Uniform and Stable Aerosol Jet Printing of Carbon Nanotube Thin-Film Transistors by Ink Temperature Control

Abstract

Semiconducting carbon nanotube (CNT) networks exhibit electrical, mechanical, and chemical properties attractive for thin-film applications, and printing allows for scalable, economically favorable fabrication of CNT thin-film transistors (TFTs). However, device-to-device variation of printed CNT-TFTs remains a concern, which largely stems from variations in printed CNT thin-film morphology and resulting properties. In this work, we overcome the challenges associated with printing uniformity and demonstrate an aerosol jet printing process that yields devices exhibiting a hole mobility of μ_h = 12.5 cm²/V · s with a relative standard deviation as small as 4% (from over 38 devices). The enabling factors of such high uniformity include control of the CNT ink bath temperature during printing, ink formulation with nonvolatile and viscosifying additives, and a thermal treatment for polymer removal. It is discovered that a low CNT ink temperature benefits aerosol jet printing uniformity and stability in both short-term (~1 min) and long-term (~1 hour) printing settings. These findings shed light on the effect of a commonly overlooked dimension of CNT aerosol jet printing and provide a practical strategy to large-scale, high-consistency realization of CNT-TFTs.

Graphical Abstract

INTRODUCTION

Semiconducting carbon nanotube (CNT) thin films are promising for electronics applications due to their superior electrical, chemical, and mechanical properties.^1–4 Their carrier mobility and air stability are high compared to organic semiconductor alternatives;⁴ their large surface-to-volume ratio suggests possible implementation in the field of sensing;^{5, 6} and their mechanical flexibility enables their use in flexible electronics.⁷ So far, there have been experimental demonstrations of logic circuits,^8–10 display backplanes,¹¹ and sensors^{4, 6} based on the use of carbon nanotube thin-film transistors (CNT-TFTs). Among a wide variety of solution-processing techniques, printing has received growing attention for CNT-TFT fabrication due to its low cost and compatibility with a great variety of substrates, both rigid and flexible.^{4, 7}

While template-based printing (e.g., gravure printing) offers high throughput; direct-write printing, including inkjet printing and aerosol jet printing, has demonstrated potential in rapid prototyping and individual print customization.¹² The recent proliferation of research into printed CNT-TFTs has largely focused on improving printing throughput,^{7, 12} broadening substrate versatility,¹³ expanding applications,^{6, 14} and enhancing the performance of resulting devices.¹⁵ However, while CNT-TFT wafer-scale uniformity via dip coating and photolithography has been well studied,^{16, 17} there is a lack of studies investigating the performance uniformity of printed devices even though homogeneity is pivotal for large-scale and practical applications.¹⁸ Limited observations have shown that printing offers inferior device uniformity compared to other CNT deposition methods like drop casting or dip coating.^{8, 19}

Compared to other direct-write printing methods, such as inkjet printing, aerosol jet printing offers the potential for high process stability,¹⁸ which is a crucial factor influencing device-to-device variation in printed electronics. The incorporated sheath gas flow prevents direct contact between the ink and the nozzle sidewall, and thus reduces the risk of clogging (Fig. S1),²⁰ which is a common issue faced by other direct-write methods.^{11, 12} The relationship between process parameters, such as sheath gas flow rate, and the resulting printed film morphology has been studied for printed conductive Ag and CNT traces, with focus on the impact of flow rates, platen speed, number of printing passes (as illustrated in Fig. S2), and platen temperature on film properties and variation.^{21, 22} Meanwhile, ink temperature, with its strong relationship to the ink rheology properties and volatility,²³ has not been expressly considered.

In this work, we demonstrate an aerosol jet printing process that yields high uniformity and expanded printing stability for carbon nanotube thin films in CNT-TFTs. Our devices exhibit transistor performance commensurate with the best previously reported CNT-TFTs, both printed and dip-coated, while achieving the highest device-to-device uniformity. Fine tuning of the CNT ink bath temperature, an often-overlooked parameter for aerosol jet printing, is one of the key factors enabling the realization of such uniformity. This was achieved by a reduction in the bath temperature from 32 °C to 8 °C, which improved ink atomization, resulting in an increase in film density and a hole mobility of 12.5 ± 0.5 cm²/V · s. In addition, we found that a lower ink temperature improves short-term (~1 min) printing uniformity and printing stability over the timescale of around 1 hour. Other enabling aspects, including ink formulation and the use of rapid thermal annealing (RTA) for polymer removal, also are shown to improve CNT thin-film channel definition and film morphology. These investigations open the door for the direct-write fabrication of highly-uniform, wafer-scale CNT-TFTs.

RESULTS AND DISCUSSION

The CNT-TFTs explored in this study were fabricated on p++ Si wafers with 300 nm SiO₂ top layer. As shown in Fig. 1a, the devices present a bottom-gate structure with the heavily doped silicon substrate acting as the gate. All other device components were deposited by aerosol jet printing. A detailed description of the printing process is shown in Fig. S3 and the MATERIALS AND METHODS section. As illustrated in Fig. 1b and Fig. S1, inks containing electronic materials are loaded in an ink vial and atomized into aerosol mists by ultrasonic transduction. The ultrasonic bath is connected to a chiller such that the bath temperature can be precisely controlled throughout the printing process, and the actual ink temperature is tuned correspondingly (Fig. 1b). The aerosolized ink is carried by a nitrogen (N₂) gas flow and directed to the deposition head where an annular N₂ sheath flow is applied to focus the aerosol beam for deposition onto the substrate and to prevent direct contact between the ink and the sidewall of the nozzle, which alleviates the risk of nozzle clogging and is thus beneficial for stable printing.²⁰

Figure 1.

Printed CNT-TFT device structure and aerosol jet printing parameters. Schematic diagrams illustrating: a) the printed CNT-TFT device, including AgNP source/drain contacts, a substrate gate, and semiconducting CNT channel; and b) an ultrasonic atomizer for the aerosol jet printer (connected to a chiller to control bath and thus ink temperature) used to print the CNT and AgNP thin films.

The CNT ink used in this work was diluted from the commercial semiconducting CNT ink (Nanointegris iso-sol S-100, 99.9% semiconducting) using toluene; terpineol was added as a secondary solvent. The addition of this viscous, high-boiling point solvent ensures that the microscale ink droplets within the aerosol mist do not dry out before contact with the substrate.²⁴ It also significantly suppresses the unwanted coffee ring effect commonly observed during the deposition of carbon nanotubes either by aerosol jet²² or inkjet printing,¹⁹ allowing for more continuous CNT films (Fig S4).

In addition to high-boiling point solvent additives, polymer surfactants are frequently used both to separate the semiconducting and metallic nanotubes and to stabilize the dispersion of CNTs in solvents like toluene. However, once printed, the existence of polymer surfactants in the CNT thin film decreases on-current by screening hole mobility in CNT-to-CNT and CNT-to-electrode junctions.^{25, 26} The removal of polymer residue is commonly achieved by annealing,^{25, 27} solvent rinsing,^{28, 29} and/or chemical treatments.^{26, 30} Rinsing and other chemical treatments (including the subsequent drying steps) are possible sources of non-uniformity in printed devices due to their frequent directionality and manual implementation (Fig. S5). Hence, when the substrate is compatible, annealing is a more ideal choice as it provides uniform removal of polymer surfactants. The effectiveness of RTA was examined using an atomic force microscope (AFM). As shown in Fig. S6, the average nanotube thickness (and diameter) of CNTs drops from ~2.3 nm to ~1.6 nm after the RTA process, indicating the successful removal of wrapping polymer. Once the RTA is finished, silver particle (AgNP) source and drain electrodes were printed atop the CNT channel to complete the fabrication process of the CNT-TFTs.

It is known that exposure of CNT channels to air likely introduces systematic variation in performance among devices,^{31, 32} which is undesirable for the study of printing-related device consistency. One frequently utilized solution to the negative impact of air-exposed channels is to keep and characterize the devices under high vacuum;^{32, 33} yet, this considerably increases the measurement complexity and, moreover, is impractical for commercial use. Another option is to deposit a 30-nm Al₂O₃ capping layer on top of the samples using atomic layer deposition (ALD). The capping layer isolates the CNT channels from the ambient conditions, thus minimizing the effects of atmospheric exposure.^{31, 34} As illustrated in Fig. S7, the capping layer introduces ~5x decrease in hysteresis, indicating significantly less interaction with ambient air and moisture.

The electrical performance of the printed TFTs, including the transfer and subthreshold behaviors, are shown in Fig. 2a. The TFTs were printed under various CNT ink bath temperatures while the rest of the printing parameters were the same. It was observed that TFTs printed at lower CNT ink temperature exhibit higher mobility as well as larger on-currents (Fig. 2a & b), whereas the threshold voltages (V_T) remain relatively independent of the ink temperature (Fig. 2c). As shown in Fig. S8, TFTs printed at different temperatures exhibit a similar trend in output characteristics, despite that their absolute values of current are different. This indicates that the contact qualities of the TFTs are similar to each other. The increase in on-current is attributed to higher film density facilitated by increased ink deposition at lower CNT-ink temperatures, as illustrated in Fig. S9. Both ink viscosity and surface tension change at different ink temperature, thus influencing the formation of ultrasonically atomized aerosol.³⁵ The color variation shown in the optical images of Fig. 2d indicates a larger volume of nonvolatile CNT ink components (terpineol, CNTs, and polymer surfactants) transported onto the substrate when printing at lower temperatures. In a CNT film with a randomized network structure, carrier transport is dominated by percolation transport across CNT-to-CNT junctions. Thus, a denser film offers more percolation paths between the source and the drain electrode, giving rise to the higher on-state current and an increased effective field-effect mobility seen in Fig 2a & b.^{36, 37}

Figure 2.

Film morphology and CNT-TFT performance dependence on CNT ink bath temperature. a) Transfer characteristics and (inset) subthreshold curves of CNT-TFTs as a function of ink bath temperature and characterized in ambient air (devices are covered with 30 nm Al₂O₃ to minimize ambient effects). b) Mobility (left axis) and the relative variation σ(μ_h/μ_h) (right axis), both as a function of CNT ink bath temperature. Each data point represents average ± standard deviation. d) Optical images of CNT films immediately after printing at different ink bath temperatures.

Notably, mobility variation increases as the CNT ink temperature rises. This is preeminently caused by a decrease in CNT density with increasing ink temperature. As the CNT density approaches the percolation threshold (below which no conduction path would be available from source to drain), a small change in film density causes a significant shift in performance,³⁸ which in turn could introduce large variations in mobility from device-to-device. In addition, as the film density increases, the random variations that cause such a significant variance in devices with densities near the percolation threshold are overwhelmed by the already dense CNT films. This is evidenced by examining the CNT-TFTs printed with different numbers of CNT printing passes, which is proportional to the CNT densities. As shown in Fig. S10, the relationship between the mobility and the number of CNT printing passes is nearly linear, with the x-intercept close to 1. This suggests that the film density of the 1-pass devices is close to the percolation threshold. Correspondingly, the 1-pass devices exhibit a significantly higher variation in mobility (the standard deviation of mobility σ(μ_h) is close to 100% of the average value) compared to the 2-pass and 3-pass TFTs (σ(μ_h)/μ_h < 20%).

To further explore the temperature-uniformity dependence, we modified the carrier gas flow rate (18 sccm for 8 °C and 22 sccm for 32 °C) such that the resultant devices achieve similar on-currents and hole mobilities among 67 devices. These devices were printed in intervals to examine the impact of printing time on the resultant uniformity and performance, and the printer was kept operational (printing on dummy wafers without running any pattern files) during the intervals. In an aerosol jet printer, the ink is always printing once the various gas flows are established and stabilized, with a shutter blocking or opening the path between the printed ink and the substrate. Therefore, as the printing time increases, changes in the ink volume and composition are possible due to any number of factors, including inhomogeneous mixing in the ultrasonic bath. As shown in Fig. 3, we examined devices printed at the following time points and respective ink bath temperatures: 12, 24, 36, and 60 minutes at 8 °C; and 12, 24, 36, 48, and 60 minutes at 32 °C. All chips printed at 8 °C, regardless of the time printing was performed, have a smaller relative mobility standard deviation (1.9 – 4.0%) compared to those printed at 32 °C (4.2 – 6.7%) (Fig. 3c & 3d). This indicates that over the time scale of printing each set of devices (~1 min), the CNT-TFTs from low-temperature ink exhibit better device uniformity, regardless of mobility and CNT density. Interestingly, threshold voltages and their standard deviations show no apparent relationship to the CNT ink temperature (Fig 3e). While the mechanism remains unclear, the observation nevertheless indicates a promising tendency that the performance uniformity could be considerably enhanced via optimization of the CNT ink temperature.

Figure 3.

CNT printing stability with respect to ink bath temperature, tested under high vacuum conditions to minimize ambient effects. a) and b) Transfer characteristics of CNT-TFTs printed at ink bath temperatures of 8 °C and 32 °C, respectively, at various printing durations. Each time point shows 7 to 8 CNT-TFTs. c) Mobilities, d) relative variation of mobility, e) threshold voltage, and f) absolute variation of the threshold voltage of CNT channels printed at various time slots at 8 °C (blue) and 32 °C (red).

Interestingly, when printing is carried out for extended periods of time (~1 hour), temporal device performance trends vary differently at low and high CNT ink bath temperatures (Fig. 3c). At 8 °C, the mobility of the resulting devices tends to rise asymptotically as printing continues, with the most significant variation occurring in the first 30 minutes of printing. By contrast, the CNT-TFT mobility from devices printed at 32 °C shows a decreasing trend as a function of printing time. Such variation could be explained by examining the amount of ink delivered onto the substrate as the printing proceeds. As shown in Fig. S11, at higher CNT ink bath temperature, the volume of the ink delivered onto the substrate tends to decrease throughout the printing process. This phenomenon is attributed to the fact that only a finite amount of CNT ink (~2 mL) is loaded into the atomizer prior to printing, and that over time, the ink volume in the atomizer will decrease as a larger fraction of ink is transported to the substrate. As the ink volume in the atomizer drops, less ultrasonically atomized aerosol is generated per unit time, giving rise to lower-density films with a decreased field-effect mobility. However, at 8 °C the trend reverses, which suggests some factors counterbalance the decreasing ink deposition caused by depletion of ink volume. If appropriately tuned, ink bath temperature can be utilized to counterbalance the ink volume drop and to realize a steady printing process. The aggressive increase in mobility during the first half-hour of printing is likely due to the three reservoirs (ink, ultrasonic atomizer, and ultrasonic bath) gradually reaching thermal equilibrium, as shown in Fig. S12. Therefore, by ensuring all reservoirs have reached thermal equilibrium before printing, the initial rise in mobility can be suppressed (Fig. 4). Among the 5 time points during a 1-hour CNT printing process, no notable drift in mobility or threshold voltage was observed (Fig. 4b & c), and the 38 CNT-TFTs in total yield highly uniform performance, with μ_h = 12.5 ± 0.5 cm²/V · s and σ(μ_h)/μ_h ~4%.

Figure 4.

High-uniformity and high-stability printing of CNT-TFTs. a) Transfer characteristics of printed CNT-TFTs with ink bath temperature of 8 °C at various printing durations (devices are covered with 30 nm Al₂O₃ to minimize ambient effects). b) Mobility (left axis) and its relative variation (right axis) of CNT channels printed at various time points. c) Threshold voltage (left axis) and its absolute variation (right axis) of CNT channels printed at various time points.

To benchmark the device performance and consistency, a comparison between this work and previous works on CNT-TFTs can be seen in Fig. 5.^{7, 8, 11, 16–19, 39, 40} Field-effect hole mobility is chosen as the metric because it is most widely reported in previous works and is extensively studied and discussed in this work. CNT deposition method, as well as the substrate, are also considered due to their potential impact on TFT performance and consistency. Detailed information from all devices is shown in Table S1. In general, while non-printing techniques, such as drop-casting and dip-coating, offer higher mobilities on average, none of the CNT deposition methods or substrate selections guarantees device-to-device uniformity. The CNT-TFTs realized in this study show a remarkable device-to-device uniformity – better than any of the previous works regardless of the deposition method or the substrate, as well as a mobility higher than most of the printed CNT-TFTs and some of the non-printed CNT-TFTs from previous works.

Figure 5.

CNT-TFT performance and uniformity benchmarking. Relative standard deviations for hole mobilities of CNT-TFTs plotted with respect to their mobilities. Details for the devices benchmarked in this plot are provided in Table S1.

CONCLUSION

In summary, we discovered that CNT ink temperature plays a significant role in the uniformity of CNT-TFT and the stability of printing nanotube thin films with aerosol jet printing. The combination of proper ink formulation, ink bath temperature optimization, and the polymer removal process gives superior device-to-device uniformity and printing stability. It was discovered that low CNT ink bath temperature (~8 °C) benefits the device consistency in both short-term (~1 min) and long-term (~1h) printing. It was also observed that the atomization output of toluene+terpineol-based CNT-ink rises at a lower ink temperature. These findings address device consistency issues that have previously plagued printed CNT-TFTs and boost the potential for development of TFT fabrication towards large-scale, high-uniformity production.

MATERIALS AND METHODS

Substrate cleaning and PLL functionalization

The SiO₂/Si substrates were sonicated in Acetone and IPA for 5 minutes each, rinsed with DI water, blow-dried with nitrogen, and underwent a 3-minute oxygen plasma treatment using an Emitech K-1050X plasma asher. The power of the plasma was set to be 100 W. The substrate functionalization includes a 5-min immersion in PLL solution (0.1mg/mL, Sigma-Aldrich), DI water rinsing, and nitrogen blow-drying.

CNT channel deposition

CNT channels were printed using an Optomec AJ300 aerosol jet printer with a 150-μm nozzle. 100 μL of 99.9% semiconducting CNT ink (0.1mg/mL, IsoSol-S100, NanoIntegris), 1900 μL of toluene, and 20 μL of terpineol were loaded into the atomizer sequentially. The sheath flow rate, ultrasonic current, platen motion speed, and platen temperature were set to be 30 sccm, 410 mA, 8 mm/s and room temperature (~23 °C), respectively. The carrier gas flow and the ink bath temperature varied. Unless specified, 2 passes of CNTs were printed for the channel.

RTA-based polymer removal

RTA was performed on a Jipelec JetFirst 100 rapid thermal anneal system. Once the sample had been loaded, the chamber was purged with nitrogen twice, then was held at a vacuum level of 1 Torr. The chamber temperature was subsequently increased to 450 °C in 2 minutes, held for 8 minutes, and cooled down to room temperature within 4 minutes. The RTA chamber was continuously purged with N₂ during cooldown.

Fabrication of source and drain electrodes

Commercial AgNP ink (UTDAgX, UT Dots Inc.) was used to print the source and drain electrodes. Prior to printing, the ink was mixed with terpineol in a 9:1 volume ratio to modify the ink viscosity, and 2mL of the as-prepared ink was loaded into the atomizer. Aerosol jet printing of AgNP was carried out using an Optomec AJ300 aerosol jet printer with a 100-μm nozzle. The sheath flow rate, carrier gas flow rate, and ultrasonic current were set to be 25 sccm, 18 sccm, and 350 mA, respectively. The platen temperature was held at 60 °C, with a printing speed of 3 mm/s. 1 pass of AgNP was printed for the electrodes. After printing the electrodes, the samples were placed in an Isotemp 281A oven and baked for an hour at 200 °C under ambient pressure.

Al₂O₃ deposition

After the fabrication of CNT-TFTs, 30-nm of Al₂O₃ was deposited at 120 °C and 1 Torr using a Kurt Lesker ALD-150LX ALD system. Trimethylaluminum (TMA) was selected as the aluminum precursor, and water vapor was selected as the oxygen precursor. The pulse time of 70 ms and 40 ms was selected for water vapor and TMA, respectively, and the purge time was 10000 ms for both precursors. 370 cycles to grow 30-nm Al₂O₃.

Microscopic and Electrical characterizations

Atomic force microscopy (AFM) imaging was performed using a Digital Instruments Dimension 3100 AFM. Scanning electron microscopy (SEM) was carried out with an Apreo S SEM from ThermoFisher Scientific. Electrical characterizations under ambient conditions were performed on a tabletop probe station, and a Lakeshore vacuum probe station was used for electrical characterizations under vacuum (< 10⁻⁵ Torr). Both probe stations were connected to an Agilent B-1500 semiconductor parameter analyzer during the measurements.