Capillary‐Assisted Self‐Assembly of Carbon Nanotubes for the Self‐Powered Photothermoelectric Detector

The development of mid‐infrared (MIR) detectors has become a hot research topic with significant progress in low‐dimensional materials and clean‐room fabrication strategies. Some of the applications of MIR detectors include industrial non‐destructive testing, wearable safety monitoring, and other Internet of Things. Photothermoelectric (PTE) mechanism, as a room‐temperature free‐bias conversion mode, is comprehensively developed in the MIR regimes in the last decade. Although carbon nanotubes (CNTs) and their related materials are demonstrated as effective PTE conversion materials, the large‐area scalable detector fabrication based on the Si substrate is still underdeveloped, thus limiting further PTE device designs and industrial applications. Herein, the self‐assembly CNT‐based detectors driven by the capillary force are fabricated to achieve sensitive and rapid IR detection, and photoresponse measurements of PTE detectors are experimentally performed at room temperature and atmospheric conditions. This work reveals that the PTE mechanism can play a key role in the IR response, thereby broadening horizons about high‐performance IR detectors in industrial applications.


Introduction
Mid-infrared (MIR) detection has drawn tremendous attention because it can satisfy the requirements for numerous applications such as medical tracking, gas sensors, security imaging, and Internet of Things (IoTs). [1][2][3] Nowadays, industrial MIR detectors are mainly based on the mercury cadmium telluride (MCT, HgCdTe), palladium sulfide (PdS), and III-V quantum-well/dot (QWIP/QWID). [4,5] However, MCT and PdS detectors suffer from the high toxicity of heavy metals, and QWIP/QWID detectors require cryogenic working conditions. [6] DOI: 10.1002/admt.202300309 Although these detectors can show a respectable response in IR detection, other issues, such as high-cost fabrication and incompatible fabrication with the complementary metal-oxidesemiconductor (CMOS) technique, also hinder their further applications. Thus, the novel MIR detectors operated on Si substrates are urgent.
Beyond the limitations of bandgap, photothermoelectric (PTE) detectors are considered a powerful tool for MIR detection. [7,8] PTE detectors can be also designed for a simple configuration such as the asymmetric electrodes and operate at room temperature with zero bias and low power consumption. [9] Additionally, PTE detectors, dominated by Johnson-Nyquist thermal noise, suppress the shot noise and 1/f noise of detectors. [10,11] In contrast, common IR detectors suffer from a low signal-to-noise ratio (SNR) effect and the presence of a dark current. PTE detectors based on low-dimensional materials offer advantageous thermodynamics and optoelectrical properties. [12][13][14][15] Many nanomaterials have been explored for PTE detectors in the IR regime. Cai et al. investigated the PTE detector based on single-layer graphene and achieves a responsivity of over 10 V W −1 in the terahertz (THz) regime. However, the fabrication of single-layer graphene is complicated, the absorption is relatively low, and the cost is high. [8] Although microporous graphene enhances the inherently poor IR absorption of single-layer graphene, their related PTE detectors show a 4 mV W −1 responsivity. However, it is challenging to pattern the array of microporous graphene, which limits further industrial applications. [16] Graphene fibers are also used in MIR optical communications, but the PTE detector based on them requires high bias up to 6 V and cannot achieve a self-powered response. [17] Black phosphorus (BP) is one promising PTE material with high carrier mobility, but it suffers from unstable ambient problems. Its current fabrication process depends on exfoliation, and the device is generally fabricated in one single BP flake, [18] which makes it challenging to achieve large-area production. Carbon nanotubes (CNTs) can be regarded as an effective 1D material for absorbing IR radiation, inherently combining free carrier-based intraband and exciton-based interband absorption processes. [13,19] With strong metallic properties, multi-walled CNTs (MWCNTs) can show a heat-induced photoresponse. [20]  The broadband absorption ability of CNTs, combined with highmobility carriers and fast response, open opportunities for photodetectors, as well as other optoelectronic applications. [21][22][23] Despite these competitive advantages, state-of-the-art PTE detectors still twist some challenges. First, mainstream sensitive PTE detectors rely on nanometer-level channel length control, which needs the expensive electron beam lithography (EBL) technique. Thus, reducing the cost is urgent for sensitive PTE detectors. Second, current PTE detectors based on thermoelectric junctions need the chemical doping dropped on one spot, finally resulting in a macroscopic film with PTE junctions. [24] For such devices, it is significantly difficult to drop micrometer-level liquid spots and achieve small-area pixel control, which further prevents the miniaturization and integration of devices. Third, in the vertical PTE detectors, metal electrodes are mostly fabricated on the top of photoactive layers, which decreases the IR absorption of active materials.
In our work, we fabricate a high-performance scalable PTE detector driven by capillary-assisted self-assembly CNTs. Lithography processes are performed before the growth of CNTs, and therefore contaminations coming from photoresists are avoided. By dropping dimethyl sulfoxide (DMSO) on the top of the whole device, the thermoelectric junction is naturally produced without any further CNT transfer. Furthermore, the PTE detectors exhibit a potential industrial non-destructive testing (NDT) application because the components in the PTE detectors can work under a high temperature of about at least 700°C.

The Capillary-Assisted Self-Assembly of the CNT Junction Induced by DMSO
The fabrication process of the device is shown in Figure 1. As we use a plasma-enhanced method to grow CNTs, some gaps exist among the vertical CNTs. When the liquid enters the gaps, the capillary forces induce the liquid into the buffer layer or even the electrode layer and Si substrate. Although the metal surface is hydrophobic, the repeated drop-casting along with the flowing direction control of the liquid assists the CNTs to soak among the interspaces. The existence of DMSO may help optimize the energy filtering effect, thus leading to enhanced thermoelectric performance. [25] DMSO also can induce a non-overlapped CNT thermoelectric junction. According to previous experimental results, an increased overlapped area of the thermoelectric junction will significantly reduce the PTE performance. [26] Thus, our device without overlapped junction will not cause a thermal loss in the junction area. Previously, researchers mainly focus on the aligned CNTs in parallel or perpendicular to the illumination, [24,26] and few papers refer to the capillary forcedriven self-assembly CNT clusters. Additionally, pristine aligned CNTs show a strong polarization sensitivity ranging from a broad spectrum with low noise. However, they are mechanically unstable and easily broken. Their polarization sensitivity also provides challenges for the shape or area design of PTE detectors. Therefore, we here use the CNT cluster to satisfy global IR illumination requirements. In our configuration, the CNT channel addressed by DMSO acts as a bridge to connect two asymmetric heightdifference electrodes and create the circuit (Figure 2), where Ti and Cr are selected as electrodes because they can endure high temperatures up to 700°C, which can enable future industrial applications. Furthermore, this configuration can benefit the miniaturization and integration of PTE sensors.

IR Response Mechanism
Given the IR response mechanism of our detector, the pyroelectric mechanism can be excluded because of its general typical capacitor configuration and pyroelectric material. The bolometric and photoconductive mechanism needs external bias, but our detector works under zero bias. As for the photovoltaic mechanism, the relatively low photon energy in the MIR regimes is difficult to induce photocarriers. When electromagnetic radiation illuminates the device, the CNTs will absorb the photon energy and the temperature of CNTs will increase. We tested the temperature difference of the CNT-1 area and CNT-2 area with the incident power intensity of 2.1 × 10 4 μWmm −2 (Table S1, Supporting Information), which demonstrates our detector is dominated by the PTE mechanism. Due to the different growth conditions, the dimensions and lengths of CNTs are various, which will show different Seebeck coefficients. [27] For the CNT clusters, it is challenging to measure the Seebeck coefficient because their metal buffer layer may cause a short circuit of the measurement setup. Given that the skin depth of the metal electrode Ti is over 37 nm in the 100 THz (3 μm) regime (Table S2, Supporting Information), the Ti electrode of 40 nm is highly IR reflective as a heat-sink role. The thickness of the Cr electrode is about 10 nm, which is under the skin depth of Cr in the THz regime. Thus, the Cr electrode will act as the heat source combined with the top CNTs. [28,29] Thus, the CNT-2 area and Cr electrode will form the CNT-Cr composite. Here, we evaluate the Seebeck coefficient difference ΔS = V p ΔT , where V p is the photovoltage of the detector, ΔT is the temperature difference. The calculated Seebeck coefficient difference is 14 μV K −1 , which can be further optimized by changing the concentration of DMSO and the physical parameters of CNTs.

I-V Curve of PTE Detectors
During the measurement, the detector was irradiated by the filtered continuous-wave (CW) IR waves with about 2.1 × 10 4 μW mm −2 input power, and the photocurrent was directly recorded using the DMM6500 current-time module (Figure 3a). Figure 3b shows the current-voltage (I-V) curve performance of the PTE detector under dark or radiated conditions. The I-V measurement would be performed at room temperature and ambient atmospheric conditions. The blackbody temperature was set at 1173 K. The linear characteristic could demonstrate that both electrodes and CNTs form Ohmic contact in a largely imposed voltage range from −120 to 120 mV. Additionally, the thermoelectric junction of CNT interfaces did not generate the rectification effect because the plasma-enhanced chemical vapor deposition (PECVD)-grown CNT might consist of many semiconducting and metallic CNTs, which might activate the metallic properties of CNTs. [26] The PTE current could be observed without any external bias. These results illustrated that PTE detectors could be used as IR energy harvesters or further applied for collecting waste energy.

Response Time
Response time is also an important parameter for evaluating PTE detectors, including rising and fall time, which can be defined as the period when the PTE current increases from 10% to 90% on the rising time or the falling times of one pulse amplitude, respectively. [30] During the on-off curve measurement, we put one thick aluminum foam between the IR-filtered window and the CNT PTE detector. When we turn on the blackbody radiation, the DMM6500 machine will record the photocurrent of the off status. When we remove the foam quickly, the photocurrent will increase and be recorded as the on-status photocurrent.  Our detector exhibits a rise time of 40 ms and a falling time of 60 ms without external bias, substantiating the real-time IR signal tracking capacities (Figure 4a-c), where the input power of the filtered CW IR waves was 2.1 × 10 4 μW mm −2 . Figure 4d shows a reproducible current response with resolved time. Such a fast response speed may originate from the localized electromagnetic field enhancement induced by the nanometer-level height difference. Additionally, the thermoelectric junction is created by no-overlapped conductive CNTs, which may reduce thermal loss. Furthermore, the experimental model can be applied to other nanowire-based systems designs, which is expected to further optimize the performance of PTE detector arrays.

Responsivity and Specific Detectivity
In the view of practical applications, specific detectivity is also one property used to evaluate the performance of the PTE detector, which is characterized as the capacity to track the weak signals compared with the noises and can be represented as, . [31] R v is the voltage responsivity of the device and is expressed as where V is the photovoltage and P in is the total incident radiation power. V n is the mean root square of the noise voltage and is defined as V n = √ 4k B TR , where k B is Boltzmann constant, T is temperature, and R is device resistance. [32] For the PTE detector, the thermal Johnson-Nyquist noise dominates. Other noises will be discussed in the Supporting Information. The total incident power is calculated by the integral of spectral radiant emittance with a wavelength of over 1.5 μm because the cut-off wavelength of the IR window is 1.5 μm. In our measurement, we assume that all incident is absorbed by the detector, and thus the practical responsivity or detectivity should be larger than the calculated value. With different blackbody radiation temperatures, Figure 5a shows the zero-bias PTE response under broad illumination. The detectivity can reach a 10 8 -level Jones when the black-body radiation is set up from 573 to 1173 K, which matches a center wavelength from 5.06 to 2.47 μm, which demonstrates that our device can satisfy the demands for IR broadband detection. Such a broadband detection capacity matches the intrinsic properties of IR thermal detectors. We also characterize the blackbody power intensity versus responsivity and noise equivalent power (NEP) in Figure 5b, showing a roughly linear relationship. According to Stephan-Bolzmann's law, the blackbody power intensity should be proportional to the fourth power of the temperature. We attribute this linear correlation to two reasons. First, the significant temperature difference between the device system and the surrounding temperature may cause thermal losses. If the detector is placed under a high-level vacuum system, the PTE current is expected to reach a higher level because of the reduced thermal losses. [33] Second, with a higher blackbody temperature, the peak wavelength will change. This detector using the novel thermal junction may show typical absorption for different wavelengths.

Long-Term Stability and Durability
Long-term stability is also a key figure of merit in evaluating the performance of PTE detectors. Normally, the performance of IR  detectors can be affected because they directly contact air and can be moist and metamorphic, eventually resulting in degeneration under ambient conditions. Thus, we test the PTE response of our device every 7 days. After 56 days, the detector still demonstrates a highly advantageous reproducibility and matches well with the original experimental results, indicating excellent stability and durability in the air. We contribute the continuous high photoresponse to the oxidization of the endpoints of CNTs, that is, the addition of DMSO has oxidized one side of CNTs and prevents the further photoresponse degradation of CNT-based PTE detectors (Figure 6).

Conclusion
In conclusion, this paper proposed a scalable PTE detector design utilizing self-assembly CNT thermoelectric junction among microscopic CNT components, providing high-performance and stable IR photoresponse. Typically, the detector shows the broadband MIR response and stable photoresponse within 2 months.
During this design, the naturally grown CNTs act as an active element of the PTE detector. Our work opens significant opportunities for large-area IR detection integrated with CNTs, and it also offers new design sights for other Si-based sensors.

Experimental Section
Device Fabrication: The PTE detectors were driven by the self-assembly CNT thermoelectric junction, fabricated followed by the procedure described as follow: 1) one prime P-type doping Si wafer was immersed in H 2 O 2 and H 2 SO 4 piranha solution (volume ratio 1:4) for 15 min under room temperature, and the P-type doping Si wafer was rinsed with deionized water and dried by the spin rinse dryer (SRD) machine system; 2) hexamethyldisilazane (HMDS) of 5 nm thickness was initially coated on the cleaned Si wafer at 150°C for 15 min to enhance the adhesion between photoresist and Si wafer substrate; 3) spin-coated the positive photoresist Shipley 1805 (S1805); 4) maskless aligner (MLA) photolithography method was used to pattern the insulating suspended layer and the top electrode layer, followed by a 120°C post bake process for 90 s; 5) 40 nm Ti electrode was deposited under room temperature without bias by magnetron sputtering system, AJA Inc; 6) unfinished device was first placed immersed into remover PG solution for PMMA removal under 80°C for 60 min and under room temperature overnight to achieve a lift-off process; 7) the device performed a soft bake process under 120°C for 2 min to remove the water left inside the wafer; 8) MLA photolithography method employing PMGI/S1805 bilayer photoresist was used to pattern the insulating layer, followed by a 120°C post-bake for 90 s; 9) Typically, 60 nm SiO 2 insulating layer was first deposited under room temperature without bias by AJA magnetron sputtering system to separate two electrodes and achieve a height difference between two electrodes; 10) 10 nm Cr as a catalyst buffer layer and the electrode layer were deposited, followed by a reliable lift-off process; 11) The device performed a post-bake process under 120°C for 2 min to remove the water left inside the wafer; 12) MLA photolithography method employing PMGI/S1805 bilayer photoresist was used to pattern the electrode layer and CNT growth layer, followed by a 120°C prebake for 90 s; 13) 2 nm Ni catalyst film was deposited by E-beam evaporation (Angstrom E-beam Deposition System) onto P-doped Si substrates at 1.5 Å s −1 growth rate, followed by the same aforementioned lift-off process; 14) The device was cut into the single individual device using DISCO dicing saw system; 15) The cold wall PECVD machine with a 2″ graphite heater was used to grow CNTs. Vertically aligned CNTFs were grown by a PECVD reactor at 695°C, 5 mbar system pressure, and 75 W DC plasma power in an Aixtron Black Magic 2 system. Acetylene (C 2 H 2 ) and ammonia (NH 3 ) flowed at rates of 50 and 200 sccm, respectively. For a typical 15 and 30 min growth, the CNTF height was about 7 and 15 μm, respectively; 16) 10 μL DMSO solution (# 472301, purchased from Sigma-Aldrich) was dropped cast on the top of the CNT forest along with the arrow direction. By controlling the flow direction of the DMSO solution, the alignment direction of the whole CNT forest can be controlled, and the CNT forest will fall toward the top electrode to form an ultrashort suspended "CNT bridge;" 17) The device was dried at a 100°C hotplate to evaporate DMSO and enhance the charging CNT network; 18) The device can be connected to the external circuit by using conductive silvery epoxy (MG Chemicals 8331D Silver Conductive Epoxy) for further performance measurement.
Device Characterization: The morphology of CNTs was characterized using images in a JEOL JSM 7200F field emission scanning electron microscopy (SEM) at 10-15 kV voltage with 10 nA beam current. Energydispersive X-ray spectroscopy (EDX) images were taken in the same SEM system at 14 kV voltage with 15 nA beam current. Raman was investigated by a Raman microscope Bruker Senterra with a 785 nm laser at 1 mW intensity.
Device Measurement: The black-body radiation (Newport Oriel Model 67030) was adopted to act as the IR source. In the setup, an IR window that cuts off the IR radiation below 1.5 μm was placed between the detector and the source. The measurement was carried out under atmospheric pressure and room temperature (295 K).
I-V curves with black-body illumination on or off were measured by combining the Keithley 5 1/2-Digit Model 6487 Picoammeter/Voltage source (Keithley 6487) and Keithley DMM6500 6 1/2-Digit Bench/System Digital Multimeter (Keithley 6500). Here, Keithley 6487 can provide stable DC voltage, and Keithley 6500 can directly record the DC value with or without black-body illumination. The CNT PTE detector was directly connected with Keithley 6487 and Keithley 6500 without using an amplifier, and the photovoltage can be measured.
Statistics: There was no pre-processing of data in this manuscript. All experimental data of this manuscript was processed by Python.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.