Flexible organic transistors for neural activity recording

The profound mysteries of the brain have attracted considerable research interest to seek the basic mechanisms of sensory input collection and information processing. Containing 86 × 10⁹ neurons that form a highly complex interwoven neural network, the brain allows dynamic tasks from sensation to thought to action. Recording of the neural activity serves as the foundation for understanding the brain's working mechanism and treating the neurodegenerative or psychiatric conditions caused by nervous system disorders. Hemodynamic-based approaches, including functional magnetic resonance imaging (fMRI) and functional near-infrared spectroscopy (fNIRS), are widely applied in clinics for neuroimaging and brain metabolism measurement. High spatial and low temporal resolutions are typical characteristics of the technologies mentioned above.¹ 

Neural activity occurs mainly in the electric form through changes in their membrane potential caused by the ion flux through ion channels across the membrane.² They communicate via action potentials or spikes, and the aggregate synaptic and spiking activity are defined as local field potential (LFP). In other words, LFPs are generated by the spatiotemporal summation of current sources and sinks caused by the ion flux in a given brain volume.³ Important data about brain status is coded in different time or frequency domain components and their interactions. Electrophysiological recording method constitutes a crucial complementary strategy to imaging technologies, with the capability of decoding “brain waves” and revealing electrical activity. Noninvasive electroencephalogram (EEG) was innovated in 1929 by H. Berger, the German psychiatrist, a historical breakthrough for neurologic and psychiatric diagnosis back then. Today's analysis of EEG signals has moved from a simple visual inspection of the waveform, i.e., amplitude and frequency modulations to a more in-depth exploration of signals' temporal and spatial characteristics, thanks to the rapid development of information technologies.⁴ On the other hand, using electrodes placed on the scalp, the signal attenuation due to the resistive skull and the relatively low spatial resolution impose challenges in inferring the exact location of the brain areas generating the target neuronal activity. Electrocorticography (ECoG)⁵ provides an interface with higher resolution and fidelity, greater noise resistance, and a broader frequency band using subdural grid electrodes on the cortical surface directly. EEG and ECoG allow for capturing relatively low-frequency signals, i.e., <100 and <200 Hz, respectively. Furthermore, stereotactic depth electrodes (silicon probes, Utah and Michigan arrays, and tetrodes)⁶ can be localized with submillimeter anatomical precision, and their tissue-penetrating configuration demonstrates the ability to record action potentials (around 1 kHz) in concert with LFP (<200 Hz) and to reveal single-unit activities.⁷ 

Recording tools with high spatiotemporal resolution and minimized invasiveness are highly desired for the precise localization from populations of neurons over a large area.^8,9 Arrays of stereotactic depth electrodes and ECoG probes have been used in the clinic to localize epileptogenic zones and provide brain mapping before the surgery, assistive for the diagnosis and treatment of neurological disorder-related diseases.^10–12 Despite their application potentials, conventional microelectrode arrays (MEAs) still have limitations in their signal processing capability and rigid form factor. In particular, as the electrode size decreases, the interfacial impedances often increase, leading to increased thermal noise and reduced signal-to-noise ratio (SNR).¹³ This suggests a trade-off between the density of passive electrodes and the signal quality. State-of-the-art passive electrode arrays allow measuring extracellular action potentials with 60 μm resolution.¹⁴ On the other hand, as neurons and brain networks generate small electric potentials which are challenging to extract from noise, a high SNR is always a must to realize high precision activity recording. In this regard, a built-in pre-amplification/processing system is highly demanded for reaching high SNR.

Transistor-based sensors capable of performing in situ amplification of the recorded signals have inherent advantages. In addition, their working channel can be miniaturized to the nanometer scale, allowing integration into high-density arrays. With rational design, transistors are expected to simultaneously provide a more excellent spatial resolution with sufficiently high SNR. Conventional silicon (Si)-based field-effect transistors (FETs) have been integrated into in vivo probes to address hundreds of electrodes simultaneously.^9,15 However, Si lacks biocompatibility, and its mechanical mismatch with tissues precludes direct contact with the brain for a long-term application. In addition, rigid encapsulation is required to protect Si-based transistors in physiologic environments. An optimal neural interface device is expected to be minimally invasive yet effective in precisely localizing relevant physiological or pathological activities. In the brain, glial cells (microglia and astrocytes) tend to migrate to a foreign body.⁶ It is known that implanted materials of lower moduli incur less immunological rejection.¹⁶ Therefore, approaches must be taken to minimize the causes and exacerbations of the neuroinflammatory response.

Organic bioelectronics is a vibrant research field blooming these years quickly, with considerable attention focusing on interfacing flexible organic electronics with the biological world.^17–20 The tunable mechanical flexibility and biocompatibility are the top advantages of organic electronics over their inorganic counterpart devices. With a reduced risk of implant rejection and minimizing adverse events, organic transistors have proven to allow reliable in vivo recording. The earliest organic transistors date back to the 1980s when the semiconducting channel was based on the thin film of conjugated polythiophene, exhibiting a hole mobility of around $10−5 cm2 V−1 s−1$⁠.²¹ This has triggered considerable research inputs in developing novel conjugated materials and engineering the device interfaces.^22–25 Today, documented mobilities in organic field-effect transistors (OFETs) have exceeded 10 $cm2 V−1 s−1$⁠,^26–30 beyond that of the amorphous Si transistors and gradually approaching the polycrystalline Si-based technologies.³¹ Simultaneously, organic transistors possess great feasibility in their structural design and low-temperature fabrication processing, fully compatible with the ultrathin plastic substrates, leading to variable form factors that are desired for wearable and implantable applications.^32–34 Herein, we provide a timely summary of the design principle and working mechanism of flexible and biocompatible organic transistors that demonstrate superior capabilities in neural activity recording in vitro and in vivo.

Transistors typically have a lateral-channel with a three-terminal configuration and work as amplifiers or electronic switches. The function of on-site signal amplification is especially preferred for electrophysiological sensing to enhance the SNR before signal transmission effectively. Classified by the device architecture, in particular, their working mechanism, transistors are divided into two major categories: dielectric-gate transistors and electrolyte-gated transistors. The latter category includes electrochemical transistors as well. The structure and working mechanisms of the three kinds of organic transistors, i.e., organic field-effect transistors (OFETs), electrolyte-gate organic field-effect transistors (EGOFETs), and organic electrochemical transistors (OECTs), are depicted in Fig. 1.

In OFETs, the gate electrode is in direct contact with a dielectric that is interfaced with a thin layer of an organic semiconductor. The metallic source and drain electrodes are electrically connected through the organic semiconductor layer, i.e., the active channel [Fig. 1(a), p-channel in this case), and are meant to inject and collect positive charge carriers. When a voltage is applied between the gate and the channel (V_GS), the organic semiconducting channel is doped and polarized charges form two plates of a parallel plate capacitor. The gate voltage thus modulates the concentration of charge carriers in the channel. The amount of induced charge carriers (Q) in the channel is a product of the dielectric capacitance (C_i) and the gate voltage (V_GS), $Q=CiVG$⁠. When V_GS exceeds a specific value, the so-called threshold voltage (V_th), there are excessive carriers in the channel in addition to the trapped carriers. As illustrated in Fig. 1(b), a few positive charges reside in the organic semiconductor layer due to the trace amount of p-type dopants in not fully ordered materials when no gate bias is applied (left panel). When a negative V_GS is imposed, the generated potential well allows positive charges to be confined and accumulated at the organic semiconductor/dielectric interface, forming a conductive channel with an in-plane configuration (right panel). Under an imposed bias V_DS, the current (I_DS) starts to flow between the source and drain electrodes.

EGOFETs consist of an electrolyte solution (or an ion gel) in direct contact with the semiconductor and gate electrode, which acts as the dielectric layer. A nonpolarizable gate electrode is preferred for effective gating, and the most utilized is Ag/AgCl pellet. Take a p-channel for example, as illustrated in Fig. 1(c), a negative gate bias attracts mobile cations in the electrolyte solution, establishing an electrical double layer (EDL) at the gate/electrolyte interface. Consequently, anions are repelled from the gate and accumulate at the electrolyte/semiconductor interface. The resulting shift in the chemical potential of the semiconductor at the interface causes holes to be swept into the semiconductor to balance the negative charge of the anionic layer, establishing the second EDL [Fig. 1(d)].³⁵ Therefore, two EDLs generate in series at the gate/electrolyte and electrolyte/semiconductor interfaces, respectively. The gate bias is still capacitively coupled to the organic semiconductors across the electrolyte. As depicted in Fig. 1(d), the excess of cations (anions) decreases with their distance from the gate (organic semiconductor), forming a sequence of the compact Helmholtz layer (denoted by HL), the diffuse layer, and the bulk solution layer. The potential drop at the interfaces occurs only within the double layer where an electric field is formed [left panel in Fig. 1(d)].³⁵ The ions flow in the gate/electrolyte/semiconductor channel structure can be described by an ionic circuit consisting of two EDL capacitors in series, i.e., $CG$ and $CCH$⁠, together with a resistor corresponding to the electrolyte $(RE$⁠) [right panel in Fig. 1(d)]. The total capacitance of the circuit (C) then equals to $CCHCG/(CCH+CG)$⁠. The capacitance is mainly dominated by the EDL capacitor at the electrolyte/semiconductor interface, as the EDL capacitance at the gate/electrolyte interface is generally larger than that at the latter interface.

Unlike OFETs, ionic charges in the electrolyte compensate for the induced electronic charges in the channel of EGOFETs. The dielectric thickness is reduced to an extremely thin dimension on par with the ionic radius, resulting in a high capacitance (on the order of 1–10 $μF cm−2$⁠). In this regard, EGOFETSs are viewed as an extreme case of OFET capable of operating at low bias conditions.³⁶ When the gate voltage is changed due to the neural activity, the concentration of the induced holes (electrons) in the channel varies, leading to a readable variation in the output current. EGOFETs generally possess excellent sensitivity, as the capacitive coupling amplifies subtle electrostatic potential variations at the semiconductor/electrolyte interface by up to three orders of magnitude. Any minuscule potential change is translated into a noticeable difference in the transistor parameters. Therefore, EGOFETs appear as a powerful sensing platform, readily adapted to the detection of not only biologically relevant species³⁷ but also the electrophysiological signals.

The identifying characteristic that distinguishes OECTs from OFETs and EGOFETs is that OECTs effectively use the ion injection from the electrolyte to the organic semiconducting channel to modulate its bulk conductivity [illustration in Figs. 1(e) and 1(f)].^19,36,38,39 Take p-type depletion-mode OECTs as example, the channel most commonly comprises poly(3,4-ethylenedioxythiophene) doped with poly(styrenesulfonate) (PEDOT:PSS). As cations from the electrolyte enter the channel, dedoping of the channel occurs that replaces charge-carrying holes, and thus the channel works in the depletion mode.^14,40,41 Accumulation-mode OECTs are mostly based on intrinsic organic semiconductors, with typically higher operation voltages.^42–44 Representative active materials in OECTs include polypyrrole,⁴⁵ polyaniline,⁴⁶ etc. Comprehensive summaries of the material design principles and latest advances can be found in several decent reviews,^47–50 which is out of the main scope of this article.

In OECTs, individual polymer chains or domains provide a capacitive interface, leading to a volumetric capacitance in bulk. It is up to three orders of magnitude greater than the planar EDL capacitance formed at the organic semiconductor/electrolyte interface. Comparisons in the documented capacitance range of OFETs, EGOFETs, and OECTs are summarized in Tables I^,29,51–58 and II.^{39,43,44,59–68} The volumetric capacitance of OECTs permits an even lower operating voltage while requiring a longer response time. Nevertheless, OECTs effectively capture the ion fluxes, constituting an optimal architecture to measure electrophysiological signals, i.e., electric field fluctuations triggered by the directional movement of ions in biotic environments.

The steady-state characteristics of organic transistors are mainly reflected by their transfer curve and output characteristics. The former depicts the variation of I_DS with the applied V_GS at a certain V_DS, and the latter shows the variation of I_DS with V_DS under a certain V_GS. Major figures of merit extracted from the transfer curve include the on/off ratio, the threshold voltage $Vth$⁠, the carrier mobility μ, and the transconductance $gm (gm=δIDS/δVGS)$ that reflects the sensitivity of channel conductivity in response to a change in electric field. For transistors working as transimpedance amplifiers, the magnitude of the amplification is proportional to its transconductance, $gm$⁠, making $gm$ the most essential figure of merit. The transconductance $gm$ can be expressed as follows:

where μ and C* denote the electronic charge carrier mobility and the volumetric capacitance, $Wd/L$ indicates the channel geometry with W, L, and d representing the channel width, length, and thickness, respectively, and V_th is the threshold voltage related to the work function of the gate electrode and the oxidation or reduction onset of the semiconducting channel. Much effort has been devoted to innovating materials and device geometries to enhance the maximum value, $gmmax$⁠, and to tune the voltage at which it occurs, $Vgmmax$⁠. When used for recording electrical signals, it is desirable to have the $gmmax$ achieved at 0 V to simplify the circuitry and minimize the need for continuous biasing.

Output curves are characterized by a linear regime at $VDS≤VGS−Vth$ and a saturation regime on condition of $VDS>VGS−Vth$⁠. For OECTs, $gm$ depends on both the geometric properties of the device and the intrinsic properties of materials. The product of electronic charge carrier mobility (μ) and the volumetric capacitance (⁠$C*$⁠), i.e., $μC*$⁠, has been viewed as a figure of merit of the active materials in OECT,⁶⁹ independent of the lateral scale and thickness of the channel. For instance, PEDOT:PSS with added ethylene glycol (EG) shows a $μC*$ of 75 $Fcm−1 V−1 s−1$⁠.⁷⁰ Another representative p-channel material, p(g2T-TT), achieves a $μC*$ of 228 $Fcm−1 V−1 s−1$⁠.⁴³ Currently, the highest reported $μC*$ has reached 559 $Fcm−1 V−1 s−1$⁠.⁷¹ Notably, a high $μC*$ implies fulfilled requirements in efficient electronic transport supported by rigid backbones with π-stacked crystallites and simultaneously effective ion transport benefiting from an open and porous morphology.

The transient characteristics are mainly obtained by measuring the variation of I_DS with time when a square voltage pulse is applied to the gate. The response time (τ) is defined by the time taken for the I_DS to cross from a specified low value to a specified high value. The specified low and high values usually refer to 10% and 90% of the final steady-state value, respectively. From the response time, parameters such as cutoff frequency and bandwidth can be further obtained. The cutoff frequency is expressed as $fT=1/2πτ$⁠. For OECTs, the response time is mainly determined by the ion injection speed from the electrolyte to the organic film and the extraction speed of holes/electrons at the source/drain electrodes, according to the device model suggested by Bernards and Malliaras,⁷² where the former one dominates. Rivnay et al.⁷⁰ have found that τ follows the trend of $RsC$⁠, i.e., $τ=RsC$⁠, where $Rs$ and C denote the resistance of the equivalent ionic circuit and the capacitance of the film, respectively. The $CCH$ is proportional to the channel volume, i.e., $CCH$ ∝ $WdL$ and the $Rs$ scales as $Rs∝ 1/WL$⁠. Therefore, the response time τ is proportional to $dWL$⁠, i.e., the smaller the channel lateral size, the quicker the device response. On the other hand, transconductance $gm$ depends on $Wd/L$⁠, i.e., inversely proportional to L, the reduced L in vertical OECTs (vOECTs) benefits the transistor amplification property. One can clearly find a crucial role the device geometry plays in determining the overall electrical performance, both the steady and transient characteristics.

The past decade has witnessed a rising trend in the major figures of merit (Fig. 2), including an enhancement of the transconductance and shortening in the response time, out of the rapid development in materials design and device engineering. Practical device engineering, for instance, introducing a vertical structure^73–75 or interdigitated geometry of electrodes as the source and drain^76,77 have led to improved $gm$⁠, pushing the highest value toward 180 mS.⁷⁶ With the rational design of an internal ion-gated structure,^78,79 the response time has been shortened from 100–300 to 2.6 μs, which will be addressed explicitly in Sec. II B.⁷⁸ 

Different device architectures have their pros and cons in neural activity recording. OFETs are usually employed for electrical stimulation, based on forming an electric field at the dielectric/organic semiconductor interface,^37,63 or used as the active matrix-driven array for in vivo recording.^83,94 On the other hand, OFETs are often susceptible to water or ion penetration damage and must be fully encapsulated to ensure chronic use in physiologic environments.⁹⁵ In addition, most OFETs require relatively high input voltage (on the order of tens of volts) to switch on, leading to undesired hydrolysis or generating too much heat, thus causing damage to tissues. The limitations mentioned above suggest that OFETs are more suited to be integrated into the amplifying circuit for in vitro applications instead of functioning as the recording site in vivo. EGOFETs and OECTs liberate such limitations with their channel directly contacting the electrolyte solution. Notably, the structural reduction impedes the independent gating ability, as the electrolyte is an integral part of the transistor, and ions in the electrolyte are shared by all device units on the same supporting substrate. In this regard, it becomes challenging to build bioelectronic circuits by integrating EGOFETs or OECTs.⁷⁸ On the other hand, OECTs exhibit a longer response time, and $gm$ of OECTs often starts to roll-off for frequencies higher than 1 kHz, while OFETs can be operated at higher frequencies.³⁶ 

Transistors and EGOFETs have been demonstrated to measure electrophysiological signals in vitro from cell cultures and tissue slices.^96,97 An early demonstration of EGOFETs for monitoring the neural cells in vitro was reported by Cramer et al. in 2013, on the basis of an ultrathin pentacene channel.⁶³ In an ex vivo testing configuration, murine neural stem cells were adhered to the device surface and differentiated into neuronal networks. The EGOFET sensor demonstrated sensitivity to small potential changes, exhibiting stability for up to nine days in its cell medium under standard cell culture conditions.

The first demonstration of an organic transistor for in vivo neural activity recording was reported by Khodagholy et al.^40,80 back in 2013, where OECTs with an active channel of PEDOT:PSS were fabricated [Figs. 3(a) and 3(b)] and ECoG on the somatosensory cortex of rats were carried out [Figs. 3(c) and 3(d)]. From in vitro characterization, the OECTs exhibited a maximum $gm$ of 900 μS at a given V_GS = 0.42 V and V_DS = −0.4 V [Fig. 3(e)]. The in vivo performance of the transistors was then evaluated and compared with a PEDOT:PSS surface electrode [Fig. 4(d)] and Ir penetrating electrodes, using an experimental model of epileptiform activity in rats (detailed discussion in Sec. V). The OECTs and Ir penetrating electrodes were sensitive to pickup similar signals that the PEDOT:PSS surface electrodes were unable to detect. OECTs demonstrated higher SNR (22.3 dB) than the penetrating Ir electrodes (18.2 dB). This study implies that the combined features of OECTs, including high gain, biocompatibility, and conformability, make them well-suited for implantable applications.

For chronic in vivo applications, undesired electrochemical processes would lead to membrane rupturing and even death of cells, which should be prevented. An effective strategy is to operate transistors at a considerably low gate voltage. A desired safe window of the gate bias is often considered as below 0.3 V.^98,99 This imposes limitations in the channel material selection, as only those that exhibit considerably low threshold voltage (pentacene in most documented EGOFET sensors) can be applied. Therefore, engineering the device structure to minimize the applied voltage across tissues is of paramount importance. Zhang and colleagues designed liquid–solid dual-gate organic field effect transistors which maintain a safe voltage bias across the cell membrane (i.e., lower than 0.3 V).¹⁰⁰ As shown in Fig. 4(a), a bottom gate and an Al₂O₃/SiO₂ bilayer gate dielectric were introduced. The $Vth$ can be linearly tuned in a voltage window from −0.475 V down to −0.074 V, suggesting the devices can function under a cellular-safe voltage and that a variety of organic semiconductors with different surface potentials can be used. The dual-gate design also demonstrates its applicability in accumulation-mode OECTs, enabling a maximum $gm$ achieved at a gate bias below 0.3 V.

More recently, a novel operation mode was designed for electrolyte-gated organic transistors to realize the functioning at a net zero bias in the cerebrospinal fluid.¹⁰¹ The classical common-source, common-ground configuration of transistors was reverted, giving a common-drain and grounded source (CDGS) configuration [Fig. 4(b)]. In a p-channel device, a positive V_DS drives a positive current, i.e., mobile holes dominating. On the other hand, the potential of the gate electrode controlling the bath, V_GD, refers to the drain itself instead of the source. Figure 4(c) displays an overlay of in vivo transfer characteristics taken from the two configurations, showing agreement in their I–V responses. Maximum $gm$ can be achieved when applying an equivalent while opposite voltages at the drain and the gate, respectively, resulting in a net zero bias in the cerebrospinal fluid. The proposed scheme was proven effective for recording evoked activities and shows good compatibility with the living systems.

As electrolyte solution is an essential part of OECTs, the OECTs' operating speed is dominated by the time taken for ions to migrate between the electrolyte and the polymeric channel. The relatively long response time imposes challenges for OECTs to capture high-frequency neuron firing.^102–106 In addition, although OECTs are efficient transducers, the conventional structures cannot be integrated into bioelectronic circuits, as the integral electrolyte cannot provide individualized input to specific transistors. To overcome such limitations, Khodagholy et al. developed a novel architecture named internal ion-gated organic electrochemical transistor (IGT).^78,79 In this peculiar structure, mobile ions are embedded in the conducting polymer channel, creating a self-modulated doping/dedoping process that no longer relies on ion exchange from a shared external electrolyte [Fig. 5(a)]. A biocompatible sugar alcohol, D-sorbitol, was introduced in the transistor channel consisting of PEDOT:PSS, allowing the creation of an “ion reservoir.” The bottom panel in Fig. 5(a) illustrates that ions within the ion reservoir (in green) are mobile in the vicinity of PEDOT-rich regions (in light blue). Out of its hydrophilic nature, D-sorbitol also enhances the elongation of PEDOT-rich domains and improves the charge percolation pathway in the polymer matrix and thus the overall conductivity. The resulting ion reservoir-contained structure allows individualized gating of the transistors and a faster time constant than electrolyte-based transistors. Ions with different hydrated radii were compared, and the results show that monovalent ions (Na⁺ and K⁺) exhibit the fastest modulation in the drain current. The authors further decreased the channel dimension and reported the fastest time constant of 2.6 μs from a channel having length of 12 μm and width of 5 μm [Fig. 5(b)], corresponding to an effective bandwidth of up to 380 kHz.

Standard integrated circuits are based on both depletion and enhancement mode transistors. To resemble the function of standard integrated circuits, the same group later developed an enhancement mode IGT, named “e-IGT,” consisting of PEDOT:PSS-polyethylenimine (PEI) as the active channel.⁷⁹ Electrons are transferred from the amine groups of PEI to PEDOT:PSS, which decreases the intrinsic conductivity of PEDOT segments by several orders of magnitude. Charge balance is maintained now by the ionic bonds between protonated PEI⁺ and PSS⁻. The initially reduced PEDOT:PSS film then allows reversible ion-based doping, working as an enhancement-mode transistor. The e-IGT having 5 μm channel length exhibited a response time of 2.9 μs and a $gm$ of ∼1.5 mS. Notably, the authors suggested using $gm$/τ ratio as a measure of gain-bandwidth product. As illustrated in Fig. 5(c), $gm$/τ of e-IGTs ranks at the top among multiple architectures of organic transistors, demonstrating simultaneous achievement of a high amplification and high speed, which often are tradeoffs due to the geometrical design. In fact, internal mobile ions inside the channel generally enable higher $gm$/τ compared with external ions supplied by the electrolyte.

Combining IGTs with enhancement and depletion modes allows for designing integrated bioelectronic circuits. The authors further demonstrated successful constructing of digital logic gates and cascaded amplifiers.⁷⁹ In short, a rational design incorporating independent ion reservoirs and an ion membrane between the gate and channel beautifully realizes the volumetric capacitance, shortened ionic transit time, and integration capability. Applications of flexible IGTs and IGT-based circuitry in neural activity recording will be addressed further in Sec. V.

Reduction in the transistors' physical size is an effective strategy to better utilize the substrate space, achieve greater integration density and lower power requirements.¹⁰⁷ However, most OECTs reported to date have a large lateral size (micrometer scale), limiting the transistor performance and their integration density. Donahue et al. reported a vertical OECT (vOECT) consisting of PEDOT:PSS active layer.⁷³ Instead of having a planar configuration, the source and drain contacts are vertically stacked and thus the channel length, L, is defined by the dielectric layer thickness [Figs. 5(d)–5(f)]. In this sense, L of vOECTs can be engineered down to hundreds of nanometers, which is orders of magnitude shorter than in conventional planar OECTs. In addition, since $gm$ depends on $Wd/L$⁠, i.e., inversely proportional to L, the reduced L in vOECTs benefits the transistor amplification property. The use of vOECT architecture allows for the preservation of high gain, occupying less than half of the substrate area required for planar counterpart devices. In particular, an enhanced inherent $gm$ of 57 mS was achieved, corresponding to a geometry-normalized transconductance of 814 S m⁻¹. The reduced channel length further shortens the time required for the ion to travel from the electrolyte to the active channel, thus to realize a shorter response time and an increased cutoff frequency (exceeding 1.5 kHz in this work). In a follow-up study, a novel cofacial pair complementary inverter was fabricated, with two vOECTs facing one another that share a single channel. As a prototype demonstration, the vOECTs-based inverter demonstrated a peak gain of around 28 while having a footprint equivalent to that of a single planar OECT.^74,108 The design of vOECTs provides a novel transistor architecture for bio-interfacing applications, with great potential toward increased channel number and recording sites, much higher spatial resolution, and improved electrical performance.

The mechanical mismatch between the neural interface and tissue will cause severe problems, including tissue damage and premature failure of the probe. Conventional Si-based FETs can be fabricated into MEAs with high density. At the same time, the large mechanical mismatch between rigid Si and the soft tissue, and a lack of biocompatible interface have precluded its application as a chronic neural interface.¹⁰⁹ It is thus essential to fabricate flexible transistors with a bending stiffness similar to brain tissue.

In an electronic device consisting of a multilayer structure, the strain estimation takes Young's modulus and thickness values of all stacking layers into consideration. A neutral mechanical plane, b, also called the zero-strain position, is given by the following formula:¹¹⁰ 

where b is the distance form the top surface to the neutral mechanical plane, and $E¯i$ and $hi$ denote Young's moduli and thickness of each layer [Fig. 6(a)]. Kim et al. combined the multilayer neutral mechanical plane layouts and “wavy” structural configurations and successfully demonstrated Si-based complementary logic gates, ring oscillators, and differential amplifiers with mechanical flexibility.¹¹⁰ These Si-based integrated circuits exhibited good electrical properties even when folded in half. In fact, the multilayer neutral mechanical plane model has been widely adopted in different device categories, including FETs,^32,111,112 organic light emitting diodes (OLEDs),^113,114 flexible battery,¹¹⁵ wearable device,¹¹⁶ and electrochemical devices,¹¹⁷ etc. For multilayer devices, soft adhesives between brittle functional layers can be used to split the neutral plane into multiple ones, thus protecting brittle materials at various locations.^118–120 

The importance of minimizing device thickness in enhancing the overall flexibility is well revealed by the definition of flexural rigidity, D, as shown in the following equation:¹²¹ 

where E, t, and ν are Young's modulus, thickness, and Poisson's ratio of the film, respectively. As the flexural rigidity is proportional to the cube of the thickness t, the most direct approach to decrease D is to reduce the device thickness. This has often been seen in rigid material systems, for instance, combining prefabricated thinned and small Si parts onto plastic substrates.^110,122 Kaltenbrunner et al. combined both strategies, i.e., using an ultrathin device structure and placing the relatively brittle layers close to the neutral mechanical plane, and demonstrated successfully ultraflexible OFET-based active-matrix array.¹²³ Such structure withstands repeated bending to a radius down to 5 μm [Figs. 6(b) and 6(c)] and accommodates stretching for up to 230% on pre-strained elastomers, showing unprecedented mechanical compliance. A similar design strategy can be found in flexible OECTs. For instance, the first in vivo OECT for brain activity recording^40,80 adopted an ultrathin design, with a total thickness down to 4 μm, resulting in excellent conformability.

Aside from flexibility, stretchability is another essential property to realize a seamless tissue-electronics interface for accurate signal acquisition. The most direct strategy is to develop intrinsically stretchable materials.¹²⁴ In particular, Bao group has contributed enormously to innovating materials and fabrication processes toward intrinsically stretchable electronics.^125–129 For instance, a design concept for stretchable semiconducting polymers was presented in 2016, which involves introducing chemical moieties, i.e., 2,6-pyridine dicarboxamide (PDCA), to promote non-covalent cross-linking of the conjugated polymers, hydrogen bonding in this case [Fig. 6(d)].¹²⁵ Such non-covalent cross-linking moieties then allow energy dissipation through breakage of dynamic bonds under strain, with retaining high charge carrier transport intrinsically. Another effective design strategy involves introducing conjugated fused rings with bulky side groups, thus maintaining a highly conjugated polymer backbone responsible for high electrical performance.¹²⁷ Later, the same group reported a strain engineering toward strain-insensitive stretchable transistor arrays.¹³⁰ They incorporated patterned elastomer layers with tunable stiffnesses into the transistor structure, thus to reduce the strain on the active regions of the devices. This approach has led to an integrated array having a density of 340 transistors cm⁻². The transistors exhibit performance variation of less than 5% under 100% strain. Dai et al. reported an intrinsically stretchable OECT, based on the stretchable polymer semiconductor p(g2T-T), exhibiting high $gm$ and a biaxial stretchability for 100% strain.¹³¹ Above innovations have shown great feasibility in material and processing engineering toward intrinsically stretchable electronics. With such, the electronics can form intimate contact with the human body and record the physiological signals precisely during the body movement.

Compared with polymers, commonly used metallic electrodes are more likely to crack. In general, the device failure under stretch is mostly due to metal contact fatigue. Therefore, stretchable electrodes are indispensable for fabricating a fully stretchable device. Vertical metal nanowires embedded in an elastomer matrix have been reported to realize intrinsically stretchable organic transistors.¹³² In addition, silver nanoparticles form in situ from silver flakes embedded in an elastomer, demonstrating benchmark conductivity when being stretched up to 400%.¹³³ Comprehensive summary of the latest progress on stretchable electrodes for stretchable transistors can be found in recent reviews,¹³⁴ thus will not be further addressed herein.

Microstructure engineering constitutes an effective indirect method to endow the system with stretchability. For instance, porous structure is widely adopted to realize stretchable conductors and electronics.^135–137 Recently, an electrochemically redox-stable OECT was reported, showing negligible performance degradation under tensional strains up to 30%–140%.¹³⁸ This intriguing property was realized by combining the semiconducting polymer's uniform honeycomb film morphology with a biaxially pre-stretched film/device architecture. The porous structure [Fig. 6(e)] increases the film's specific surface area and facilitates ion exchange between the electrolyte and the polymer channel, thus reaching a negligible hysteresis. Under the deformation of the porous honeycomb film, it forms an out-of-plane buckling structure [Fig. 6(f)]. This out-of-plane buckling deformation facilitates the geometric expansion of the honeycomb and benefits the release of local stress concentrations during stretching [Figs. 6(f) and 6(g)]. Similar buckling effects were also widely used for intrinsically non-stretchable materials.^110,139,140 A common way to get the buckling structure is to transfer the freestanding film to a pre-stretched elastomer and then release the strain, leading to sinusoidal waves out of the plane.

Two-dimensional arrays of transistors are highly desired to map brain activity on a large scale and unravel propagation of neural signals and complex neuron networking. Arrays have been built up with different configurations. As depicted in Figs. 7(a)–7(c), arrays having the transistors individually wired [Fig. 7(a)], multiplexed arrays with common data and scan lines [Fig. 7(b)], and arrays in which each pixel contains a sensing transistor and a multiple-addressing transistor [Fig. 7(c)] are main documented categories in constructing multielectrode arrays.¹⁴¹ The first type can tune the biases of each transistor individually to overcome heterogeneity in device metrics, making the readout signals in different sites more meaningful, but this configuration also requires high energy consumption and wastes space to individually wire these transistors [example shown in Fig. 7(d)].¹⁴² The second type effectively reduces the total number of wirings but suffers a high multiplexer crosstalk [example shown in Fig. 7(e)].⁹³ Switching of the circuit is controlled by the drain bias, V_DS, instead of V_GS. The crosstalk can be minimized further by reducing the load resistance and wire resistance. Adding an addressing transistor to control the sensing transistor on and off allows the energy consumption and crosstalk to be reduced simultaneously [Figs. 7(f) and 7(g)].⁸³ 

Conventional multielectrode arrays request individual wiring of each passive electrode, incapable of sampling large brain areas with high spatial resolution. Integrating flexible Si nanomembrane transistors into the electrode array was first demonstrated by Viventi et al. in 2011.⁹ A 360-channel active electrode array was fabricated with a reduced number of wires ninefold, demonstrating capability in sampling a large region of the brain (10 $×$ 9 mm²). The active array exhibits a relatively high spatial resolution (500 μm) and a temporal resolution beyond $10 kS s−1$⁠.

For large-area mapping which requires thousands of electrodes or transistors integrated, multiplexing configuration is a must. Compared to their inorganic counterparts, the development of organic active arrays is still left behind. To increase the spatial resolution, the footprint of each transistor and the space for wirings must be reduced. The architecture of vOECTs mentioned above demonstrates a highly effective strategy to reduce the spatial footprint of the transistor while maintaining its high electrical performance.⁷³ Based on this architecture, the OECT array having each transistor wired individually still shows a high spatial resolution,⁷⁵ as the footprint of these wires is considerably minimized by vertically stacking the source and drain connecting wires. In this particular case, the spacing between adjacent OECTs is 80/160 $μm$⁠, comparable to that of the state-of-the-art multielectrode array.¹⁴³ 

To increase the collective sampling rate in an array, as the signal continuously varies with time, the sampling interval should be significantly shortened to better restore the characteristics of input signal and realize high temporal resolution. The sampling frequency, defined as the reciprocal of the sampling interval with a unit of samples, is also an indicator of how accurate the measured signal waveform is within a unit time. Sampling results can be affected by the quality of single devices, the crosstalk, and signal losses during transmission. Lee et al. integrated OECTs and OFETs, and realized an ultraflexible electrophysiology array (5 $×$ 5), exhibiting high temporal resolution and a cutoff frequency of around 3 kHz.⁸³ With a total thickness down to 2.0 μm, the integrated multiplexing array can be laminated on the rat's gracilis muscle surface, exhibiting negligible mechanical interference during the dynamic muscle motions. Later, the same group demonstrated an active OECT array with nonthrombogenicity, stretchability, stability, and blood compatibility.⁸⁵ This design has taken advantages of high gain OECT processed on microgrid substrates and a coating of poly(3-methoxypropyl acrylate), exhibiting antithrombotic properties with maintaining high ionic conductivity [Fig. 7(h)]. Although the channel number is still behind the Si-based arrays, this design pushes a step forward to organic multiplexed arrays for chronic in vivo applications.

Bioelectronics should optimally merge a soft and biocompatible tissue interface with high spatiotemporal signal processing capacity. As mentioned earlier, flexible organic transistors have been applied in vivo for epilepsy diagnosis and cortical mapping.⁴⁰ The first documented ECoG probe was based on OECTs consisting of PEDOT:PSS active channel, placed on the somatosensory cortex of rats [Figs. 8(a) and 8(b)]. Before that, PEDOT-based electrodes were applied as recording electrodes in vivo, showing superior performance in chronic experiments compared to traditional metal electrodes.¹⁴⁴ By triggering epileptiform activity with bicuculline, OECTs in vivo displayed excellent SNR (44 dB) due to local amplification capability, compared with PEDOT:PSS-based surface electrodes (24.2 dB) [Fig. 8(c)]. Furthermore, pathological epileptiform activity was recorded in an experimental model with the Genetic Absence Epilepsy Rat from Strasbourg. As detailed in Fig. 8(d), the transistor again outperformed the surface electrodes by having an SNR of 52.7 dB over 30.2 dB. A higher SNR generally translates into a shorter overall recording time required to obtain the same amount of information. Therefore, using OECTs in vivo constituents a breakthrough in recording more local activities with high precision.

Advanced neural interfaces are desired to be capable of stimulating individual neurons in a minimally invasive manner while recording the neural activity simultaneously. OECTs with an ultraflexible configuration, i.e., having a total thickness down to 4 μm, demonstrated the ability to inject current pulses from their channel to the hippocampus, which then allows the stimulating of targeted populations of neurons [Fig. 9(a)].¹⁴⁵ As observed, the evoked intracellular calcium levels of pyramidal neurons correlated directly with the applied stimuli. The device elicited minimal glial scarring after being implanted in the brain for one month. Under recording conditions, neurons close to the transistor (20 μm away) were barely affected by the current flow in the channel. The results implied application potentials of OECTs in recording the neural activity and combing biochemical stimulation with electrical stimulation, creating a novel tool for future in-depth research of neuroscience.

Another effective strategy for simultaneous recording and stimulation is constructing devices with high optical transparency. This has been well demonstrated by arrays consisting of transparent OECTs and transparent Au grid wirings [Fig. 9(b)].⁸⁴ Notably, the linewidth of Au grid wiring, i.e., 3 μm, is much smaller compared to the typical size of neuron cells. The transparent active OECT array realizes the spatial mapping of ECoG signals from an optogenetic rat. With laser stimulation, the evoked response can be recorded simultaneously [Fig. 9(c)]. This work allows a thorough investigation of neural activity with direct light stimulation.

Neural interface electronics are expected to map over a large area and listen to the individual neural activities. Following the development of IGTs (working mechanism discussed earlier in Sec. II), which are high-performance building blocks designed to function in biotic environments, Khodagholy et al. further demonstrated the outstanding transient response of the enhancement-mode IGTs (e-IGTs).⁷⁹ Conformable arrays of e-IGTs placed on the cortical surface enable recording of LFPs, as shown by the time-frequency spectrogram in Fig. 9(d), taken two weeks after the implantation. In the recordings, representative spectral features of variable states, including wakefulness, non-rapid eye movement sleep (NREM), and rapid eye movement (REM) sleep, were all distinguishable. Furthermore, the e-IGTs can acquire neural spiking activity, i.e., action potentials, precisely from deep cortical layers. As depicted in Fig. 9(e), high-pass filtering (>250 Hz) of the neural signals show a consistent waveform morphology and reveal large amplitude transients ascribed to action potentials. Autocorrelation of their occurrence times demonstrates a pattern consistent with the physiologic refractory period of an individual neuron, validating the ability of e-IGTs to record the action potential in vivo. The above demonstrations reveal an effective neural interface simultaneously providing high speed and amplification.

Finally, in situ processing of the acquired biological signals rely on integrated bioelectronic circuits. This is challenging to realize, as OECTs suffer from their commonly shared electrolyte while Si-based components lack biocompatibility and stability in biotic environments. The only documented neural processing circuit using soft organic transistors is based on the combined d-IGT (depletion-mode IGT) and e-IGT units [Fig. 9(f)], which yield a non-linear rectification circuit, offering high SNR recording, real-time detection of epileptic discharges, and in situ processing of the detected signals.⁷⁹ In particular, a non-linear rectifier could suppress lower but variable-amplitude non-target signals, thus improving the detection precision. As demonstrated by Khodagholy et al., the IGT-based non-linear rectification circuit was implanted in the hippocampus of an epileptic rat, exhibiting accurate detection of epileptic discharges [Fig. 9(g)]. The receiver operating characteristics [Fig. 9(h)] of IGT-based circuit greatly surpassed traditional methods, including the bandpass filter or amplitude thresholding.

In short, the soft IGT-based circuitry, a counterpart of traditional Si-based circuits, unprecedentedly combines features of mechanical compliance, biocompatibility and stability, and most importantly, the signal acquisition and processing capability, well-suited for capturing and processing neurophysiological signals in a real-time manner. With such, more complicated bioelectronic circuits capable of safely capturing and stimulating neural activities, in situ amplifying and processing their resulting signals seem within reach.

To summarize, organic transistors possess unique advantages for neural signal recording, given their designable flexibility, biocompatibility features, and inherent amplification capabilities for in situ signal processing. The designed features have shown potential in diagnostic and neuromodulation applications.

Combining electronic and ionic conductivity to create enhanced functionality in devices is an emerging development trend in innovating material and device structures. This is particularly essential in building up an effective and chronic neural interface, as ions dominate the electrical processes in living matter. Fundamentally, classic OFETs are unsuitable for sensing, as their driving voltages may result in undesired electrochemical reactions that could potentially arouse tissue damage. They are good candidates to be integrated as the active-matrix addresser. Electrolyte-gated organic transistors have been demonstrated successfully as transducers in vivo, given their working principle functionally integrating the electrolyte biological environment and their relatively low operating voltage. In particular, OECTs have shown capability as neural interfaces in vivo, combining high volumetric capacitance, high transconductance, and biocompatibility. On the other hand, electrolyte-gated structure and organic mixed conductors bring disadvantages, foremost, the slower response time. To this end, novel device architecture was proposed, for instance, the internal ion-gated structure^78,79 and the vertical OECTs with decreased channel volume.⁷³ 

Over the past decade, organic transistors for neural activity recording have developed rapidly, while numerous challenges are still confronted. First, an ideal ECoG system is expected to achieve a high spatial resolution (∼10 μm) and temporal resolution (in the sub-millisecond range) over a large scale and excellent ability in the bionic/biotic environment for weeks or even years. The spatial resolution of current active arrays is in the millimeter range,^83,84,106 far from its inorganic counterparts.⁹ The design of vertical OECTs⁷³ has effectively reduced the device footprint, yet how to scale it up into a high-density active array remains unexplored. Furthermore, real-time processing of biological signals requires integrated bioelectronics capable of signal amplification and stimulation.⁷⁸ Second, organic materials are designed to be intrinsically flexible, while metal electrodes and external wires are relatively brittle. Scaling across large brain regions requires integrating electronics to multiplex many electrodes to a few external wires. Flexibility at a system level requires the rational design of the device architecture to minimize strain at the relatively brittle layers.

To design fully implantable neural interfaces, system-level integration remains a big challenge. Functional neuroelectronic devices allowing simultaneous recording and stimulation of cellular electrical activity are essential for an in-depth understanding of neuronal and neurodegenerative diseases. Apart from the signal acquisition module, crucial components of the fully implantable system, i.e., the data transition module including radio frequency (RF),¹⁴⁶ ZigBee,¹⁴⁷ Bluetooth,¹⁴⁸ or WiFi,¹⁴⁹ and the power supply module including wireless rechargeable battery,¹⁵⁰ inductive coupling,¹⁴⁶ and other energy transfer forms, such as infrared¹⁵¹ and ultrasound¹⁵² are urgently needed. For instance, Dai et al. recently demonstrated an intrinsically stretchable neuromorphic array based on OECTs showing excellent computational characteristics, and realized in situ classifications of electrocardiogram signals without constraining the subject's movement.¹⁵³ This technique makes it possible to develop a fully on-body neural signal monitoring system. Power dissipation of the system and the electromagnetic absorption of the tissue often lead to a temperature increase in surrounding tissues, which should be maintained at less than 1 °C. Such requirement brings significant challenges in the system design, as tradeoffs between channel count, power consumption, and device lifetime should be thoroughly considered. Finally, successfully translating organic transistors as sensors and transducers to a clinical setting requires guaranteed safety and stability. Optimal neural implants should demonstrate their conformability to brain tissues during natural motions associated with pulsatile blood flow and respiration, thus minimizing undesired glial scar formation or cellular damage at the interface.¹⁰⁹ 

In short, the following key features need to be achieved to bring the organic neural interfaces into actual play with efficient, safe, and prolonged capability. To precisely detect low-amplitude signals at the physiologically relevant time scales, transistor building blocks should exhibit high transconductance and high speed. In addition, they are expected to have excellent conformability to the soft tissue and stability of constituent materials to survive biotic environments. Furthermore, independent gating that enables circuit computations and compatibility with other electronic components allowing system-level integration are challenges that require considerable research input.

Today's organic transistors can reliably obtain high-quality neural signals, but more efforts are needed toward a fully implantable and large-scale, long-term stable neural interface. It is believed that with the help of such integrated and flexible systems, the mystery of the brain will be greatly unveiled, and the application of the brain–computer interface will be within reach.

The authors acknowledge the support from the National Natural Science Foundation of China (No. 52003141), Natural Science Fund of Guangdong Province (No. 2021A1515010493), Outstanding Youth Basic Research Project for Shenzhen (No. RCYX20210609103710028), Shenzhen Science and Technology Innovation Committee—Stable Support Key Project (No. WDZC20200818092033001), Tsinghua Shenzhen International Graduate School (SIGS) (Nos. HW2020007 and QD2021006N).

The authors have no conflicts to disclose.

Wei Xu: Conceptualization (equal); Investigation (equal); Writing – original draft (lead). Jingxin Wang: Investigation (supporting); Writing – original draft (supporting). Simin Cheng: Investigation (supporting); Writing – original draft (supporting). Xiaomin Xu: Conceptualization (lead); Funding acquisition (lead); Investigation (equal); Methodology (equal); Supervision (lead); Validation (equal); Writing – original draft (equal); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.