Complementary integration of organic electrochemical transistors for front-end amplifier circuits of flexible neural implants

Abstract

The ability to amplify, translate, and process small ionic potential fluctuations of neural processes directly at the recording site is essential to improve the performance of neural implants. Organic front-end analog electronics are ideal for this application, allowing for minimally invasive amplifiers owing to their tissue-like mechanical properties. Here, we demonstrate fully organic complementary circuits by pairing depletion- and enhancement-mode p- and n-type organic electrochemical transistors (OECTs). With precise geometry tuning and a vertical device architecture, we achieve overlapping output characteristics and integrate them into amplifiers with single neuronal dimensions (20 micrometers). Amplifiers with combined p- and n-OECTs result in voltage-to-voltage amplification with a gain of >30 decibels. We also leverage depletion and enhancement-mode p-OECTs with matching characteristics to demonstrate a differential recording capability with high common mode rejection rate (>60 decibels). Integrating OECT-based front-end amplifiers into a flexible shank form factor enables single-neuron recording in the mouse cortex with on-site filtering and amplification.

Front-end integration of organic amplifiers allow processing of neural signals on-site in a minimally invasive manner.

INTRODUCTION

The development of neural probes, capable of on-site amplification and signal conditioning of neuronal signals, has been an increasingly important focus of neurotechnology research in the past few decades (1–3). However, the current state-of-the-art, silicon-based technology is limited by the rigidity of the implants where the hard electrodes do not match the softness and the constant, dynamic movement of the brain, creating damage upon implantation and chronic inflammation (4). Here, we instead develop integrated circuits for this front-end analog signal processing in neural recording applications using soft, flexible semiconductors with dual ionic-electronic conductivity.

Owing to their intrinsic softness, biocompatibility, and mixed ionic and electronic charge transport, conjugated polymers (CPs) have experienced tremendous attention for various applications requiring a direct interface with aqueous electrolytes, such as neural interfaces and biomimetic platforms (5–11). Among CP-based devices, the organic electrochemical transistor (OECT) is particularly suitable for neural interfacing as a miniaturized amplifier of ionic signals. The OECT uses the CP in its channel gated through an aqueous electrolyte (such as the cerebrospinal fluid). When a gate voltage (V_GS) is applied, ions are injected inside the CP channel, which electrostatically compensate for the electronic charges present in the channel upon application of a drain-source voltage V_DS. The resulting drain current (I_D) is proportional to the number of mobile charge carriers (holes or electrons for a p-type or n-type semiconductor, respectively) in the semiconductor film (12). The magnitude of voltage at the gate electrode controls the number of ionic charges that enter or exit the channel and dope or dedope the film. This precise modulation of the CP doping state (conductivity) can be used for various modes of operation, depending on the type of the CP. For example, if the channel is made of an intrinsically conducting (doped) CP, such as poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS), the device will operate in depletion mode (13). In the PEDOT:PSS film, the negatively charged sulfonate groups electrostatically balance the holes on PEDOT, rendering the film conductive. With a positive V_GS, cations are injected into the PEDOT:PSS channel and compensate for the PSS anions, while holes are extracted at the contacts, leading to “de-doping” of the channel. Alternatively, if a nondoped, intrinsically semiconducting polymer is patterned in the channel, the OECT operates in enhancement mode. In contrast to commonly used CPs in solar cells or light emitting diodes, these new generation, water-compatible CPs used in the enhancement mode OECTs generally have ethylene glycol (EG) side chains to aid ion penetration and transport in the polymer bulk (14, 15). In an OECT configuration, the polarity of the applied V_GS dictates the type of ions injected in the CP channel, where anion and cation injection are used to electrochemically switch on p- and n-channels, respectively (14, 16).

However, most of these OECT structures have only been tested as in vitro platforms for biomolecular sensing applications and were not implemented into tissue-facing interfaces. In addition, circuit structure has been largely limited to single-transistor amplifiers (17–19). This contrasts with the complexity of comparable complementary metal-oxide semiconductor front-ends which include multitransistors differential amplifiers and filters (20–22). The construction of similar circuits with OECTs requires complementary transistors with controllable threshold voltage and transconductances (g_m). However, the electronic and ionic conductivity of CPs, and the resulting OECT characteristics, have considerable variability with n-type–based devices lagging behind their p-type counterparts (23). OECTs are often fabricated at millimeter size scales to mitigate this mismatch, which is completely incompatible with the size scale required for neural recording (24, 25). In addition to these challenges with performance mismatch, amplifier topologies based on complementary p- and n-OECTs usually lack the electrical performance required for neural recording since transistors typically have g_m values only in the range of μS (26, 27). Very recently, amplifiers combining high g_m p- and n- type OECTs have been reported, but their performance has not been tested in physiologically relevant media, and such amplifiers have never been fabricated on flexible substrates (28). Consequently, these limitations in existing devices and circuits highlight the need for the generation of micro-scale high-performance complementary (p and n) OECT-based circuits and their integration and application in in vivo in implantable form factors.

In this work, we combine PEDOT:PSS OECTs with n-and p-type enhancement mode OECTs fabricated using two recently developed CPs with EG side chains. This device combination is used to fabricate voltage amplifiers that are small enough to be accommodated under a recording electrode of 20-μm diameter. By scaling the channel geometries with submicron precision within a vertical device architecture, we are able to tune the electronic and ionic characteristics of the individual OECTs. We demonstrate two amplifier topologies. We first use complementary enhancement-mode p- and n-type OECTs as a push-pull amplifier topology for voltage amplification, achieving a maximum small-signal gain of 31 dB at a power consumption of 160 μW. We next show that the combination of enhancement mode p-type OECT with a depletion-mode p-type OECT load can be made to operate with overlapping output characteristics with the same input voltage. This way, we can achieve a differential amplifier topology with common-mode rejection capability. For both architectures, by using common pads for the supply voltages, we are able to integrate arrays of 16 amplifiers onto a flexible shank form factor. The push-pull amplifiers we demonstrate allow recording neural signals in a wide frequency range from local field potentials (LFPs) to single unit activity (SUA) in mouse visual cortex. Furthermore, our differential amplifiers show that LFPs can be rejected as common-mode signals, allowing SUA to be detected in the absence of high-pass filtering. We envision that the OECT-based complementary circuits we demonstrate here leverage the properties of CP-based materials to generate a new power-efficient, minimally invasive and high-performance platform for implantable bioelectronics.

RESULTS

Push-pull amplifier architecture

To generate a push-pull amplifier architecture capable of voltage gain, we connect complementary p-type and n-type enhancement-mode OECTs in series. The p-type material we choose is poly[2-(3,3′-bis{2-[2-(2-methoxyethoxy)ethoxy]ethoxy}-[2,2′-bithiophen]-5-yl)thieno[3,2-b]thiophene], p(g-2 T-TT) (29, 30). For the n-type CP, we use a naphthalene diimide thiophene-based (NDI-T) backbone functionalized with EG side chains, namely, p(C₆NDI-T) (14, 16, 17). Both polymers have been recently developed specifically for OECTs and shown high g_m with stable performance in biochemical sensing. The mode of operation of this architecture is illustrated in Fig. 1A. Here, we define the supply voltage (V_DD) as the potential applied to the source of the p-type OECT with the source of the n-type OECT (V_SS) grounded. We measure the potential at the connected drains of two OECTs (V_out) as the amplifier output. The input potential, (V_in), is applied over the common electrolyte and with respect to the source of the n-type OECT. At the point at which V_in ≈ V_out in the voltage characteristics, both channels are partially electrochemically doped through cation migration in the case of p(C₆NDI-T) and anion migration in the case of p(g2T-TT). For V_in greater than this value, the cation migration inside the p(C₆NDI-T) leads it to be further doped whereas p(g2T-TT) gets de-doped due to anion depletion. Application of a V_in less than this value reverses this ionic flow, resulting in the doping and de-doping of the p(g2T-TT) and p(C₆NDI-T) channels, respectively. The small-signal the gain of the amplifier (A_V) at any operating point is determined by

$$A_{\mathrm{v}}=-\frac{g_{\mathrm{m},\mathrm{p}}+g_{\mathrm{m},\mathrm{n}}}{g_{\text{ds},\mathrm{n}}+g_{\text{ds},\mathrm{p}}}$$ (1)

where g_m,p and g_ds,p are the transconductance and output conductance of the p-type OECT, and g_m,n and g_ds,n are the transconductance and output conductance of the n-type OECT. Here, by tuning the OECT device geometry, we seek to find an operating regime where this gain is maximized for the V_in≈V_out operating point. For in vivo applications, V_DD must be low enough (<1.5 V) to avoid hydrolysis and minimize power consumption. Gain in these amplifiers is superior at lower supply voltage than designs with passive loads that require larger V_DD to achieve sufficient gain (31).

Fig. 1. Structure and operation principle of the organic push-pull amplifier.

(A) Operation principle of the push-pull amplifier. p- and n-type polymers, namely, p(g2T-TT) and p(C₆NDI-T), are connected in series between rail voltage pads, V_DD and V_SS, respectively. In steady state, V_in is set at center potential of V_DD and V_SS leading to both polymers in a partially electrochemically doped state. On the left (blue brackets), shifting the V_in toward positive values leads to de-doping of the p(g2T-TT) film and doping of the p(C₆NDI-T) film. On the right (red brackets), shifting V_in toward negative values leads to doping of p(g2T-TT) film and de-doping of p(C₆NDI-T) film. In either case, V_out reflects the changes of V_in as a function of g_m of each OECT. (B) Top image represents the top view of the amplifier. Interconnects for V_DD, V_SS, and V_out electrodes are patterned as separate layers with parylene C insulation in-between them. Dashed black lines represent the OECT channels patterned vertically between the top V_out electrode and corresponding rail electrodes. Bottom image represents the cross section of the amplifier. The channels for the n-type and p-type OECTs are defined by the polymers connecting V_SS-V_out and V_DD-V_out pads vertically. (C) Optical image of the push-pull amplifier with optimized dimensions. The n-channel area is highlighted in green, and the p-channel area is highlighted in purple for better illustration. To increase the W/L ratio of the n-channel, we pattern the output pad in a saw-tooth pattern to increase the channel width. The inset illustrates the subsegment of the saw-tooth output pad.

Tuning channel geometries for complementary operation

The layout of the push-pull amplifier is illustrated in Fig. 1B. Parylene C layers serve as the dielectric separator between the OECT contacts (i.e., V_out pads, V_DD, and V_SS supply electrodes), enabling vertical polymer channel geometries. Vertical OECTs have been shown to achieve much higher g_m and speed of operation compared to planar configuration at a given geometry, because the channel length (L) is not limited by photolithography resolution but instead by the dielectric thickness in vertical desings (32). In this way, the g_m of individual components can be tuned by adjusting the L (determined by the thickness of the parylene C insulator) and the channel width (W, determined by the diameter of the top electrode).

First, we generate single-element vertical OECTs using p(g2T-TT) and p(C₆NDI-T) channels separately with identical geometries (W = 5 μm and L = 1 μm) and measure the resulting current-voltage (I-V) characteristics (fig. S1). The g_m of the p(g2T-TT) OECT is approximately two orders of magnitude higher than of the p(C₆NDI-T) OECT. To match the performances of the devices, we tune the p(C₆NDI-T) channel dimensions to achieve a W/L ratio of ~200 by using a 500 nm parylene C separation between metal contacts (L = 500 nm) and by giving the top electrode a “saw-tooth” design (resulting in an effective W of ~100 μm), as illustrated in Fig. 1C. For the p(g2T-TT) channels, we decrease the initial channel dimensions to a W/L ratio of ~4 (W = ~16 μm, L = 4 μm), resulting in a ratio of n-type W/L to p-type W/L of ~50. For each OECT type, we keep the polymer coating thickness identical (d = 100 nm). To achieve the complementary characteristics on a large scale, we carefully monitor polymer and parylene thickness during the fabrication process and adjust the layout of the active areas accordingly (see Materials and Methods and fig. S2, A to D). Figure S3 (A and B) provides additional examples of I-V curves obtained from distinct wafers using the optimized design layout to illustrate the reproducibility of our approach.

Complementary operation of OECTs in a push-pull amplifier

With the optimized device geometry (illustrated in Fig. 1C), we achieve the I-V characteristics of the n-channel p(C₆NDI-T) and p-channel p(g2T-TT) OECTs, shown in Fig. 2 (A and B). For in-series integration, we choose the V_in ≈ V_out operating point for a V_DD of 0.8 V resulting in a ∣V_DS∣ of 0.4 V for both OECTs and allowing operation in the saturation regime for both devices to maximize small-signal voltage gain (Eq. 1). For the corresponding operation point (V_SS at ground potential), the I_D of both devices is ~0.2 mA, and g_m values for both devices exceed 1.5 mS (Fig. 2, C and D). At this configuration (at which the power consumption is ~160 μW), we calculate the A_V of the amplifier to be ~32 dB according to Eq. 1 using the corresponding g_ds,n and g_ds,p values, both of which are less than 0.1 mS with saturation biasing (fig. S4). In addition, we conduct tests to assess the stability of our amplifiers over a period of 10 days (fig. S5A). First 7 days, we do not observe any notable decay in the performance with the small-signal gain changes remaining in the range of ~5%. After day 8, the gain degredation exceeds 30%, primarily as a result of changes in the I-V characteristics of the p(C₆NDI-T) OECT (fig. S5B).

Fig. 2. Electrical characterization of n-type and p-type OECTs with optimized geometries.

(A) Graph shows I-V characteristics of a p(g2T-TT) based OECT with a W/L ratio of 16/4. The section marked in purple illustrates the linear operation regime, while the section marked in orange illustrates the saturation regime. (B) Graph shows I-V characteristics of a p(C₆NDI-T) based OECT with a W/L ratio of 100/0.5. The section marked in purple illustrates the linear operation regime, while the section marked in orange illustrates the saturation regime. (C) g_m of the p(C₆NDI-T) (in green) and p(g2T-TT) (in purple) OECTs for a V_DD of 0.8 V and a V_SS at ground potential, respectively, as a function of V_in. (D) I_D of the p(C₆NDI-T) and PEDOT:PSS OECTs for a V_DD of 0.8 V and a V_SS at ground potential as a function of V_in. (E) Voltage-to-voltage amplification using a push-pull topology. Left illustrates the applied input pulses with an offset at 0.5 V. One-millivolt sinusoidal pulses are applied within a frequency range of 0.1 to 100 kHz. Plot on the right side shows the A_V of the push-pull amplifier within this frequency range.

We also measure the frequency response of these amplifiers since, for electrophysiology, the device bandwidth must be sufficient to amplify fast single neuronal activity (~1 kHz). To investigate the transient response, we apply 1-mV-amplitude V_in sinusoidal inputs into our amplifier circuit to represent biological signals (Fig. 2E). The corresponding low-frequency small-signal A_V of ~31 dB is consistent with that calculated above from the transistor I-V characteristics. The frequency response of the amplifier has a −3-dB bandwidth of approximately 4 kHz (Fig. 2C), enabled by the short device channel L (500 nm for the n-type OECT) owing to the vertical architecture and the choice of CP film thickness (100 nm).

In vivo neural recordings with push-pull amplifiers

For in vivo recording applications, these organic push-pull amplifiers are fashioned into a shank structure using a 2-μm flexible parylene C layer as the carrier substrate (Fig. 3A). We incorporate 16 circular amplifiers layouts, each with an overall individual recording pixel diameter of 20 μm on a 40-μm pitch to match single neuronal size and density, all while maintaining an overall device thickness of 8 μm. First, two layers of metals are vertically patterned to act as common V_DD and V_SS contacts for all the amplifiers, followed by the top deposition of individual V_out pads. p(g2T-TT) OECT channels are patterned vertically between V_out pads and a common V_DD electrode, whereas p(C₆NDI-T) channels are patterned vertically between V_out pads and a common V_SS pad. The input voltage, V_in, is applied through an external electrode, which defines the potential of the tissue. Therefore, to minimize noise coupling, we reference all potential to the surgery setup, setting the V_in to match this potential which we establish as the ground reference, shifting V_DD and V_SS to 0.3 V and −0.5 V, respectively.

Fig. 3. In vivo implantation using a push-pull amplifier array integrated into a flexible shank.

(A) An image of the flexible shank consisting of 16 push-pull amplifiers. Right inset illustrates the array layout. Amplifiers with a 20-μm overall diameter are aligned with a 40 μm pitch. (B) In vivo implantation process of the flexible shanks. Flexible shanks are inserted inside the tissue with a rigid guide. (C) Power spectra of the recorded LFPs by amplifiers (red) and passive electrodes (black) extracted from the recordings in (D). (D) In vivo activity recorded by the amplifiers (red) and passive electrode (blue). Amplifier recordings are inverted. Color plots show frequency response within the 0- to 40-Hz band. (E) Spontaneous single-unit activity recorded by push-pull amplifiers. Left side illustrates the amplifier pixels located with a 40-μm pitch. The spike waveforms acquired from corresponding amplifier sites and averaged from 100 spikes are plotted on the right site to illustrate the spatial fall-out as a function of distance. The corresponding autocorrelogram of the spike is plotted on the bottom part. a.u., arbitrary units.

We measure the Young’s modulus of these flexible amplifiers to be 800 MPa, which is significantly lower than rigid shanks that typically have moduli on the order of hundreds of GPa (fig. S6) (33). In addition, we do not observe any performance degradation in either of the n-type or p-type OECTs when subjected to a 1- to 5-mN force, which is required for the insertion of the probes into the brain tissue (fig. S7) (34).

For implantation, we perform a cranial window (2 mm by 2 mm) in the visual cortex of a wild-type (WT) mouse. The dura mater is removed to facilitate the insertion of the shank. The flexible shank is laminated on the exposed surface of the brain before the insertion with the assistance of a micromanipulator-controlled rigid guide, down to 1 mm (Fig. 3B). To be able to precisely benchmark the resolution of the recorded activity through the push-pull amplifiers, we place a commercial tungsten microelectrode next to the implantation site with a footprint (20 μm) identical to that of the amplifiers. A stainless-steel electrode, at ground potential, is placed in the cerebellum and serves both as a reference for the recording electrode and as a common gate electrode for all the voltage amplifiers. The measured voltage transients for both the active amplifiers and the recording electrode are plotted in Fig. 3, C and D. Both recordings show overlapping LFP frequency response peaking at 1 Hz, as commonly observed for slow wave sleep under anesthesia (35). As anticipated, the signal from the amplifier is more pronounced, displaying additional features at 4 and 6 Hz that are not visible in the passive recordings. The active shank recordings showcase a ~28.5 dB of voltage amplification with a 23.7 dB signal-to-noise ratio (SNR), demonstrating similar performance in vivo as observed in benchtop testing, whereas the passive electrode exhibits a significantly lower SNR of 17 dB. We further demonstrate the recording capabilities of our push-pull amplifier by extracting spike clusters at each amplifier (see Materials and Methods) (36). The detected units, with autocorrelograms typical of pyramidal neurons observed in the cortex, show undistorted waveforms with high locality. The high SNR demonstrated here, along with the ability to record neural activity from large-scale LFPs to SUA, demonstrates the high-resolution recording capability of these amplifiers.

Differential amplifier circuit architecture

A differential voltage amplifier topology is also achievable using complementary OECTs. In vivo neural signals predominantly manifest themselves as LFPs, which reflect dendritic inputs generated by synchronized synaptic activity and, therefore, lack spatial locality on a micro-scale (<100 μm) (37). We propose that differential amplification within this scale will allow the rejection of this common LFP background, leading to improved use of the amplifier dynamic range.

A differential amplifier topology, as shown in Fig. 4A, combines a depletion-mode PEDOT:PSS p-channel OECT load with a p-type enhancement-mode OECT p(g2T-TT) common-source amplifier (Fig. 4A). PEDOT:PSS OECTs show high g_m and reach saturation at positive V_GS biases as effective cation injection, and therefore, de-doping occurs in this regime (fig. S8A). p(g2T-TT) OECTs, on the other hand, exhibit a peak g_m at negative V_GS owing to electrochemical doping due to anion injection (fig. S8b). We leverage this to operate OECTs in series with complementary characteristics with a common V_in. The source of the p(g2T-TT) OECT is connected to V_SS, while the drain of the PEDOT:PSS OECT is connected to V_DD. Shifting V_in toward V_DD causes both polymers to be further electrochemically doped, while shifting V_in toward V_SS results in the de-doping of the channels. In this topology, the small-signal common-mode voltage gain is given by

$$A_{\mathrm{v}}^{\text{cm}}=\frac{g_{\mathrm{m},\mathrm{d}}-g_{\mathrm{m},\mathrm{e}}}{g_{\mathrm{m},\mathrm{d}}+g_{\text{ds},\mathrm{e}}+g_{\text{ds},\mathrm{d}}}$$ (2)

Fig. 4. Structure and operation principle of the organic differential amplifier circuit.

(A) Operation principle of the differential amplifiers. Both p-type polymers, PEDOT:PSS and p(g2T-TT), are connected in series between rail voltage pads V_DD and V_SS, respectively. At the optimized operation regime where V_DD is at negative bias, V_SS is at positive bias, and V_in is at ground potential, both polymers are partially electrochemically doped. As shown within blue brackets, shifting the V_in toward positive values leads to the dedoping of both films. As shown within red brackets, shifting V_in toward negative values leads to doping of both polymers. In either case, V_out rejects the V_in fluctuations, whereas I_D reflects the changes of V_in as a function of g_m of the OECTs. (B) Top view of the amplifier with PEDOT:PSS channel shown in blue and p(g2T-TT) channel shown in purple. Bottom image is the cross section of the amplifier. Channels for both OECTs are defined by the polymers connecting V_DD-V_out and V_out-V_SS pads vertically. (C) An image of the amplifier array is illustrated. Interconnects for V_DD and V_SS are patterned on the same layer, and V_out interconnects are patterned on a separate top layer with parylene C insulation in between them. At the cross section, PEDOT:PSS and p(g2T-TT) are patterned vertically. All channels have the same length, defined by the 500-nm parylene separator thickness, and the width of the channels is tuned by varying the width of the etched openings for PEDOT:PSS (W_D) and p(g2T-TT) (W_E) between 6 and 20 μm. Scale bar, 50 μm.

Here, g_m,e and g_ds,e represent the transconductance and conductance of the p(g2T-TT) OECT, respectively, while g_m,d and g_ds,d represent the transconductance and conductance of PEDOT:PSS film, respectively. For matching the device transconductances, $A_{\mathrm{v}}^{\text{cm}}$ is zero.

For differential operation, selective gating of the p(g2T-TT) OECT results in a differential amplifier gain given by

$$A_{\mathrm{v}}^{\text{diff}}=-\frac{g_{\mathrm{m},\mathrm{e}}}{g_{\mathrm{m},\mathrm{d}}+g_{\text{ds},\mathrm{e}}+g_{\text{ds},\mathrm{d}}}$$ (3)

whereas selective gating of the PEDOT:PSS channel results in a differential amplifier gain given by

$$A_{\mathrm{v}}^{\text{diff}}=\frac{g_{\mathrm{m},\mathrm{d}}}{g_{\mathrm{m},\mathrm{d}}+g_{\text{ds},\mathrm{d}}+g_{\text{ds},\mathrm{e}}}$$ (4)

In each case, considering a g_m>>g_ds, A_v is equal to one, leading to either an inverter with selective gating of p(g2T-TT) or a voltage follower with selective gating of PEDOT:PSS channel. Large gain cannot result from this topology if common-mode rejection is desired.

Differential amplifier operation

To test the performance of the differential amplifier design, we first generate an array of PEDOT:PSS and p(g2T-TT)–based OECTs with varying W/L ratios and identical CP thickness (100 nm) (Fig. 4, B and C). To replicate the vertical integration as we previously showed for the push-pull amplifiers, we pattern the V_DD and V_SS rail electrodes on the same bottom metal layer and the V_out pads on the top metal layer. We define the width of the etched openings for polymer coatings at various dimensions (6 to 15 μm) for further performance optimization.

We achieve overlapping current-voltage characteristics if the W/L ratio of the depletion-mode OECT is 1.2 times that of the enhancement-mode OECT (Fig. 5, A and B). The corresponding transfer curves of the OECTs show that for the differential amplifier topology with V_SS of 0 V and V_DD of −1.2 V, both OECTs have similar g_m values over a wide range of V_in operating points (0 to −0.3 V) (Fig. 5, C and D). g_ds values follow a similar trend over the same operating point range (fig. S9).

Fig. 5. Electrical characterization of differential amplifier topology.

(A) Graph shows I-V characteristics of a PEDOT:PSS-based OECT with a W/L ratio of 12/0.5 (normalized for a V_GS range of 0 to −0.6 V). The section marked in purple illustrates the linear operation regime, while the section marked in orange illustrates the saturation regime. (B) Graph shows I-V characteristics of a p(g2T-TT)–based OECT with a W/L ratio of 10/0.5 (for a V_GS range of 0 to −0.6 V). The section marked in purple illustrates the linear operation regime, while the section marked in orange illustrates the saturation regime. (C) g_m of p(g2T-TT) OECT (in purple) and PEDOT:PSS OECT (in blue) as a function of V_in for a V_DD of −1.2 V and a V_SS at ground potential, respectively. (D) I_D of p(g2T-TT) OECT (in purple) and PEDOT:PSS OECT (in blue) as a function of V_in for a V_DD of −1.2 V and a V_SS at ground potential, respectively. (E) Common-mode rejection using combined enhancement and depletion mode OECTs. Left side illustrates applied V_in (offset at −0.2 V) with various pulse amplitudes between 100 μV and 200 mV. V_out response corresponding to each V_in pulse is illustrated on the right side. The inset shows the response time of the amplifier in the case of local gating of individual OECTs. The values on each pulse illustrates the response time of the associated OECT and are calculated using an exponential decay function fit (in orange dashes).

To test the common-mode rejection capability, we apply varying amplitude V_in pulses with an offset of −0.2 V at which both transistors have a g_m of 4.5 mS (Fig. 5E). We observe that with a V_in pulse amplitude of over 10 mV, the gain exceeds −20 dB and the amplifier is not able to reject common-mode inputs. However, for V_in pulses below 10 mV, amplitudes commonly observed for LFPs, the amplifier has a significant common-mode rejection rate (>60 dB).

We then measure the gain of the amplifier for differential inputs in Fig. 5E (inset). In this case, we fabricate individual OECTs with 200-μm spacing and place stimulation electrodes within 20-μm distance to each OECT. We drive the amplifier with operation biases as determined in Fig. 5E and use the electrodes to generate local 10-mV pulses with 1-ms width. Selective gating of p(g2T-TT) OECT results in an inverter, whereas selective gating of PEDOT:PSS OECT results in a voltage follower, consistent with Eqs. 3 and 4 (fig. S9). We further characterize the response time for the amplifier; V_out reaches steady-state within 100 μs in the case of the p(g2T-TT) OECT and 20 μs in the case of the PEDOT:PSS OECT.

In vivo implementation of the differential amplifier topology

For in vivo testing of the differential topology, we generate amplifier arrays in the form of an implantable shank (Fig. 6A). We use common V_DD and V_SS electrodes patterned adjacently on the bottom metal layer and tune their potential to be −1 V and + 0.2 V, respectively, to be able to keep the external V_in electrode at the ground reference of the surgery setup as we did for the push-pull amplifer topology. We pattern 16 amplifiers in a manner identical to the previous push-pull amplifier design on the same 40-μm pitch. In this case, the geometry is rectangular, and the spacing between enhancement-mode and depletion-mode devices is 4 μm. We place a recording PEDOT:PSS coated (10-μm diameter, 100-nm thickness) Au electrode on the shank in the vicinity of the amplifiers for independent measure of the recorded activity (38–40). We implant the flexible shank as described above and measure the neuronal activity in anesthetized WT mice. We observe that the I_D flowing through both OECTs (averaged from all channels due to common V_DD and V_SS pads) matches the LFPs recorded by the passive electrode (Fig. 6B), since the respective doping level of each polymer channel is effectively controlled by the LFP modulations (Fig. 6B). V_out, on the other hand, rejects this common-mode LFP response.

Fig. 6. In vivo implantation using an array of differential amplifiers integrated into a flexible shank.

(A) An image of the flexible shank consisting of 16 differential amplifiers. A passive electrode is placed on top of the array for in vivo activity evaluation. Right inset illustrates the array layout. Amplifiers, with a 12 μm by 20 μm overall size, are placed with a 40-μm pitch. (B) In vivo V_out recorded by individual amplifiers (top), average I_D transients (inverted) recorded by the amplifiers (center), and voltage transients recorded by the passive recording electrode (bottom). Color plots show frequency response within the 0- to 40-Hz band.

To test the capability of our circuits to record very local SUAs, we fabricate an array of differential amplifiers with an extended layout (Fig. 7A). We place eight amplifiers with individual OECTs located at 100-μm spacing (Fig.7B). We also pattern an array of recording electrodes adjacent to the PEDOT:PSS OECTs to independently measure the local V_in. We implant the shank and record neural data at all electrodes and amplifiers.

Fig. 7. Single neuronal recordings using differential amplifiers in vivo.

(A) An image of the flexible shank consisting of 16 differential amplifiers and 16 passive recording electrodes. The inset illustrates the array layout. PEDOT:PSS and p(g2T-TT) OECTs of the same amplifier are located with 100-μm distance for input differentiation. (B) A cross section of the implanted shank. Interconnects for recording electrodes are patterned on the bottom metal layer. Recording electrodes are placed adjacent to PEDOT:PSS OECTs for validation of spike activity. (C) Blue traces on the left side show spike waveform of the single neuronal activity (averaged from 100 spikes over the course of 10 min) as a function of electrodes placed with 40-μm pitch. Average waveforms acquired from the bottom electrode and the adjacent amplifier at the same time points are plotted on the right side.

We first spike-sort the recordings from passive electrodes (see Materials and Methods). An identified neuron is in proximity of the bottom recording electrode, and the amplitude of the recorded signal decays within 70 μm (Fig. 7C). Subsequently, we plot the V_out of the differential amplifier that is located adjacent to the bottom electrode at the times of detected spikes. These curves exhibit similar spike-like waveforms with the same polarity as the electrode recordings. The measured circuit noise within the 0.15-Hz to 3-kHz frequency range is 10.2 μV root mean square (RMS), comparable to electrode recordings after high-pass filtering at 300 Hz (7 μV RMS).

DISCUSSION

In this work, we demonstrated that OECT devices (n-channel and p-channel, enhancement-mode and depletion-mode) to realize voltage amplifiers in an area efficient vertical device architecture. The proposed models bring the advantage of implementing voltage-to-voltage conversion directly at the front-end recording pixel for in vivo applications. The push-pull amplifier leads to a A_v in the range of 31 dB, allowing electrophysiological signals to be amplified directly at the electrode site, before they experience the coupling noise associated with long interconnects down the recording shanks (41). In addition, the differential amplifier enables common-mode LFPs to be rejected with electrode spacings on the order of ~100 μm while providing limited gain for local single-unit and multi-unit activity. Such differential amplifiers may also find application in sensing in which one of the two electrodes can be selectively functionalized for differential detection of species important for investigations of brain metabolism, such as lactate or glucose (42, 43). In particular, enhancement mode OECTs have been shown to detect such species with performance that exceeds that of the state of the art but have yet to be integrated in vivo (44, 45). Future integration of such strategies to the differential amplifier circuit would allow in vivo biomolecular sensing with a rejected common neural response. In both cases, front-end voltage conversion allows the integration of these amplifiers with conventional large-scale recording systems, which are based on acquiring voltage. This bypasses the need to have external electronics such as transimpedance amplifiers to generate voltage outputs, which is usually the case for single-element OECTs as pre-amplifiers.

A significant challenge in the spin coating of solution-processable electronics is achieving precise control over polymer thickness to enable complementary operation between OECTs. We anticipate that in the future, these challenges could be addressed by depositing polymer solutions onto wafers using more precise techniques, such as inkjet or screen printing (46, 47). This innovation has the potential to enhance the uniformity of coatings on a wafer level, enabling large-scale implementation of these complementary circuits. A secondary limitation lies in the comparatively lower performance and stability of newer n-type polymers. p-Type devices, on the other hand, have a well-established track record and have been used in neural interfaces for over a decade. We expect that ongoing advancements in the field, including the development of new-generation of n-type polymers, will help mitigating this gap in the performance.

Overall, we show that enhancement mode n-type and p-type CPs, in an OECT configuration, can be tuned to work in accordance with the existing state-of-the-art depletion mode p-type modalities. The amplifiers generated with complementary OECT operation allow on-site amplification and signal conditioning of neuronal signals, which is otherwise not possible with single-element OECTs, and pave the way for in vivo integration of active organic electronics.

MATERIALS AND METHODS

Device fabrication

A 2 μm of parylene C layer, coated over a 75-mm glass wafer through chemical vapor deposition (CVD) (Specialty Coating Systems coating), is used as a carrier substrate for all the amplifier arrays. To pattern the bottom layer gold, a lift-off process is followed. A 1-μm-thick film of LOR3A and a 1-μm-thick film of S1811 are spin-coated and cured (130°C for 5 min and 110°C for 2 min, respectively). The cured photoresist stack subsequently is exposed to ultraviolet light using a Suss MA6 Mask Aligner or, for finer resolution patterns (<2 μm), S1811 is exposed directly using a Heidelberg DWL66 Laser Writer. Photoresist was later developed with AZ300MIF (MicroChemicals, Merck) developer for 30 s. The wafer is then transferred to an Angstrom EvoVac Multi-Process Evaporator to complete the deposition of a 10-nm-thick titanium (Ti) adhesion layer, followed by a 100-nm-thick gold (Au) layer. Last, the wafer is soaked in remover PG (Kayaku Advanced Materials Inc., Westborough, MA) to complete the base metal lift-off. As insulation between vertical metal layers stacks, a second layer of parylene C film (0.5 to 2 μm depending on the amplifier structure) was deposited using CVD using 3-(trimethoxysilyl) propyl methacrylate (A-174 silane, Sigma-Aldrich) as an adhesion promoter. The top metal layer is deposited and patterned as described above. As insulation over the overall device to minimize the effects of electrolyte, a thick layer of parylene C (2.2 μm) is deposited. To achieve polymer patterning between vertical metal stacks, a sacrificial parylene C layer (2 μm) is deposited following a spin-coating step of soap solution diluted to 1% concentration in deionized water. To define the polymer patterns, an 8-μm-thick layer of AZ10XT (MicroChemicals, Merck) is spin-coated at 5000 rpm, baked at 115°C for 2 min, exposed using a Suss MA6 Mask Aligner, and developed with AZ400K developer (MicroChemicals, Merck). A dry etch process is performed for 15 min in a plasma-reactive ion etcher (Oxford Plasmalab 80) with parameters of 180 W, 60 standard cubic centimeter per minute (SCCM) O₂ and 2 SCCM SF₆ till the bottom gold layer. Depending on the amplifier design, a solution of p(C₆NDI-T) dissolved in chloroform (1 mg/ml), or an aqueous dispersion of PEDOT:PSS is spun at 500 to 3000 rpm for 45 s. PEDOT:PSS solution is prepared using 95 wt % PEDOT:PSS (Clevios PH1000) mixed with 1 wt % (3-glycidyloxypropyl) trimethoxysilane (Sigma-Aldrich), 0.1 wt % 4-dodecyl benzene sulfonic acid (Sigma-Aldrich), and 4.9 wt % EG (Sigma-Aldrich). The sacrificial layer of parylene C is carefully peeled-off to complete polymer patterning in-between top and bottom metal layers to define the vertical OECT channels. A final layer of sacrificial parylene C is deposited following a spin-coating step of soap solution diluted to 1% concentration in deionized water to achieve p(g2T-TT) patterning. The top parylene C layer is then patterned, as described above, to define p(g2T-TT) OECT channel openings. p(g2T-TT), dissolved in chloroform (5 mg/ml), is spun at 500 to 3000 rpm for 45 s. The sacrificial layer of parylene C is carefully peeled-off to complete p(g2T-TT) patterning in-between top and bottom metal layers to define vertical OECT channels that are adjacent to the previously patterned OECT channels. To make the arrays in the shank format, the contour of the shanks was patterned by using an IPG Photonics excimer laser cutter.

In vivo experiments

Our laboratory, where animal experiments are conducted, and the animal housing facilities have been subject to inspection by The Institutional Animal Care and Use Committee (IACUC). The IACUC has carefully reviewed and granted approval for the animal protocols used in this research, which is conducted under the project name AC-AABE5554 (development of high-density, implantable recording, imaging, and stimulating arrays). For the implantation procedure, WT mice (Strain: 00664, obtained from The Jackson Laboratory) were used. Surgery protocol is explained elsewhere (38). Once the cranial opening is performed, the flexible amplifiers were then laminated onto the brain. To facilitate the insertion process, a borosilicate glass pipette (1.5-mm diameter) was prepared using a Sutter Instrument Model P-97 to have an outer tip diameter of 10 μm. This glass carrier pipette was used to insert the flexible shank via a through-hole at the tip of the shank.

OECT geometry optimization

The key to achieving reproducible amplifiers lies in optimizing the geometry of the OECTs. A critical factor is maintaining the dimensions of length (L) and width (W) at their specified, optimized parameters. Before fabricating each device, we conducted a thorough characterization of the parylene thickness and proceeded with fabrication accordingly to ensure that we achieved the desired thickness. Thickness mapping was performed by coating parylene (via CVD) and or spin-coating semiconducting polymer films onto silicon wafers at target thicknesses of 2-μm and 100-nm, respectively. These samples were considered as air/polymer film/Si substrate stacks, with initial estimates for the thicknesses and refractive index values of the polymer films. Reflectance measurements were taken at normal incidence across a wavelength range of 500 to 900 nm using a calibrated X-Prep Vision reflectometer from Allied High Tech Products Inc. A motorized stage translated the sample in 5-mm increments, creating a grid of thickness values over the wafer surface. Polymer film thicknesses were determined from either a fast Fourier transform of the reflectance spectra and by fitting curves. Repeated measurements at the same locations showed no variation, and the results were consistent across multiple wafers (n = 4).

We observe that parylene thickness shows an incremental increase along the wafer axis (fig. S2, A and B). Therefore, we took a meticulous approach by carefully situating all the active transistors within a ~1-cm-diameter zone. This strategic placement helps us avoid variations in device performance that could result from the thickness gradient. Similarly, the thickness of the polymer coating can pose particular challenges due to the usage of spin-coating process. Both polymers are dissolved in chloroform, which evaporates rapidly during the spinning step. In addition, the low wetting properties of the parylene coating hinders the homogeneous deposition of the polymers over a large scale. To circumvent these challenges, we chose the 1-cm zone to be in the exact center of the wafers. This central region has demonstrated a high level of uniformity, as depicted in fig. S2 (C and D).

Amplifier characterization

All current-voltage and current-time measurements are done using an electrolyte (Phosphate Buffered Saline, 1× solution purchased from Fisher BioReagents). Current-voltage curves are measured using a semiconductor device analyzer (B1500A, Agilent Technologies). An Ag/AgCl electrode (1 mm by 3 mm, purchased from World Precision Instruments) is used as the gate external electrode. Current-time measurements are performed using an oscilloscope (DSO-X 4034A, Agilent Technologies). The time constant of the amplifiers was extracted by measuring the current response and fitting the current-time profile with a single exponential decay equation fit. For amplifier measurements, V_DD and V_SS biases were applied by a power supply (HP E3620A) connected to a custom-made voltage regulator circuit. V_in pulses were applied, and V_out pulses were recorded by using an oscilloscope (DSO-X 4034A, Agilent Technologies).

Mechanical testing

To measure the Young’s modulus of the flexible shanks, we used a TA Instruments DMA 850 equipped with a film tension clamp. We applied an oscillating load of 0.25 N along with a 0.25 N preload and conducted a frequency sweep ranging from 1 to 10 Hz. For the analysis of the impact of insertion force during neural implantations, we gradually applied a load to the shank, ranging from 0 to 5 mN, over the course of 1 min.

Electrophysiology recordings and analysis

All device characterization and bias optimization were carried out using a phosphate-buffered saline solution. However, we noticed a gradual increase in solution resistance, particularly during the placement of the shanks on the surface of the brain and their subsequent insertion inside the brain, as shown in fig. S10. This increased ionic resistance inside the brain can lead to minor shifts in the optimized device characteristics. To ensure that we remained within the complementary regime, the biasing of the differential amplifier operation was fine-tuned for each in vivo experiment by adjusting V_DD within ±0.1-V range depending on the V_out signal.

To facilitate the comparison between our amplifiers and standard electrodes, we purchased passive tungsten electrodes (Microprobes for Life Sciences) that are commonly used for neuroscientific studies with overlapping footprint of our amplifiers. These electrodes were placed on the same side as the implanted probe during surgery. We connected all the amplifiers and electrodes to the same Intan RHS2000 amplifier to ensure that the recorded activity was exposed to the same external noise, making it directly comparable. All the electrophysiology data were acquired using the Intan system with a sampling rate of 30 kHz. Electrophysiology data were analyzed separately for spike and LFP information. The data were acquired over the range of 0.15 Hz to 7 kHz, and Butterworth was filtered between 300 Hz and 3 kHz for spike detection. Single neuronal spikes were sorted using Kilosort 2.5 and plotted with custom MATLAB code to visualize waveforms at the times of stimulation. The LFP bands were analyzed with a custom-written MATLAB code to extract LFP time-frequency information using wavelet transformations at different frequency bands upon application of a Butterworth low-pass filter with a corner frequency of 500 Hz.

Supplementary Materials

This PDF file includes:

Figs. S1 to S10