Skip to main content
NCBI home page
Science Advances logoLink to Science Advances
. 2026 Mar 6;12(10):eaeb5852. doi: 10.1126/sciadv.aeb5852

Flexible amplifier with >100-dB voltage gain enabled by intrinsic gain singularity of carbon nanotube transistors

Ruyi Huang 1,2,†, Yuru Wang 1,†, Guanhua Long 1,†, Zirui Zhang 1,3, Tingzhi Wang 1, Zheng Fang 1, Yufan Chen 1, Wanyi Wang 1, Tianshun Bai 1, Meiqi Xi 1, Xiaowei He 1,3, Xuelei Liang 1, Shiwei Wang 4, Lian-Mao Peng 1,2,*, Youfan Hu 1,2,3,*
PMCID: PMC12965310  PMID: 41790893

Abstract

For biointegrated flexible systems that acquire and process electrophysiological signals, amplifying weak biosignals from their original low amplitudes (ranging from microvolt to millivolt) to volt-level is essential for subsequent processing. Achieving this level of amplification requires a high voltage gain (>105 or 100 decibels for microvolt signals). However, realizing such gain in flexible circuits remains highly challenging because of constrained integration scale and limits in feasible circuit topologies. Here, we report flexible amplifiers that achieve ultrahigh gain by leveraging intrinsic gain singularities induced by negative differential resistance (NDR) effect in carbon nanotube–based transistors. The NDR behavior is investigated under various factors, including contacts, gate structures, and channel lengths. Guided by insights into the correlations between NDR characteristics and device-level parameters, a device-circuit codesign approach is implemented to build a flexible amplifier achieving a record-high gain of 104 decibels among all reported flexible amplifiers, with successful demonstration of electroencephalogram signals amplification.

A carbon nanotube–based flexible amplifier achieves >100-dB voltage gain, demonstrating EEG signal amplification.

INTRODUCTION

High-fidelity monitoring of surface electrophysiological signals generated by biological processes, typically involving the cardiac, muscular, or nervous systems, has broad and impactful application potential in disease diagnosis, early risk detection, and personalized health management (1–4). These signals, including electrocardiograms, electromyograms, electroencephalograms (EEGs), and others (5–7), are typically characterized by low amplitudes. In particular, EEG signals are exceptionally weak, down to several to tens of microvolts (8), for which a voltage gain >100 dB is required to bring the signals to volt-level to meet subsequent signal processing requirements (Fig. 1A) (9).

Fig. 1. Flexible amplifier for weak electrophysiological signal amplification.

Fig. 1.

Open in a new tab

(A) Conceptual illustration of amplifying weak electrophysiological signal via a high-gain flexible amplifier for subsequent signal processing. (B) Schematic illustration and (C) optical microscope image of a flexible CNT-TFT. Scale bar, 10 μm. (D) Scanning electron microscopy image of the CNT networks in the channel. Scale bar, 200 nm. Photograph of flexible CNT-TFTs attached to the surface of (E) a cortical phantom and (F) a badge. Scale bars, 2 cm (E) and 5 mm (F). (G) Transfer characteristics of a flexible CNT-TFT before and after 40, 100, and 200 bending cycles at a fixed radius of 2.5 mm.

In silicon-based rigid chips, amplifiers with >100-dB voltage gains can be achieved using complex circuit topologies such as multi-cascode configurations and bootstrapping techniques (10). However, implementing these circuit topologies in flexible amplifiers to achieve comparable high gain is very challenging because of the limited integration scale of flexible integrated circuits (ICs) and relatively large process variations of flexible transistors compared with mainstream silicon technologies. Unique behaviors exhibited in emerging material-based transistors not seen in conventional silicon devices offer opportunities for innovative solutions. Intrinsic gain reflects the maximum possible voltage gain achievable from the transistor in an ideal configuration. In an emerging material-based transistor exhibiting negative differential resistance (NDR) effect, theoretically, intrinsic gain singularities with infinite gain can be obtained at the negative-to-positive difference resistance (NDR/PDR) transition boundaries (Av = gm/gds → ∞ when gds → 0, where gm is transconductance and gds is output conductance). The existence of singularities makes it possible to achieve ultrahigh gain using simple circuit architectures if the singularity can be properly leveraged in the circuit. Previous work has demonstrated that by exploiting the NDR effect in negative-capacitance (NC) transistors, a flexible amplifier with a gain up to 9.7 × 103 (≈79.8 dB) (11) can be realized using a simple common-source topology. However, the NC transistors are limited by their inherent hysteresis and instability (12, 13).

Carbon nanotubes (CNTs) are superior for flexible electronics, distinguished by their exceptional intrinsic properties, including an ultrathin body, high carrier mobility, outstanding mechanical properties, and compatibility with solution-based processing (14–17). Recently, ambipolarity-based inherent NDR effect has been demonstrated in flexible CNT thin-film transistors (CNT-TFTs) (18), where the NDR effect is induced by competitions between two types of carriers in the channel. Comparing to previously reported ambipolarity-based NDR effect observed in graphene (19) and black phosphorus (20), due to CNT’s appropriate bandgap, symmetric band structure, and immunity against Fermi pinning, a substantially larger NDR window and an extended NDR/PDR transition boundary are obtained in CNT-TFTs, which facilitates biasing the device operation point near the intrinsic gain singularities and makes circuit-level applications practical. Flexible operational amplifier with single-stage gain up to 60 dB was demonstrated. As the ambipolarity-based NDR effect is governed by the competition between two types of carriers, which theoretically can be modulable by various device-level factors, device-circuit codesign emerges as a viable strategy to exploit the tunable NDR characteristics and intrinsic gain singularities for realizing ultrahigh-gain flexible amplifier, specifically targeting application-driven requirements of >100 dB. However, research on the correlations between NDR behavior and device-related parameters remains scarce.

In this work, we systematically investigated the NDR behavior and its impact factors in CNT-TFTs. Through controlled experiments, the key role of contacts, gate structures, and channel lengths (Lch) in controlling NDR characteristics was elucidated. The tunable NDR behaviors enabled a device-circuit codesign strategy to leverage the intrinsic gain singularity in a flexible CNT amplifier that achieved a record-high gain of 1.5 × 105 (≈104 dB) and demonstrated amplification of EEG signals.

RESULTS

Flexible CNT-TFTs with NDR effect

Bottom-gate CNT-TFTs (Fig. 1, B and C) based on high-purity semiconducting CNT networks (Fig. 1D) are fabricated on 2-μm-thick polyimide (PI) substrates. Palladium (Pd), with its high work function, is used as the contact metal to form ohmic contact for hole injection, thus achieving p-type devices. Yttrium oxide (Y2O3) is selected as the gate dielectrics due to its excellent wettability with CNTs and high dielectric constant, providing a good interface and high gate-control efficiency (21, 22). The detailed fabrication process of flexible CNT-TFTs is shown in fig. S1. Following fabrication, the devices on the PI foil were delaminated using previously reported capillary-assisted electrochemical delamination method (23). The obtained ultrathin electronic foil demonstrates a high degree of conformability and transparency (Fig. 1, E and F). Furthermore, the device’s electrical performance remains nearly unchanged under cyclic bending (Fig. 1G and fig. S2) and different bending radii (fig. S3), confirming the excellent mechanical flexibility.

Figure 2A illustrates the transfer characteristics of a typical flexible CNT-TFT with an Lch of 5 μm and a channel width (Wch) of 50 μm. At a drain-source voltage (Vds) of −0.1 V, the device exhibits an on-off current ratio exceeding 106 and a subthreshold swing of 93 mV/dec, with nearly negligible hysteresis (fig. S4A). The device achieves a maximum width-normalized on-state current exceeding 7 μA/μm and good uniformity (fig. S4, B and C), competitive with state-of-the-art flexible transistors (17, 24, 25). As shown in Fig. 2B, in the subthreshold region, when turning the device from off state to on state with decreased gate-source voltage (Vgs), the output curves transit from N-shaped three-section NDR curve, to Λ-shaped two-section NDR curve, and finally to a normal PDR curve. This gate-modulated NDR effect is consistent with our previous report (18) with large NDR window (the Vds range of the NDR section) in the gds plot, as shown in Fig. 2C. At the transition boundary between the NDR and PDR zone, gds of the transistor approaches zero, giving the intrinsic gain singularities, which is the basis for its application in high-gain amplifiers. Specifically, the upper NDR/PDR transition boundaries, which are distinguished as current supersaturation region, are more favorable for exploiting in circuit design.

Fig. 2. Electrical performance and underlying mechanism of NDR behavior of flexible CNT-TFTs.

Fig. 2.

Open in a new tab

(A) Transfer characteristics of a representative flexible CNT-TFT with Lch = 5 μm and Wch = 50 μm. (B) Output characteristics of the CNT-TFT in/around the subthreshold region. The gray arrow indicates the sweep direction of Vgs. (C) Two-dimensional (2D) color plot of output conductance gds as a function of Vds under different Vgs of the CNT-TFT. The NDR/PDR transition boundary is indicated by gray dashed lines. (D) Schematic band diagrams illustrating the hole depletion and conduction type inversion near the drain side with increased |Vds|. (E) Output characteristics of a CNT-TFT (Lch = 20 μm) at Vgs = −5 V (top) and the false-color near-infrared EL intensity images at corresponding Vds (bottom). Scale bar, 5 μm.

The NDR behavior in CNT-TFTs primarily originates from minority carrier inversion near the drain side and determined by the competition between the two types of carriers in the channel. As shown in Fig. 2D, for p-type devices in subthreshold region, increasing |Vds| raises the drain-side energy band, which pulls the Fermi level into the bandgap at certain voltage level and forms a hole-depleted, charge neutrality region with high resistance near the drain side. With further increased |Vds|, the barrier for electron injection from the drain side is thinned. This enables minority electron injection and establishes conduction type inversion near the drain side, pushing the charge neutrality region toward the source side and forming a p-i-n configuration in the channel. Initially, the electron density is low, creating a high-resistance n region that replaces the previous low-resistance p region and resulting in the NDR. As |Vds| increases further, the electron density and distribution expand, reducing the overall channel resistance and leading to the reappearance of PDR. The modulated p-i-n configurations (i.e., carrier distribution in the channel) underpins the emergence and evolution of NDR behavior.

Evolution of carrier distribution in the channel

In these in-channel p-i-n–configured devices, the coexisting holes and electrons can form Coulomb-bound excitons that generate electroluminescence (EL) through radiative recombination (26, 27). Figure 2E demonstrates this phenomenon. The modulated p-i-n configurations with their corresponding NDR behavior were visualized through the dynamic evolution of the near-infrared EL intensity images synchronized with the device’s output characteristics. At minimal |Vds|, no EL is detected because of negligible minority carrier (electron) injection at the drain side. With increasing |Vds|, localized EL emission emerges near the drain side and intensifies progressively, suggesting enhanced electron injection at the drain side and the formation of a hole-depleted intrinsic region as represented by the emission zone where the electron-hole recombination process mainly occurs. The further increase of |Vds| transitions the device from PDR to NDR regime, accompanied by a migration of the EL emission zone from the drain side to the source side. This indicates the migration of the intrinsic region as electrons gradually penetrate into the channel and compress the original hole-rich region, which is consistent with the evolution of carrier distribution and p-i-n configuration in the channel predicated by our interpretation of the NDR mechanism. Last, with sufficiently large |Vds|, the EL emission zone reaches near the source side, confirming the transition to electron-dominated transport in the channel after NDR. The above observations offer direct insight into the correlations between the evolutions of NDR behavior and in-channel carrier distribution.

Modulation of NDR behavior

Since carrier injection influences the carrier distribution in the channel, the contact plays a crucial role in shaping the NDR behavior. To investigate this effect, titanium (Ti) is used as a replacement for Pd due to its lower work function of 4.33 eV (28), which lies between the conduction and valence bands of CNTs. This alignment leads to the formation of Schottky barriers at the contacts for both holes and electrons injection. As shown in Fig. 3A, the Ti-contacted device exhibits inferior on-state performance compared to its Pd-contacted counterpart, attributed to the presence of injection barriers at the contacts (fig. S5). Regarding NDR behavior, the output characteristics of the Ti-contacted device (Fig. 3B, with statistical results in fig. S6) show a pronounced increase in the number of N-shaped curves, along with substantially shorter NDR segments on the Vds scale and an expanded distribution on the Vgs scale compared to Pd-contacted devices (Fig. 3C). This behavior is attributed to the enhanced ambipolar characteristics induced by the Ti contacts. As illustrated by the energy band diagram in Fig. 3D, compared to the device with Pd contacts, the device with Ti contacts has a higher injection barrier for holes at the source and a lower injection barrier for electrons at the drain, which suppresses hole injections while facilitating electron injection. In Ti-contacted devices, the competition between these two types of carriers becomes more balanced, enabling the in-channel p-i-n configuration—responsible for NDR effect—to be maintained over a wider Vgs range. At the same time, the balanced carrier injection promotes the switching of dominant conduction type in the channel, resulting in a narrower NDR distribution on the Vds scale. Adjusting the work function of contact metal can effectively alter the relative injection of electrons and holes. Allowing one type of carrier injection to dominate tends to broaden the NDR window, while more balanced carrier injection expands the Vgs range over which the NDR persists.

Fig. 3. Modulation of NDR behavior through contact, gate structure, and Lch.

Fig. 3.

Open in a new tab

(A) Transfer characteristics of a representative Ti-contacted CNT-TFT with Lch = 5 μm and Wch = 50 μm. (B) Output characteristics of the CNT-TFT in/around the subthreshold region on a logarithmic scale. (C) Comparison of 2D color plots of output conductance gds for Pd-contacted and Ti-contacted CNT-TFTs. The area outlined with the red dashed line represents the NDR zone of the Pd-contacted CNT-TFT. (D) Schematic band diagrams of Pd- and Ti-contacted devices illustrating the mechanism of contact-dependent NDR behavior. (E) Schematic illustration of CNT-TFTs with air-gap and overlapping gate structures. (F) 2D color plots of output conductance gds for CNT-TFTs with air-gap and overlapping gate structures. The area outlined with the red dashed line represents the NDR zone of the CNT-TFT with overlapping gate structure. (G) Comparison of PVR versus Vgs for CNT-TFTs with different gate structures. (H) Schematic band diagrams of CNT-TFTs with air-gap and overlapping gate structures illustrating the mechanism of gate structure–affected NDR behavior. (I) 2D color plots of output conductance gds for CNT-TFTs with different Lch of 5, 10, and 20 μm. (J) Comparison of NDR window and (K) PVR for CNT-TFTs with different Lch. The subthreshold region is marked with a dashed box in (J). (L) Schematic band diagrams of CNT-TFTs with different Lch illustrating the mechanism of Lch-dependent NDR behavior.

The gate structure can also notably influence the gate-controlled NDR behavior. As shown in Fig. 3E, two kinds of gate structure were investigated: air-gap gate structure (gate lengths Lg = 5 μm, Lch = 15 μm) and overlapping gate structure (Lg = 20 μm, Lch = 15 μm). Two key metrics—the NDR window on the Vds scale and the peak-to-valley current ratio (PVR)—are analyzed to evaluate NDR behavior with different gate structures (fig. S7A). Compared to the overlapping-gate device, the air gap gate device exhibits a shrunk NDR region (Fig. 3F, with output characteristics shown in fig. S7, B and C) and decreased PVR (Fig. 3G), with statistical results provided in figs. S8 and S9. The PVR also shows a monotonic decrease with increasing Vgs and gradually approaches 1, which corresponds to the device approaching the current supersaturation. The air-gap gate structure, where the gate only partially covers the channel, results in limited gate control over the channel, forming ungated, high-resistance area that takes more voltage drop (band diagrams in Fig. 3H). Therefore, the voltage drop near the drain side is reduced, suppressing the minority carrier injection and the inversion of the conduction type, leading to a weakened NDR. The modulation of the electric field distribution by the gate structure substantially affects the NDR behavior. The enhanced gate control efficiency enables a larger NDR zone, resulting in a wider upper NDR/PDR transition boundary, i.e., a wider current supersaturation region given the same |Vds| range.

We further investigated the scaling effect of the NDR behavior by comparing devices with different Lch of 5, 10, and 20 μm at a fixed Wch of 50 μm. As shown in Fig. 3I, the results reveal that devices with different Lch exhibit similar NDR patterns; however, as Lch increases, the NDR region shrinks on the Vgs scale. |gds| is also lowered in the NDR zone due to a notable reduction in device current (gds ∝ Id, fig. S10, with statistical results in figs. S11, S12, and S13). In the subthreshold region (marked with a dashed box in Fig. 3J), corresponding to the near-on state, the NDR window gradually narrows as the Lch increases. Meanwhile, at a fixed Vgs, the PVR continuously decreases with increasing Lch (Fig. 3K). The band diagrams in Fig. 3L elucidate the underlying mechanism. Under the same Vds, devices with shorter Lch exhibit lower channel resistance, resulting in a reduced voltage drop across the channel and a correspondingly larger voltage drop at the contact. Consequently, the injection barrier at the drain side is more substantially thinned. This facilitates electron injection, substantially enhances inversion near the drain side, and thereby intensifies the NDR effect. Although Lch affects the NDR characteristics, the overall shape of the NDR zone is more strongly governed by the contact and gate structure. Variations in Lch primarily fine-tune the position of the upper NDR/PDR transition boundary and adjust the width of the NDR window.

Device-circuit codesign for flexible CNT-based amplifier with >100-dB voltage gain

The above investigations provide insights into NDR modulation mechanisms, enabling a device- circuit codesign strategy to leverage the tunable NDR effect for ultrahigh-gain amplifier. Because the NDR characteristics substantially modify the transistor’s output resistance, the core of this codesign framework is to adopt a circuit topology in which the voltage gain is predominantly determined by the output resistances of the NDR transistors (text S1). The circuit topology establishes the relationship between voltage gain and key device parameters, thereby defining targets for device-level optimization, such as the usable Vds window, the position of the NDR/PDR transition boundary, and the corresponding bias conditions, which can be achieved through contact engineering, structural design, and sizing optimization of the transistor.

To prove the codesign concept, the single-stage common-source amplifier is used for simplicity. Since the NDR/PDR transition exists in subthreshold region, a zero-Vgs connected transistor (29) is used as the load device (Fig. 4A). Such circuit topology provides a voltage gain predominantly determined by the output resistance of the devices (text S2 and fig. S14) and a self-defined bias configuration of Vgs = 0 for the load device, setting the operation current of the amplifier. Device-level coengineering can modulate the NDR behavior so that the NDR/PDR transition boundary falls around the zero-Vgs bias conditions.

Fig. 4. Flexible amplifier with >100-dB voltage gain.

Fig. 4.

Open in a new tab

(A) Circuit diagram and optical microscope image of the zero Vgs-load amplifier based on flexible CNT-TFTs. Scale bar, 20 μm. (B) Output characteristics of a CNT-TFT illustrating the NDR-to-PDR transition, with the supersaturation curve that produces intrinsic gain singularities adjusted to be around Vgs = 0 V via device-level design. (C) Output characteristics and (D) gain characteristics of the flexible high-gain amplifier. (E) Comparison of the gain of flexible single-stage amplifiers versus supply voltage (VDD). For references, see table S1. (F) Playback EEG signals and their amplified outputs from the amplifier (top), along with the corresponding power spectral density (bottom).

In our case, the Pd contact, overlapping gate structure are adopted, the Lch of transistor is sized as 10 μm, and the threshold voltage is tuned by adjusting the work function of the gate metal (30). As shown in Fig. 4B, with device-level design, the supersaturation curve that produces intrinsic gain singularities can be observed at around Vgs = 0. Using the gain singularities, the ultrahigh-gain characteristics is confirmed in multiple amplifiers (Fig 4, C and D, and fig. S15) with negligible degradation under bending (fig. S16), which achieve a record-high voltage gain up to 1.5 × 105 (≈104 dB) under an operation voltage of 5 V, surpassing all reported flexible amplifiers (Fig. 4E, see table S1 for details) (11, 15–17, 24, 29, 31–44). For comparison, we also implemented several non-codesigned amplifiers, including an amplifier based on the same circuit topology but constructed with Ti-contacted devices and amplifiers constructed with Pd-contacted devices but without structural optimization of devices or based on different circuit topologies, all of which exhibited inferior gain performance (fig. S17). A statistical comparison specifically among flexible zero-Vgs amplifiers with different configurations is presented in fig. S18. These results underscore the critical role of device-circuit codesign in advancing future NDR-based circuit implementations. The obtained more than 100-dB gain satisfies the critical gain requirements for amplifying weak signals, such as EEG. A playback signal derived from an EEG dataset (45) was applied to the amplifier, and the amplified output was subsequently filtered to remove power frequency interference (50 Hz), with the results illustrated in Fig. 4F (top). The input signal amplitude was as low as 100 μV, yet it was successfully amplified to approximately 0.1 V, meeting the requirements for subsequent volt-level signal digitization and processing. The power spectral density of the recorded EEG signals is shown in Fig. 4F (bottom). Notably, as the dataset was acquired from subjects in an awake, eyes-open state, the beta rhythm (12 to 30 Hz) exhibits substantially enhanced power, which aligns with the expected physiological characteristics.

DISCUSSION

High-gain flexible amplifiers play a vital role in advanced bioelectronics by enabling the highly sensitive and stable acquisition of weak physiological signals. These capabilities provide essential technological support for next-generation wearable medical devices, continuous real-time health monitoring systems, and other emerging biomedical applications that demand both high fidelity and long-term operational reliability. Here, we developed ultrahigh-gain flexible amplifiers based on intrinsic gain singularity induced by the NDR effects in CNT-TFTs. The systematic investigation reveals that the NDR modulation behavior and underlying mechanism via device-level parameters, such as contact-metal work function, gate efficiency, and device dimensions. Through investigations of the tunability of NDR, we used a device- and circuit-level codesign strategy to properly leverage the intrinsic gain singularity in amplifiers, ultimately delivering an unprecedented gain of 104 dB, outperforming all previously reported flexible voltage amplifiers. We demonstrated amplification of playback EEG signals, providing a basis for in situ biosignal amplification within a fully flexible platform that enables high-fidelity and long-term physiological monitoring.

Next, addressing key technical challenges, such as compatibility at the recording electrode-skin interface, effective noise suppression, and robust circuit design, provides a viable pathway toward stable in situ demonstration. The subsequent circuit design incorporates negative feedback techniques to stabilize operation point and extend linearity. The monolithic integration of ultrahigh-gain flexible amplifiers with physiological signal acquisition electrodes represents a critical step toward minimizing signal transmission losses and enhancing system integration density. Here, insights into the modulation mechanisms of NDR effects through device-level design, together with the demonstrated device-circuit cooptimization strategy and previously revealed material property–based modulation mechanism (18), establish a solid foundation for the future development of NDR-based functional circuits that can be extended to other low-dimensional ambipolar semiconductors. This work introduces an innovative technological pathway for ultrahigh-gain amplification of weak physiological signals, highlighting its broad application potential in wearable health monitoring, implantable neural interfaces, and next-generation biosensing electronics.

MATERIALS AND METHODS

Fabrication of ultrathin PI substrates

Following an initial oxygen plasma treatment (250 W, 60 s) of heavily phosphorus-doped silicon wafers, a PI precursor solution (3022 L, POME Sci-Tech Company), diluted in N-methyl-2-pyrrolidone to a solid content of 8 wt %, was spin-coated onto the wafer surface at 4000 rpm for 30 s. The PI film was then thermally cured in an oven with a ramping rate of 2°C/min up to 300°C, followed by a 1-hour isothermal hold at the temperature. Subsequently, a 3-nm-thick HfO2 layer was deposited on the cured PI surface via atomic layer deposition at 90°C to enhance adhesion between the metal layer and the flexible substrate.

Preparation of Arc-CNT thin films

High-purity semiconducting CNT solution was prepared by dispersing 100 mg of raw arc-discharged single-walled CNTs (SWCNTs) (Carbon Solutions Inc.) and 200 mg of 9-(1-octylnonyl)-9H-carbazole-2,7-diyl (PCz) in 100 ml of chloroform. This mixture was ultrasonicated using a top-tip dispergator (VC500, Sonics) for 30 min (50% amplitude level). To purify the dispersion, sequential centrifugation was performed: initially at 20,000g for 1 hour (Allegra X-22R centrifuge) to remove most of the bundles and insoluble impurities, followed by a second centrifugation step at 50,000g for 2 hours to eliminate residual metallic nanotubes. The upper 90% of the resulting supernatant, enriched in semiconducting CNTs, was collected for thin-film fabrication. CNT thin films were then fabricated using a dip-coating machine. A PI substrate mounted on a silicon wafer support was vertically fixed above the purified CNT solution. The substrate was repeatedly dipped into and withdrawn from the solution at a controlled speed. After completing 20 such cycles, a uniform and randomly oriented CNT network was formed, achieving a density of ~30 tubes/μm.

Fabrication of flexible CNT-TFTs

CNT-TFTs were fabricated using ultraviolet photolithography (EVG610, EV Group) for device patterning. A bottom-gate electrode stack consisting of 0.3-nm Ti (adhesion layer) and 8-nm Pd was deposited via electron-beam evaporation. Subsequently, a 10-nm Y layer was deposited and thermally oxidized at 250°C for 30 min to form a 13-nm-thick Y2O3 gate dielectric. CNT films were then dip-coated onto the substrate. Unwanted CNTs outside the active channel region were selectively removed using inductively coupled plasma etching (Minilock, Trion). Last, 30-nm-thick Pd layers were deposited to form the source and drain contacts on the patterned CNT channel.

Electrical characterization of CNT-TFTs and amplifiers

The electrical characterization of CNT-TFTs and amplifiers was performed under ambient conditions. DC parameters were measured using a semiconductor parameter analyzer (Keithley 4200A-SCS, Tektronix) with probe station (Summit 11000, Cascade Microtech).

Preparation of (10, 5) CNT thin films

Poly[(9,9-dioctylfluorenyl-2,7-diyl)-alt-(1,4-benzo-{2,1′-3}0-thiadiazole)] (F8BT) was used to disperse raw pristine high-pressure carbon monoxide disproportionation (HiPco) SWCNTs (NanoIntegris). Ten milligrams of F8BT and 5 mg of raw SWCNTs materials were mixed with 10 ml of toluene. The prepared solution was subjected to 30 min of ultrasonication using a top-tip dispergator (Sonics, VC500). Subsequently, the solution was centrifuged at 30,000g for 0.5 hours using an Allegra X-22R centrifuge. The supernatant was further ultracentrifuged at 50,000g for 5 hours and was then redispersed in 5 ml of chloroform. The resulting dispersion was dip-coated onto a silicon wafer with a 300-nm-thick oxide layer pretreated with 150 W oxygen plasma for 3 min. Ultimately, the substrate was rinsed with toluene and acetone for 30 min each and subsequently dried under nitrogen flow for 60 min in a vacuum oven set at 120°C.

Fabrication of light-emitting CNT-TFTs

The (10, 5) SWCNTs network was transferred to a clean quartz substrate (10 mm by 10 mm, HF-Kejing) using a wet transfer process, followed by an annealing step in tube furnace for 30 min at 150°C with argon gas flow. Electron-beam lithography (EBL) (Voyager, Raith GmbH.) was used to define the channel and the CNTs outside the channel were etched through inductively coupled plasma etching (Minilock, Trion). EBL was again used to define the area of contact electrodes before electron-beam evaporation of 0.5-nm Ti and 50-nm Au. Then, ~20-nm layer of poly (methyl methacrylate) (0.6%, dissolved in n-butyl acetate, polymer source, Mw = 315 kg/mol, syndiotactic) was spin-coated onto the substrate. Twenty-nanometer layer of HfO2 was then deposited at 90°C using atomic layer deposition, completing the hybrid gate dielectric stack. Last, physical vapor deposition was used to deposit 100 nm Au as the gate metal onto the shadow mask-covered substrate.

EL characterization of light-emitting CNT-TFTs

For EL measurements, Keithley 2634B source meter was used to perform gate-voltage sampling in a constant current mode. All measurements were conducted on a homebuilt confocal microscope system equipped with a 50× objective (Olympus, N.A. 0.65). EL signals were recorded with an IsoPlane SCT320 imaging spectrometer (groove density, 600 lines mm−1) and a short-wavelength infrared imaging camera (NRV-640, Princeton Instruments).

EEG recording and frequency analysis

A playback signal derived from an EEG dataset was loaded into an arbitrary function generator (AFG31102, Tektronix) to generate input signals, while two DC voltage sources (GS200, Yokogawa) provided supply voltage. Input and output waveforms were monitored using an oscilloscope (MSOS254A, Keysight). The recorded EEG signals were analyzed using fast Fourier transforms (FFTs) to characterize their power spectral density. The FFT window function was the Hann window.

Acknowledgments

Funding:

This work was supported by the National Key R&D Program of China (grant no. 2021YFA1202904 for Y.H.) and the Postdoctoral Fellowship Program of China Postdoctoral Science Foundation (grant nos. GZC20250364 and 2025 M780494 for G.L.).

Author contributions:

Conceptualization: R.H., Y.W., G.L., L.-M.P., and Y.H. Investigation: R.H., Y.W., G.L., Z.Z., T.W., Y.C., W.W., and T.B. Methodology: R.H., Y.W., G.L., T.B., X.L., and L.-M.P. Data curation: R.H., Y.W., G.L., Y.C., W.W., T.B., and L.-M.P. Software: R.H. and Z.F. Formal analysis: R.H., Y.W., G.L., Z.F., W.W., T.B., M.X., L.-M.P., and Y.H. Visualization: R.H., Y.W., G.L., Z.F., Y.C., W.W., and T.B. Writing—original draft: R.H., Y.W., G.L., and L.-M.P. Writing—review and editing: R.H., Y.W., G.L., X.L., S.W., and Y.H. Validation: R.H., Y.W., T.W., Z.F., Y.C., W.W., M.X., X.H., X.L., L.-M.P., and Y.H. Supervision: L.-M.P. and Y.H. Funding acquisition: G.L., L.-M.P., and Y.H. Project administration: L.-M.P. Resources: M.X., L.-M.P., and Y.H.

Competing interests:

The authors declare that they have no competing interests.

Data, code, and materials availability:

All data and code needed to evaluate and reproduce the results in the paper are present in the paper and/or the Supplementary Materials. This research did not generate new materials.

Supplementary Materials

This PDF file includes:

Supplementary Text S1 and S2

Figs. S1 to S18

Table S1

sciadv.aeb5852_sm.pdf (5.4MB, pdf)

REFERENCES

  • 1.Chung H. U., Rwei A. Y., Hourlier-Fargette A., Xu S., Lee K., Dunne E. C., Xie Z., Liu C., Carlini A., Kim D. H., Ryu D., Kulikova E., Cao J., Odland I. C., Fields K. B., Hopkins B., Banks A., Ogle C., Grande D., Park J. B., Kim J., Irie M., Jang H., Lee J. H., Park Y., Kim J., Jo H. H., Hahm H., Avila R., Xu Y., Namkoong M., Kwak J. W., Suen E., Paulus M. A., Kim R. J., Parsons B. V., Human K. A., Kim S. S., Patel M., Reuther W., Kim H. S., Lee S. H., Leedle J. D., Yun Y., Rigali S., Son T., Jung I., Arafa H., Soundararajan V. R., Ollech A., Shukla A., Bradley A., Schau M., Rand C. M., Marsillio L. E., Harris Z. L., Huang Y., Hamvas A., Paller A. S., Weese-Mayer D. E., Lee J. Y., Rogers J. A., Skin-interfaced biosensors for advanced wireless physiological monitoring in neonatal and pediatric intensive-care units. Nat. Med. 26, 418–429 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Khodagholy D., Gelinas J. N., Zhao Z., Yeh M., Long M., Greenlee J. D., Doyle W., Devinsky O., Buzsáki G., Organic electronics for high-resolution electrocorticography of the human brain. Sci. Adv. 2, e1601027 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Srinivasan S. S., Carty M. J., Calvaresi P. W., Clites T. R., Maimon B. E., Taylor C. R., Zorzos A. N., Herr H., On prosthetic control: A regenerative agonist-antagonist myoneural interface. Sci. Robot. 2, eaan2971 (2017). [DOI] [PubMed] [Google Scholar]
  • 4.Mishra S., Kim Y.-S., Intarasirisawat J., Kwon Y.-T., Lee Y., Mahmood M., Lim H.-R., Herbert R., Yu K. J., Ang C. S., Yeo W.-H., Soft, wireless periocular wearable electronics for real-time detection of eye vergence in a virtual reality toward mobile eye therapies. Sci. Adv. 6, eaay1729 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Kim D.-H., Lu N., Ma R., Kim Y.-S., Kim R.-H., Wang S., Wu J., Won S. M., Tao H., Islam A., Yu K. J., Kim T., Chowdhury R., Ying M., Xu L., Li M., Chung H.-J., Keum H., McCormick M., Liu P., Zhang Y.-W., Omenetto F. G., Huang Y., Coleman T., Rogers J. A., Epidermal electronics. Science 333, 838–843 (2011). [DOI] [PubMed] [Google Scholar]
  • 6.Xu S., Zhang Y., Jia L., Mathewson K. E., Jang K.-I., Kim J., Fu H., Huang X., Chava P., Wang R., Bhole S., Wang L., Na Y. J., Guan Y., Flavin M., Han Z., Huang Y., Rogers J. A., Soft microfluidic assemblies of sensors, circuits, and radios for the skin. Science 344, 70–74 (2014). [DOI] [PubMed] [Google Scholar]
  • 7.Liu Y., Pharr M., Salvatore G. A., Lab-on-Skin: A review of flexible and stretchable electronics for wearable health monitoring. ACS Nano 11, 9614–9635 (2017). [DOI] [PubMed] [Google Scholar]
  • 8.J. G. Webster, Medical Instrumentation: Application and Design (John Wiley & Sons, 2009). [Google Scholar]
  • 9.Xiang L., Zeng X., Xia F., Jin W., Liu Y., Hu Y., Recent advances in flexible and stretchable sensing systems: From the perspective of system integration. ACS Nano 14, 6449–6469 (2020). [DOI] [PubMed] [Google Scholar]
  • 10.W. M. Sansen, Analog Design Essentials (Springer Science & Business Media, 2007). [Google Scholar]
  • 11.Luo Z., Peng B., Zeng J., Yu Z., Zhao Y., Xie J., Lan R., Ma Z., Pan L., Cao K., Lu Y., He D., Ning H., Meng W., Yang Y., Chen X., Li W., Wang J., Pan D., Tu X., Huo W., Huang X., Shi D., Li L., Liu M., Shi Y., Feng X., Chan P. K. L., Wang X., Sub-thermionic, ultra-high-gain organic transistors and circuits. Nat. Commun. 12, 1928 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.H. Wang, M. Yang, Q. Huang, K. Zhu, Y. Zhao, Z. Liang, C. Chen, Z. Wang, Y. Zhong, X. Zhang, R. Huang, “New insights into the physical origin of negative capacitance and hysteresis in NCFETs” in 2018 IEEE International Electron Devices Meeting (IEEE, 2018), pp. 31.1.1–31.1.4.
  • 13.Cao W., Banerjee K., Is negative capacitance FET a steep-slope logic switch? Nat. Commun. 11, 196 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Hu Y., Peng L.-M., Xiang L., Zhang H., Flexible integrated circuits based on carbon nanotubes. Acc. Mater. Res. 1, 88–99 (2020). [Google Scholar]
  • 15.Xiang L., Zhang H., Dong G., Zhong D., Han J., Liang X., Zhang Z., Peng L.-M., Hu Y., Low-power carbon nanotube-based integrated circuits that can be transferred to biological surfaces. Nat. Electron. 1, 237–245 (2018). [Google Scholar]
  • 16.Tang J., Cao Q., Tulevski G., Jenkins K. A., Nela L., Farmer D. B., Han S.-J., Flexible CMOS integrated circuits based on carbon nanotubes with sub-10 ns stage delays. Nat. Electron. 1, 191–196 (2018). [Google Scholar]
  • 17.Lei T., Shao L.-L., Zheng Y.-Q., Pitner G., Fang G., Zhu C., Li S., Beausoleil R., Wong H.-S. P., Huang T.-C., Cheng K.-T., Bao Z., Low-voltage high-performance flexible digital and analog circuits based on ultrahigh-purity semiconducting carbon nanotubes. Nat. Commun. 10, 2161 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Long G., Wang Y., Bai T., Li W., Zhang P., Deng X., Cai X., Xi M., Lin Y., Cheng X., Fan C., Xia F., Luo X., Zhang Z., Liang X., Zhang Z., Sun N., Peng L.-M., Hu Y., Super-saturated complementary carbon nanotube transistors with intrinsic gain singularities. Nat. Commun. 16, 3390 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Wu Y., Farmer D. B., Zhu W., Han S.-J., Dimitrakopoulos C. D., Bol A. A., Avouris P., Lin Y.-M., Three-terminal graphene negative differential resistance devices. ACS Nano 6, 2610–2616 (2012). [DOI] [PubMed] [Google Scholar]
  • 20.Cheng R., Yin L., Hu R., Liu H., Wen Y., Liu C., He J., Modulation of negative differential resistance in black phosphorus transistors. Adv. Mater. 33, 2008329 (2021). [DOI] [PubMed] [Google Scholar]
  • 21.Wang Z., Xu H., Zhang Z., Wang S., Ding L., Zeng Q., Yang L., Pei T., Liang X., Gao M., Peng L.-M., Growth and performance of yttrium oxide as an ideal high-κ gate dielectric for carbon-based electronics. Nano Lett. 10, 2024–2030 (2010). [DOI] [PubMed] [Google Scholar]
  • 22.Wei N., Gao N., Xu H., Liu Z., Gao L., Jiang H., Tian Y., Chen Y., Du X., Peng L.-M., Wafer-scale fabrication of carbon-nanotube-based CMOS transistors and circuits with high thermal stability. Nano Res. 15, 9875–9880 (2022). [Google Scholar]
  • 23.Zhang H., Liu Y., Yang C., Xiang L., Hu Y., Peng L.-M., Wafer-scale fabrication of ultrathin flexible electronic systems via capillary-assisted electrochemical delamination. Adv. Mater. 30, 1805408 (2018). [DOI] [PubMed] [Google Scholar]
  • 24.Zhang H., Xiang L., Yang Y., Xiao M., Han J., Ding L., Zhang Z., Hu Y., Peng L.-M., High-performance carbon nanotube complementary electronics and integrated sensor systems on ultrathin plastic foil. ACS Nano 12, 2773–2779 (2018). [DOI] [PubMed] [Google Scholar]
  • 25.Lee S., Cho Y. J., Han B., Lee J., Choi S., Kang T., Chu H. Y., Kwag J., Kim S. C., Jang J., Poly-Si thin-film transistors on polyimide substrate for 1 mm diameter rollable active-matrix organic light-emitting diode display. Adv. Eng. Mater. 24, 2100910 (2022). [Google Scholar]
  • 26.Avouris P., Freitag M., Perebeinos V., Carbon-nanotube photonics and optoelectronics. Nat. Photonics 2, 341–350 (2008). [Google Scholar]
  • 27.Jakubka F., Backes C., Gannott F., Mundloch U., Hauke F., Hirsch A., Zaumseil J., Mapping charge transport by electroluminescence in chirality-selected carbon nanotube networks. ACS Nano 7, 7428–7435 (2013). [DOI] [PubMed] [Google Scholar]
  • 28.Michaelson H. B., The work function of the elements and its periodicity. J. Appl. Phys. 48, 4729–4733 (1977). [Google Scholar]
  • 29.Haldar T., Wollandt T., Weis J., Zschieschang U., Klauk H., Weitz R. T., Burghartz J. N., Geiger M., High-gain, low-voltage unipolar logic circuits based on nanoscale flexible organic thin-film transistors with small signal delays. Sci. Adv. 9, eadd3669 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Zhong D., Zhao C., Liu L., Zhang Z., Peng L.-M., Continuous adjustment of threshold voltage in carbon nanotube field-effect transistors through gate engineering. Appl. Phys. Lett. 112, 153109 (2018). [Google Scholar]
  • 31.Jiang C., Choi H. W., Cheng X., Ma H., Hasko D., Nathan A., Printed subthreshold organic transistors operating at high gain and ultralow power. Science 363, 719–723 (2019). [DOI] [PubMed] [Google Scholar]
  • 32.Sekitani T., Yokota T., Kuribara K., Kaltenbrunner M., Fukushima T., Inoue Y., Sekino M., Isoyama T., Abe Y., Onodera H., Someya T., Ultraflexible organic amplifier with biocompatible gel electrodes. Nat. Commun. 7, 11425 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Kondo M., Uemura T., Matsumoto T., Araki T., Yoshimoto S., Sekitani T., Ultraflexible and ultrathin polymeric gate insulator for 2 V organic transistor circuits. Appl. Phys. Express 9, 061602 (2016). [Google Scholar]
  • 34.He Z., Dai D. S. H. S., Chen M., Zou D., Chik G. K. K., Rafael R., Lee K. H., Piao Y., Zhang S., Cheng X., Chan P. K. L., Van der Waals assembled solution-processed organic monolayer single-crystal transistor for electrocardiograph sensing. Adv. Funct. Mater. 32, 2205129 (2022). [Google Scholar]
  • 35.Schlupp P., Vogt S., Von Wenckstern H., Grundmann M., Low voltage, high gain inverters based on amorphous zinc tin oxide on flexible substrates. APL Mater. 8, 061112 (2020). [Google Scholar]
  • 36.Divya M., Pradhan J. R., Priyadarsini S. S., Dasgupta S., High operation frequency and strain tolerance of fully printed oxide thin film transistors and circuits on PET substrates. Small 18, 2202891 (2022). [DOI] [PubMed] [Google Scholar]
  • 37.Li Z., Chai Z., Wang G., Zhang J., Wang T., Zhao D., Busnaina A. A., Lu X., High-performance flexible electronics fabricated using a surface energy-directed assembly process on ultrathin polyimide substrates. Small 21, 2409458 (2025). [DOI] [PubMed] [Google Scholar]
  • 38.Vidor F. F., Meyers T., Müller K., Wirth G. I., Hilleringmann U., Inverter circuits on freestanding flexible substrate using ZnO nanoparticles for cost-efficient electronics. Solid State Electron. 137, 16–21 (2017). [Google Scholar]
  • 39.Hsu S.-M., Su D.-Y., Tsai F.-Y., Chen J.-Z., Cheng I.-C., Flexible complementary oxide thin-film transistor-based inverter with high gain. IEEE Trans. Electron Devices 68, 1070–1074 (2021). [Google Scholar]
  • 40.Pu J., Funahashi K., Chen C.-H., Li M.-Y., Li L.-J., Takenobu T., Highly flexible and high-performance complementary inverters of large-area transition metal dichalcogenide monolayers. Adv. Mater. 28, 4111–4119 (2016). [DOI] [PubMed] [Google Scholar]
  • 41.Li N., Wang Q., Shen C., Wei Z., Yu H., Zhao J., Lu X., Wang G., He C., Xie L., Zhu J., Du L., Yang R., Shi D., Zhang G., Large-scale flexible and transparent electronics based on monolayer molybdenum disulfide field-effect transistors. Nat. Electron. 3, 711–717 (2020). [Google Scholar]
  • 42.Li N., Wei Z., Zhao J., Wang Q., Shen C., Wang S., Tang J., Yang R., Shi D., Zhang G., Atomic layer deposition of Al2O3 directly on 2D materials for high-performance electronics. Adv. Mater. Interfaces 6, 1802055 (2019). [Google Scholar]
  • 43.Tang J., Wang Q., Tian J., Li X., Li N., Peng Y., Li X., Zhao Y., He C., Wu S., Li J., Guo Y., Huang B., Chu Y., Ji Y., Shang D., Du L., Yang R., Yang W., Bai X., Shi D., Zhang G., Low power flexible monolayer MoS2 integrated circuits. Nat. Commun. 14, 3633 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Huang Q., Wang J., Li C., Zhu J., Wang W., Huang Y., Zhang Y., Jiao H., Zhang S., Meng H., Zhang M., Wang X., Intrinsically flexible all-carbon-nanotube electronics enabled by a hybrid organic-inorganic gate dielectric. Npj Flex. Electron. 6, 61 (2022). [Google Scholar]
  • 45.Andrzejak R. G., Lehnertz K., Mormann F., Rieke C., David P., Elger C. E., Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: Dependence on recording region and brain state. Phys. Rev. E 64, 061907 (2001). [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Text S1 and S2

Figs. S1 to S18

Table S1

sciadv.aeb5852_sm.pdf (5.4MB, pdf)

Data Availability Statement

All data and code needed to evaluate and reproduce the results in the paper are present in the paper and/or the Supplementary Materials. This research did not generate new materials.

Articles from Science Advances are provided here courtesy of American Association for the Advancement of Science

RESOURCES