circuits (such as memristor crossbar arrays62), neural activities could also be imprinted on the hardware through designed learning rules. One existing study using CMOS-driven intracellular electrode arrays has successfully restored the functional synaptic connectivity map of the in vitro-cultured neurons on the integrated neuromorphic circuits12. Further integration of flexible BCIs and neuromorphic circuits could extend natural and artificial intelligence integration for in vivo applications.
Flexible BCIs show promise in a broad range of applications, but there are clear engineering challenges that remain to be tackled. First, further scaling up of the density of the electrodes requires the development of high-performance soft electronic materials compatible with lithographic fabrication (Fig. 4a). While some soft conductors and semiconductors (such as those based on organic materials) offer mechanical compliance, they are not compatible with lithographic fabrication, and their electrical properties such as conductivities, electron and hole mobilities, and sheet resistance are largely limited45. The moderate performance compounded with large parasitic capacitance and sensitivity to ion and water penetration45 also limits their use in high-density neural recording. To offer sufficient performance for signal transmission, amplification and multiplexing, and to maintain cellular-level mechanical properties and feature sizes, nanoscale metal and semiconductors (such as gold and silicon) are still the ideal candidates for the functional layer.
The dielectric constant that determines insulation performance is not affected by the mechanical softness of insulators in the dielectric layers, although the long-term performance, stability and robustness of dielectric organic materials such as elastomers are still limited by the progressive ion penetration from physiological solutions45. In addition, soft materials are prone to swelling in the organic solvent used in multilayer lithographic fabrication. As a result, new soft electronic materials with long-term stability in physiological solutions, high performance and compatibility with multilayer lithographic fabrication are essential for flexible BCIs (Fig. 4a). Many researchers have already studied these issues. In particular, integration of perfluorinated elastomers42,63 with conventional soft electronic materials could substantially increase material longevity in physiological solution and enable the chemical orthogonality to prevent ion and water penetration.
Realistically, high-performance flexible BCIs will most likely be built as inorganic–organic hybrid electronics, where elastomer dielectrics serve as soft dielectric layers and inorganic thin-film interconnects and computational units are employed as functional layers (Fig. 4b). While placing the rigid thin-film components on the mechanically neutral plane could sustain their intactness during mechanical bending, novel multilayer structure design is necessary to keep the overall tissue-level soft structure. It is also important to buttress the adhesion between functional layers, preventing delamination during stretching and bending. Then, new fabrication processes need to be developed to 3D stack multiple inorganic and organic layers into flexible and soft electronics. Several unconventional soft electronics fabrication procedures have already been developed43,64, but their compatibility with conventional lithographic fabrications needs to be tested further.
Device fabrication and integration
How to efficiently fabricate flexible electronics from electronic materials is also an important topic. At the current stage, we envision that lithography will play an important role in the nanofabrication of flexible electronics due to its capability to integrate millions to billions of nanometre-scaled sensors with high resolution, density and uniformity, and its compatibility with mechanically stable single-crystal silicon and metals such as gold and platinum that offer high electronic performance such as high conductivity and low interfacial impedance. Various other fabrication techniques such as printing, laser and moulding have been actively pursued in the field of flexible electronics and hold great promise in terms of cost and time efficiency65,66. However, these techniques need to first demonstrate that they can offer the aforementioned advantages before they can then be considered more applicable in future flexible BCIs.
After fabrication, flexible and soft electronics need to be integrated with ASIC chips for signal processing. Flip-chip bonding has been demonstrated to combine flexible and rigid electronics. However, the input/output resolution and throughput are limited by the bonding process, especially when the density of sensors and the complexity of 3D integrated device structures keep increasing. Future solutions require directly fabricating soft and flexible electronics by high-resolution lithography on the prefabricated ASIC chip to further increase the number and density of individually addressable flexible sensors and stimulators with multiplexing circuits (Fig. 4b). The entire process needs to be compatible with state-of-the-art CMOS circuits and silicon integrated-circuit fabrications. In addition, during the long-term implantation, it is important to prevent detachment at the flexible-to-rigid interface caused by multiple cycles of strain and stress, requiring improvement in the strain engineering at the input/output regions. Alternatively, placing the rigid ASIC chip into the skull and eliminating the wiring by wireless transmission19 can be one solution to improve the longevity of the BCIs.
Rigid brain probes can be directly implanted into the targeted brain region following craniotomy. However, it is challenging to precisely implant the tissue-level flexible brain probes into the brain (Fig. 4c). Several methods have been proposed to address this issue. For example, a neurosurgical robotic sewing machine can automatically implant thousands of flexible electrode threads into the brain19. Tissue-like electronics can be delivered into the targeted brain regions by syringe injection24. The monolithically integrated water-releasable shuttle can further deliver the unfolded mesh electronics across multiple brain regions32. To reach the deeper brain regions without introducing acute mechanical damage, delivery of the stent-electrode array through the endovascular system has been introduced67. Although stent-electrode arrays have successfully been delivered to human cortical brain regions, single-neuron action potential recording is compromised by blood vessel separation. Future implantation techniques that can distribute the sensors across multiple brain regions with minimal invasiveness are critical. For example, one study showed that the implantation in a developing brain may introduce less damage owing to brain self-recovery34.
Despite these efforts, a safe, efficient and reliable implantation method precisely targeting deeper regions of large mammalian brains such as the human brain is still required. At the current stage, flexible electronics integrated with a releasable shuttle that guides the implantation are the most clinically applicable, and how to combine such an implantation system with current flexible electronics is of particular research interest. First, flexible electronics need to be robust and durable enough to withstand potential failure during implantation using materials such as stretchable elastomers45. Second, an automatic system needs to be developed to precisely guide the electronics and controllably withdraw the shuttle without causing significant probe drift. Third, advanced imaging techniques need to be incorporated to track and confirm the location of the implanted flexible electronics in the deep brain regions. In the future, different combinations of the implantation techniques could be designed and tailored based on the specific BCI application.
Hardware and software
Issues in software and hardware also need to be addressed to obtain scalable, high-density and high-throughput flexible BCIs (Fig. 4d). The increase of neural recording in both temporal and spatial scales will generate an ever-increasing amount of data, which requires the development of advanced on-chip and online signal processing. In addition, current wireless transmission has limited throughput, thus also requiring efficient online and on-chip data compression and dimension reduction algorithms for remotely controlled, wirelessly communicated BCIs68,69,70. Examples of potential signal processing techniques to implement include the online spike detection and sorting that can rapidly decrease the data throughput by extracting only the action potential times of individual neurons and the online dimension reduction that provides the most critical information from population activities by eliminating noise.
However, on-chip processing will potentially cause high power consumption. Therefore, low-power-demanding signal processing devices should be considered to achieve fully implantable electronics. We believe the following directions could address this challenge. Advanced ASIC chips with a <10 nm CMOS fabrication process could be designed to include on-site power management modules for lower consumption. Relaxing requirements on signal quality, frequency band and noise level by modifying the design of amplifier, analogue-to-digital converter and transmitter68 could also reduce power consumption. Moreover, less power-demanding electronics such as computing-in-memory devices, neuromorphic memristive electronics, and analogue-computing-based chips that bypass the von Neumann bottleneck could be integrated for on-chip data analysis62,71.
Flexible electronics offer a number of advantages in the development of BCIs: mechanical and topological similarities between neural tissue and electronics that lead to minimal damage and enable an interlocked network for long-term tracking of the same cells, stretchability that improves the temporal coverage to the entire lifespan of the BCI users, and low-modulus soft materials that allow for high-density electronics design with scalability, multimodality and multifunctionality. Ultimately, however, fundamental innovations in BCIs require a better understanding of brain structures and functions. The diversity of the neurons needs to be precisely mapped according to their gene expression, connectivity and electrophysiology under environmental inputs and behaviours. Technologies for brain transcriptomics72 and connectomics mapping73 can now sample millions of neurons in intact tissues, but electrical mapping is still limited to a few thousand neurons.
We believe that flexible electronics can fundamentally change how the neurons are electrically interfaced and thus offer the necessary scalability for large-scale long-term stable neural electrical recording in the near future. This, however, will require the collective efforts of both academia (by solving fundamental challenges) and industry (by building scalable flexible electronics fabrication pipelines). Furthermore, flexible BCIs should not be limited by the electrical modality. While conventional optics are limited by their bulky and rigid structures, the emergence of 2D optics such as metastructures allows for CMOS-compatible fabrication processes. The further development of metastructures74 into flexible electronics could lead to scalable flexible and soft optoelectronics for BCIs. Finally, we recognize the possible hesitancy of the general public when a soft BCI does reach general applicability. Therefore, we emphasize the necessity of the pursuit of softer and smarter materials, the miniaturization of such electronics, the development of more appealing implantation procedures and the establishment of a more comfortable user experience.
Thakor, N. V. Translating the brain–machine interface. Sci. Transl. Med. 5, 210ps17 (2013).
Stanley, G. B. Reading and writing the neural code. Nat. Neurosci. 16, 259–263 (2013).
Hochberg, L. R. et al. Reach and grasp by people with tetraplegia using a neurally controlled robotic arm. Nature 485, 372–375 (2012).
Patel, S. R. & Lieber, C. M. Precision electronic medicine in the brain. Nat. Biotechnol. 37, 1007–1012 (2019).
Makin, T. R., Micera, S. & Miller, L. E. Neurocognitive and motor-control challenges for the realization of bionic augmentation. Nat. Biomed. Eng. https://doi.org/10.1038/s41551-022-00930-1 (2022).
Nicolas-Alonso, L. F. & Gomez-Gil, J. Brain computer interfaces, a review. Sensors 12, 1211–1279 (2012).
Pochay, P., Wise, K. D., Allard, L. F. & Rutledge, L. T. A multichannel depth probe fabricated using electron-beam lithography. IEEE Trans. Biomed. Eng. BME-26, 199–206 (1979).
Schwartz, A. B. Cortical neural prosthetics. Annu. Rev. Neurosci. 27, 487–507 (2004).
Flesher, S. N. et al. A brain–computer interface that evokes tactile sensations improves robotic arm control. Science 372, 831–836 (2021).
Willett, F. R., Avansino, D. T., Hochberg, L. R., Henderson, J. M. & Shenoy, K. V. High-performance brain-to-text communication via handwriting. Nature 593, 249–254 (2021).
Kleinfeld, D. et al. Can one concurrently record electrical spikes from every neuron in a mammalian brain? Neuron 103, 1005–1015 (2019).
Abbott, J. et al. A nanoelectrode array for obtaining intracellular recordings from thousands of connected neurons. Nat. Biomed. Eng. 4, 232–241 (2020).
Jun, J. J. et al. Fully integrated silicon probes for high-density recording of neural activity. Nature 551, 232–236 (2017).
Steinmetz, N. A. et al. Neuropixels 2.0: a miniaturized high-density probe for stable, long-term brain recordings. Science 372, eabf4588 (2021).
Paulk, A. C. et al. Large-scale neural recordings with single neuron resolution using Neuropixels probes in human cortex. Nat. Neurosci. 25, 252–263 (2022).
Obaid, A. et al. Massively parallel microwire arrays integrated with CMOS chips for neural recording. Sci. Adv. 6, eaay2789 (2020).
Zeng, H. & Sanes, J. R. Neuronal cell-type classification: challenges, opportunities and the path forward. Nat. Rev. Neurosci. 18, 530–546 (2017).
Chung, J. E. et al. High-density, long-lasting, and multi-region electrophysiological recordings using polymer electrode arrays. Neuron 101, 21–31.e5 (2019).
Musk, E. & Neuralink An integrated brain-machine interface platform with thousands of channels. J. Med. Internet Res. 21, e16194 (2019).
Xie, C. et al. Three-dimensional macroporous nanoelectronic networks as minimally invasive brain probes. Nat. Mater. 14, 1286–1292 (2015).
Fu, T.-M. et al. Stable long-term chronic brain mapping at the single-neuron level. Nat. Methods 13, 875–882 (2016).
Yang, X. et al. Bioinspired neuron-like electronics. Nat. Mater. 18, 510–517 (2019).
Lacour, S. P., Courtine, G. & Guck, J. Materials and technologies for soft implantable neuroprostheses. Nat. Rev. Mater. 1, 16063 (2016).
Liu, J. et al. Syringe-injectable electronics. Nat. Nanotechnol. 10, 629–636 (2015).
Kim, T. et al. Injectable, cellular-scale optoelectronics with applications for wireless optogenetics. Science 340, 211–216 (2013).
Degenhart, A. D. et al. Stabilization of a brain–computer interface via the alignment of low-dimensional spaces of neural activity. Nat. Biomed. Eng. 4, 672–685 (2020).
Salatino, J. W., Ludwig, K. A., Kozai, T. D. Y. & Purcell, E. K. Glial responses to implanted electrodes in the brain. Nat. Biomed. Eng. 1, 862–877 (2017).
Khodagholy, D. et al. NeuroGrid: recording action potentials from the surface of the brain. Nat. Neurosci. 18, 310–315 (2015).
Viventi, J. et al. Flexible, foldable, actively multiplexed, high-density electrode array for mapping brain activity in vivo. Nat. Neurosci. 14, 1599–1605 (2011).
Minev, I. R. et al. Electronic dura mater for long-term multimodal neural interfaces. Science 347, 159–163 (2015).
Tian, B. et al. Macroporous nanowire nanoelectronic scaffolds for synthetic tissues. Nat. Mater. 11, 986–994 (2012).
Zhao, S. et al. Tracking neural activity from the same cells during the entire adult life of mice. Preprint at bioRxiv https://doi.org/10.1101/2021.10.29.466524 (2021).
Rogers, J. A., Someya, T. & Huang, Y. Materials and mechanics for stretchable electronics. Science 327, 1603–1607 (2010).
Le Floch, P. et al. Stretchable mesh nanoelectronics for 3D single-cell chronic electrophysiology from developing brain organoids. Adv. Mater. 34, 2106829 (2022).
Tringides, C. M. et al. Viscoelastic surface electrode arrays to interface with viscoelastic tissues. Nat. Nanotechnol. 16, 1019–1029 (2021).
Liu, Y. et al. Morphing electronics enable neuromodulation in growing tissue. Nat. Biotechnol. 38, 1031–1036 (2020).
Guan, S. et al. Elastocapillary self-assembled neurotassels for stable neural activity recordings. Sci. Adv. 5, eaav2842 (2019).
Zhao, Z. et al. Ultraflexible electrode arrays for months-long high-density electrophysiological mapping of thousands of neurons in rodents. Nat. Biomed. Eng. https://doi.org/10.1038/s41551-022-00941-y (2022).
Fu, T.-M., Hong, G., Viveros, R. D., Zhou, T. & Lieber, C. M. Highly scalable multichannel mesh electronics for stable chronic brain electrophysiology. Proc. Natl Acad. Sci. USA 114, E10046–E10055 (2017).
Chiang, C.-H. et al. Development of a neural interface for high-definition, long-term recording in rodents and nonhuman primates. Sci. Transl. Med. 12, eaay4682 (2020).
Luan, L. et al. Ultraflexible nanoelectronic probes form reliable, glial scar-free neural integration. Sci. Adv. 3, e1601966 (2017).
Liu, Y. et al. Soft and elastic hydrogel-based microelectronics for localized low-voltage neuromodulation. Nat. Biomed. Eng. 3, 58–68 (2019).
Zheng, Y.-Q. et al. Monolithic optical microlithography of high-density elastic circuits. Science 373, 88–94 (2021).
Wang, S. et al. Skin electronics from scalable fabrication of an intrinsically stretchable transistor array. Nature 555, 83–88 (2018).
Le Floch, P. et al. Fundamental limits to the electrochemical impedance stability of dielectric elastomers in bioelectronics. Nano Lett. 20, 224–233 (2020).
Liu, J. et al. Genetically targeted chemical assembly of functional materials in living cells, tissues, and animals. Science 367, 1372–1376 (2020).
Li, Q. et al. In situ electro-sequencing in three-dimensional tissues. Preprint at bioRxiv https://doi.org/10.1101/2021.04.22.440941 (2021).
Marin, C. & Fernández, E. Biocompatibility of intracortical microelectrodes: current status and future prospects. Front. Neuroeng. 3, 8 (2010).
Kozai, T. D. Y. et al. Ultrasmall implantable composite microelectrodes with bioactive surfaces for chronic neural interfaces. Nat. Mater. 11, 1065–1073 (2012).
Cui, Y. et al. A stretchable and transparent electrode based on PEGylated silk fibroin for in vivo dual-modal neural-vascular activity probing. Adv. Mater. 33, 2100221 (2021).
Adewole, D. O. et al. Development of optically controlled ‘living electrodes’ with long-projecting axon tracts for a synaptic brain–machine interface. Sci. Adv. 7, eaay5347 (2021).
Won, C. et al. Mechanically tissue-like and highly conductive Au nanoparticles embedded elastomeric fiber electrodes of brain–machine interfaces for chronic in vivo brain neural recording. Adv. Funct. Mater. 32, 2205145 (2022).
Natraj, N., Silversmith, D. B., Chang, E. F. & Ganguly, K. Compartmentalized dynamics within a common multi-area mesoscale manifold represent a repertoire of human hand movements. Neuron 110, 154–174.e12 (2022).
Schoonover, C. E., Ohashi, S. N., Axel, R. & Fink, A. J. P. Representational drift in primary olfactory cortex. Nature 594, 541–546 (2021).
Tang, X. et al. Multi-task learning for single-cell multi-modality biology. Preprint at bioRxiv https://doi.org/10.1101/2022.06.03.494730 (2022).
Anumanchipalli, G. K., Chartier, J. & Chang, E. F. Speech synthesis from neural decoding of spoken sentences. Nature 568, 493–498 (2019).
Jazayeri, M. & Ostojic, S. Interpreting neural computations by examining intrinsic and embedding dimensionality of neural activity. Curr. Opin. Neurobiol. 70, 113–120 (2021).
Panzeri, S., Moroni, M., Safaai, H. & Harvey, C. D. The structures and functions of correlations in neural population codes. Nat. Rev. Neurosci. 23, 551–567 (2022).
Chaudhuri, R., Gerçek, B., Pandey, B., Peyrache, A. & Fiete, I. The intrinsic attractor manifold and population dynamics of a canonical cognitive circuit across waking and sleep. Nat. Neurosci. 22, 1512–1520 (2019).
Yamins, D. L. K. et al. Performance-optimized hierarchical models predict neural responses in higher visual cortex. Proc. Natl Acad. Sci. USA 111, 8619–8624 (2014).
Yang, G. R., Joglekar, M. R., Song, H. F., Newsome, W. T. & Wang, X.-J. Task representations in neural networks trained to perform many cognitive tasks. Nat. Neurosci. 22, 297–306 (2019).
Zidan, M. A., Strachan, J. P. & Lu, W. D. The future of electronics based on memristive systems. Nat. Electron. 1, 22–29 (2018).
Liu, J. et al. Fully stretchable active-matrix organic light-emitting electrochemical cell array. Nat. Commun. 11, 3362 (2020).
Jiang, Y. et al. Topological supramolecular network enabled high-conductivity, stretchable organic bioelectronics. Science 375, 1411–1417 (2022).
Dong, R. et al. Printed stretchable liquid metal electrode arrays for in vivo neural recording. Small 17, 2006612 (2021).
Afanasenkau, D. et al. Rapid prototyping of soft bioelectronic implants for use as neuromuscular interfaces. Nat. Biomed. Eng. 4, 1010–1022 (2020).
Opie, N. L. et al. Focal stimulation of the sheep motor cortex with a chronically implanted minimally invasive electrode array mounted on an endovascular stent. Nat. Biomed. Eng. 2, 907–914 (2018).
Even-Chen, N. et al. Power-saving design opportunities for wireless intracortical brain–computer interfaces. Nat. Biomed. Eng. 4, 984–996 (2020).
Jeong, J.-W. et al. Wireless optofluidic systems for programmable in vivo pharmacology and optogenetics. Cell 162, 662–674 (2015).
Park, S. I. et al. Soft, stretchable, fully implantable miniaturized optoelectronic systems for wireless optogenetics. Nat. Biotechnol. 33, 1280–1286 (2015).
Huo, Q. et al. A computing-in-memory macro based on three-dimensional resistive random-access memory. Nat. Electron. 5, 469–477 (2022).
Lein, E., Borm, L. E. & Linnarsson, S. The promise of spatial transcriptomics for neuroscience in the era of molecular cell typing. Science 358, 64–69 (2017).
Lichtman, J. W., Pfister, H. & Shavit, N. The big data challenges of connectomics. Nat. Neurosci. 17, 1448–1454 (2014).
Dorrah, A. H. & Capasso, F. Tunable structured light with flat optics. Science 376, eabi6860 (2022).
This work is supported by the Harvard School of Engineering and Applied Sciences Faculty Start Up Fund and Harvard University Dean’s Competitive Fund for Promising Scholarship. Elements in Figs. 1–4 created with BioRender.com.
J.L. is co-founder of Axoft, Inc.
Peer review information
Nature Electronics thanks Xiaojie Duan and Woon-Hong Yeo for their contribution to the peer review of this work.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Tang, X., Shen, H., Zhao, S. et al. Flexible brain–computer interfaces. Nat Electron 6, 109–118 (2023). https://doi.org/10.1038/s41928-022-00913-9