The microwire electrode is a long-established microelectrode used for electrophysiological studies, and its use dates back to the 1940s (Sscwartz, 2004). The core of a microwire electrode is a conductive metal wire, which is surrounded by an insulating material (such as polymer, ceramic, or glass) that is resistant to electrolyte solutions. (Figure 3A). Only the tiny metal tip is exposed. The glass-coated tungsten microwire electrodes can be fabricated with simple tools and their tips can reach the same size as the soma of the cortical neuron. Using this electrode, the extracellular neural activity signals close to the firing neurons have been recorded for the first time with high signal-to-noise ratios. The neural recording technology based on this electrode has made a lot of important contributions in the history of neuroscience research, with pioneering studies in brain regions such as the primary visual cortex, hippocampus, inferior temporal cortex, and the neural correlations of perceptual sensation and movement control (Hubel and Wiesel, 1959; O'Keefe and Dostrovsky, 1971; Desimone et al., 1984; Newsome and Pare, 1988). The chronic recording experiments have shown that microwire electrodes less than 20 µm in diameter significantly reduce the foreign body response to the inserted microwires.

The tetrode is a small four-channel bundle array of electrodes consisting of four microwires. The tetrode allows the simultaneous detection of neuronal signals from the same source by four electrode sites from slightly different locations (Figure 3B). The diameter of the metal contact of each electrode in this small array is typically less than 30 μm (Mcnaughton et al., 1983). The main advantage of tetrodes over single-channel electrodes is that the extracellular potentials detected by the four electrodes differ due to the different distances between each electrode and the neuron (Hubel, 1957). The classification of extracellular potential signals using a clustering approach enables the separation of the spiking events emitted by the adjacent neurons (Choi et al., 2018).

The single and multiwire electrodes are basically handmade, which are expensive to manufacture manually and cannot be mass produced (still can exceed 100 channels, over 1 mm coverage). These microelectrodes are often considered an “art” due to the low precision of processing and the low reproducibility of metrics, such as impedance and shape. However, because they can be fabricated by hand, this technique is widely used in many neurophysiology laboratories.

The Utah array is a commercially available electrode array for intracortical implantation, which adopts an out-of-plane process. The Utah array (medium scale) consisted of approximately 100 silicon needle shaped electrodes spaced 400 μm with exposed tips of diameter 10–30 μm (Figure 3C). The array is fabricated through MEMS technology with chemical micromachining, metal deposition, and polymer encapsulation (Campbell et al., 1990). Neuroscientists needed long-term stable chronic neural recording in the brain and have come up with the idea of enclosing the entire electrode array inside the skull. The Utah array is designed to solve the problem of closing the skull after implantation of a planar electrode array with the support substrate and connecting leads. Because the Utah array has the advantages of high channel count, balanced cortical coverage to electrode density, and chronic recording feature, it is mainly used in large animals, especially non-human primates. However, the tip of the silicon electrode is short (typically no more than 1.5 mm), and it is only suitable for recording neural signals from the cortex. Utah electrodes are similar to microwire electrodes in that electrode contact is only on the tip of each shaft. Inserting all nearly 100 probes of a dense array into the cortex at the same time is difficult, requires the use of a special micro air hammer to implant them (Normann, 2007; Wark et al., 2013; Choi et al., 2018) (Figure 5D).

The Utah array and its recording system have been approved by the U.S. Food and Drug Administration (FDA) for clinical investigations of intracortical brain-machine interface technology (iBMI). The Utah arrays are currently used primarily in clinical trials of motor brain-machine interfaces, which replace and restore the injured motor function in paralyzed patients by transmitting motor-related neural signals from the brain to external effectors (Figure 3D). The main approach is to implant one or more (two) Utah arrays in the region of the cerebral cortex responsible for limb movements in the patient’s brain. In non-human primate experiments, the simultaneous implantation of multiple Utah arrays with a total of more than 500 channels in the visual cortex of the same macaque monkey has been achieved so far (Collinger et al., 2013).

Another commercially available microelectrode array, the Michigan probe, is manufactured using an in-plane machining process. The “Michigan” probe was the first silicon-based neural electrode array fabricated in 1970. It is manufactured using micro-electro-mechanical system (MEMS) that includes techniques such as depositing metals on silicon and etching (Wise et al., 1970). A Michigan electrode is a flat rod with multiple recording contacts lined up along the wide side of the flat rod (small scale generally), and the contact surface area is typically between 100–400 μm². The advantage of this manufacturing process is that the spacing between electrode contacts can be precisely controlled and the diameter of the contacts can be as small as 2 μm, while the length of the entire probe can vary from a few millimeters to centimeters, making the Michigan electrode more suitable for recording brain tissue structures from the longitudinal direction.

Polytrodes probe is a variant of the Michigan probe designed for intensively extracellular multi-unit recording in a cluster of adjacent neurons (Figure 4A). The shank of the polytrode probe is arranged with multiple closely spaced recording sites, equivalent to an enhanced version of the tetrode in single-unit isolation function, and can record up to three times as many neurons as the number of electrode contacts. The high-precision lithographic process ensures uniformity in the size of each recording site, and it is easier to customize the shape of the probe (single or multiple shanks) and the location of the recording sites on the shank according to the structure of the brain region to be recorded (Blanche et al., 2005).

Multi-shank Michigan-type probes form a two-dimensional (2D) comb structure, and 2D probe combs can be further assembled into three-dimensional (3D) probe arrays. Such arrays allow dense volumetric electrophysiological recordings of neuronal networks in one brain region (Wise, 2005) (Figure 4B). One-dimensional narrow bar multi-electrode probes are expanded into high electrode density two-dimensional electrode arrays, which in turn are stacked into three-dimensional electrode matrices (often expanded to medium scale). This is a modularization solution with high scalability. The back end of the electrode matrix is connected to the integrated circuits that perform signal conditioning, multiplexing, and digitization, via high-density flexible wires. These wires effectively prevent passive electrodes from active devices in terms of mechanical, electronic, and thermal effects. In chronic recording experiments, the electrode array floating in the brain tissue and the upper side of the implanted brain tissue covered by dura and skull, are good for long-term recordings. The base of the silicon probe array for chronic recording which is not implanted into the brain tissue does impede the closing of the skull, so it has to be designed to be as short as possible. Overall, it appears that the shape of the electrode matrix for chronic recording is a hybrid style of the Michigan probe and the Utah array.

The polymer probe is a flexible polytrode probe. It is operationally flexible as a microwire electrode, while the density of recording contacts can approach that of the Michigan probe. Polymer probes are highly neuro-compatible, not only from the high biocompatibility of the polymer material but also from the ability to move with the brain tissue, thus reducing shear damage to the tissue. One improvement of the polymer probe is to cut off part of the area around the electrode contact and make the strip shape polymer probe into an open mesh structure. This design further enhanced the mechanical flexibility of the polymer probe and reduces electrical crosstalk between channel lines by disturbing the parallel interconnect wires. However, because the polymer probes tend to be soft and cannot be directly stabbed into brain tissue, auxiliary methods, such as reinforcing with water-soluble reinforcers (such as PEG) (Figure 5A), or bringing electrodes into the brain tissue with fine metal needles as the sewing machine does (Figure 5B), mesh electrode arrays can be made not only as penetrating probes but also as ECoG electrodes that adhere softly to the cortical surface without the stabbing process (Figures 5A–C).

Neurogrid is a thin-film electrode array with 4-μm parylene C as the insulating substrate. It has electrodes with poly (3,4-ethylenedioxythiophene) doped with poly (styrenesulfonate) (PEDOT:PSS) as a gate in organic electrochemical transistors and is capable of recording the action potential of neurons from the cortical surface (Khodagholy et al., 2015). The samples of this electrode array have 256 recording sites with a surface area of 10 × 10 μm² and an interelectrode spacing of 30 μm, matching the average size of neuronal bodies and neuronal density (Figure 6A). The interface material at the electrode contacts was made of PEDOT: PSS, whose high mixed electronic/ionic conductivity and high ionic mobility significantly decrease the electrochemical impedance mismatch between the brain tissue and the metal electrode (Owens and Malliaras, 2010; Stavrinidou et al., 2013). The encapsulation with parylene C has prepared a flexible film with a thickness of only 4 μm, allowing it to be tightly adhered to the surface of the irregularly shaped cerebral cortex. In addition, the neurogrid can be folded and inserted into the cerebral sulcus where conventional subdural ECoG electrodes for clinical use cannot be inserted (Figure 6B). Since such electrodes do not penetrate into the brain tissue, there is no need to worry about severe brain damage; they can be easily prepared for medium scale. However, it is also seen that the electrode contacts only cover a small area of the film attached to the cortical surface, and most of the area is occupied by the interconnect wires. The low coverage ratio of the practical records hindered the design of large-scale recording.

Under the premise of ensuring high recording electrode density, the electrical activity of a large number of neurons can be recorded across multiple brain regions simultaneously by increasing the number of recording electrodes and expanding the coverage area of brain tissue. This kind of method can provide crucial neural information for understanding the self-organized processes of neural networks and the encoding principle of specific cognitive behaviors by neuronal ensembles. Thus, large-scale recording is now an urgently needed neural technology. The following recording devices are mainly the engineering innovation of the respective basic small and medium-scale recording devices, which break through the bottleneck of the recording scale.

Compared to the conventional silicon-based Michigan electrodes, the feature of this design is that complementary metal oxide semiconductor (CMOS) technology is used to realize the integration of recording electrodes and multi-channel multiplexed circuits into the same silicon probe, and the silicon probe is also integrated with the amplification, filtering, and digital signal preprocessing circuit into the same processed silicon wafer (Figure 7A).

Version 1.0 of Neuropixel is fabricated in a 130 nm CMOS process. 960 recording electrodes and the precision electronics (or: complex electronic circuits) with multiplexed addressing and local amplification functions are integrated on a 10 mm long, 70 × 20 μm cross-section rigid silicon probe. The probe shank is connected with the base via 384 connecting wires. A configurable 384-channel integrated circuit on the 5 × 9 mm² probe base is responsible for dual-band recording (30 KHz pulse frequency and 2.5 KHz LFP frequency per channel) and to send data over 20 MB bandwidth to the acquisition board through a digital port. Although there are 960 recording sites that can be selected, only 384 signal channels can record simultaneously on neuropixel 1.0 (Jun et al., 2017) (Figures 7B,C).Version 2.0 of Neuropixel has four shanks in each probe, a total of 5,120 recording sites, which can intensively record the signals in the 1 × 10 mm plane (Figure 7D). Neuropixel 2.0 is a dual probe unit structure. So there are 10,240 recording sites and the number of simultaneously recordable channels reaches 768 but the volume of the probe shrinks to one-third of the neuropixels 1.0 (Steinmetz et al., 2021). Under the premise of ensuring thousands of recording sites, the neuropixel probe compresses the cross-sectional area and wiring amount of the probe through multiplexing and further simplifies the connection through the integration of the preprocessing chip and the probe, which greatly reduces the overall volume of the neural interface device (Putzeys et al., 2019). When recording the neural activities, this design can effectively reduce the restraint on animals.

As the longest-used microwire electrode, it can be made very slim and stiff. It not only has low insertion damage to the brain tissue but also has good recording quality. For the application of microwire electrodes in large-scale recording, it is necessary to solve the problem of connecting the electrode array composed of a large number of microwires to the amplifier array (Mal et al., 2003). The solution proposed by “Argo System” was to use a sacrificial coating to thicken the wire and then tie them into a bundle, spaced just perpendicularly coupling to the plane of the large CMOS amplifier array (Figures 8A,B).

After the butt-joint of the microwire bundle array to the amplifier array surface of the backplane, a recording device similar to the Utah array for penetrating cortex is formed. The significant progress of the Argo over the Utah array is the overall data throughput. The prototype of the Argo system has the ability to continuously record 65,536 channels at 32 kHz and 12 bits of resolution.

The prototype validates the ability to record 791 single-unit APs in the rat cortex with an array of up to 1,300 penetrating microwires. Stimulus-evoked LFPs are recorded by microwire arrays with more than 30,000 channels attached to the surface of the sheep’s auditory cortex. (Sahasrabuddhe et al., 2020). This method of grafting microwires array to the CMOS process products suitable for mass production provides a large-scale neural interface platform with further expansion.

Neuralink has specifically developed a robotic sewing machine that automatically implants each thread to stab into the brain tissue with the help of a fine tungsten needle hook (Musk and Neuralink, 2019; Tooker et al., 2012a; Tooker et al., 2012b) (Figures 9A,B). The sewing machine has the ability to locate specific cortical areas with machine vision and to automatically avoid the surface blood vessels for safe electrode implantation (Hanson et al., 2019) (Figure 9C). The neuralink provides a system-level solution, including an array of flexible polymer probes integrated with customized low-power electronics for signal acquisition, and implantation machinery that can automatically implant electrodes into the cortex (Figure 9D). The current wired transmission version (Figure 9D) of the system is a state-of-the-art large-scale scientific research platform for in vivo neuroelectrophysiology. The wireless version (Figure 9E) of the system is the first fully implanted front-end for preclinical investigation.

The Neural Matrix (Figure 10A) is a kind of thousand-channel-level array of μECoG electrodes. It is to greatly reduce the amount of wiring by integrating the powerline time-division multiplexing electronic device directly into the sensor to increase the ratio of electrode contact coverage to the total area of the array, and ensure high electrode density. In a neural matrix test prototype, the neural signals were recorded from a high-density electrode array with nearly 100 wires containing over 1,000 recording sites by setting up two layers of metal interconnect lines plus embedded multiplexers. The electrode array is in a 28-column by 36-row mode with a coverage of 9 × 9.24 mm² (Figure 10B). A total of 1,008 wires are needed for the traditional electrode array but the total number of wires required for the neural matrix is only 92 (2 times columns then plus rows), saving more than 90% of the wiring area. Because cortical surface electrodes require a relatively low sampling rate (generally do not have to exceed 1 KHz), all electrode sites within the neural matrix can be sampled in a two-dimensional scanning cycle. While each channel of the neuropixels requires a much higher sampling rate, it adopts a multiplexing method that is to select recording sites with no more than the total number of recordable channels before each recording session (Chiang et al., 2020).

Recently, another thousand-channel-level uECoG electrode array, called PtNRGrids, is developed. It is a 6.6 um thin-film electrode array with 30 um diameter electrode contacts and low impedance by platinum nanorod modification (Figures 11A–C).The spacing of recording sites for this array can be configured between 150 μm and 1.8 mm to form a coverage area ranging from 5 × 5 mm² (1,024 grids) for rodents to 8 × 8 cm² (2048 grids) for the human brain. With this high spatial resolution electrode array, the investigator observed the epilepsy discharge dynamics and spatial diffusion process in patients with epilepsy surgery (Tchoe et al., 2021). Compared to the high electrode density design of the neural matrix, the PtNRGrids design passed a large number of interconnecting wires through the wide electrode spacing and achieved the large-scale recording with that the electrode contact coverage occupies most of the total area. Since the spatial resolution of the ECoG signal does not exceed the LFP signal, the sampling density of hundreds of micron spacing does not lose a lot of brain information. For the human brain, due to the larger brain surface, reducing the spatial sampling density while covering a larger brain area may be more beneficial to obtaining richer brain information. So this large-scale recording compromise with the deployment of lower electrode density is also acceptable.

The large-scale recording electrode arrays mentioned above are integrated processing devices, with the location of each electrode contact determined and implanted into the brain as an ensemble. The idea of the next generation neural interface design has been proposed: the electrode implantation position can be set independently and flexibly, the scale can reach thousands, and it should have the bidirectional function of recording and stimulation (Lee et al., 2021). A proof-of-concept experiment of the distributed sensor system (Figure 12A) is completed on the cortex by microminiaturization of the device “neurograin” as a network node but maintaining high signal fidelity. The neurograin system is composed of implantable, submillimeter, individually addressable (Huang et al., 2019), and microchip sets (Figure 12B). Each neurograin is an independent packed module of 500 μm × 500µm × 35 µm and wireless powered by inductive coupling technology for near-field transmission (Lee et al., 2018). Neurograin microdevices do not need tether and can be implanted in specific brain regions individually or collectively for reading and writing neural information, with good prospects for clinical application.