1. Introduction

Infrared (IR) detection is a primary subject in optical sensing and is critical for a variety of industrial, military and scientific applications, including monitoring and controlling manufacturing process, optical communication, biological and military night time sensing. Carbon nanotubes (CNTs) are promising candidates for future IR detectors due to their unique band structure, excellent electronic and optoelectronic properties, and super mechanical and chemical stabilities. Moreover, the band gap of a semiconducting single-walled carbon nanotube (SWCNT) is tunable by changing its diameter and chirality [1]. CNTs not only present excellent electrical properties, ballistic transport, ohmic contact with electrodes and ultrahigh carrier mobility [2,3], but also exhibit strong infrared light absorption with broad band and fast light response up to picoseconds [4,5]. The absorption coefficient of SWCNT is at least one order of magnitude larger than that of mercury-cadmium-telluride, the most popular photoconductor for two-dimensional (2D) photodetector arrays, and over 70% incident radiation can be absorbed by a SWCNTs film with a thickness of 100 nm [6]. A commercial semiconductor based IR photodetector usually possesses a very fast response and high detectivity up to 10¹² Jones (1 Jones = 1 cm Hz^1/2/W) [7], but it needs to be cooled to low temperature which limits its applications. It would be advantageous to design a low cost CNT based photodetector with high quantum efficiency, high sensitivity and high speed over the IR spectral range without cooling. Additionally, with an important predominance due to its exceptional mechanical flexibility, CNT IR detector may be readily implemented on a flexible substrate at a low cost that is not possible for traditional group III-V semiconductors.

CNT photodetectors were previously developed based on both such thermal and photo effects. In a typical thermal detector, such as a bolometers [6,8–10] or thermopile [11], an electrical signal (resistance, current or voltage) was produced by temperature change due to the illumination. While in a photo detector, such as in a photoconductive detector or p-n junction photodiode, excitons are generated by interband transition via photon absorption of CNTs and electron-hole pairs are set free by excitons dissociation, resulting a photocurrent or photovoltage in the device [2]. The main difference for IR applications using these two mechanisms is the way that the electric signals are generated and affected by the interaction with the CNT.

Significant progresses have been made in recent years on CNT IR bolometers [8–10]. Most of the reports focus on the bolometer made of CNT films with a thickness more than tens of nanometers [6,9,10]. The energy absorption of the film is sufficient, but the device performance is usually seriously influenced by the surroundings. It was concluded that SWCNTs film should be suspended in vacuum environment to prevent heat dissipation. In additional, the performance of the CNT bolometer usually degrades with increasing temperature and some of them do not work at room temperature due to the significantly decreased signal-to-noise (S/N) ratio at room temperature [6]. Another significant drawback of such bolometers is that the change of conductivity is typically below 1% level and is therefore difficult to measure. The photoconductivity of the SWCNT based field effect transistor (FET) was also investigated but the photoresponse current was typically of the pA level even under a strong laser illumination of approximately 1 kW/cm² due to the symmetric built-in field in two contact areas resulting in low collection efficiency of photoexcited carriers [12]. Furthermore, both bolometers, which are dominated by thermal effect, and FETs which work by the exciton separation need a bias voltage to IR response measurement which also increases the power consumption of the system.

In principle photovoltaic IR detectors based on p-n diode or p-i-n diode can work under zero voltage or current as self-powered device without external power supply. While there exist many reports on CNT diodes, most of the CNT diodes were based on split gates, chemical doping and asymmetric contacts [13–16]. In this work we use a SWCNT based barrier-free bipolar diode (BFBD) device structure which is basically an asymmetrically contacted CNT, with one end being contacted by Pd (p-contact) and the other end being contacted by Sc or Y (n-contact). This device structure is simple, involves only doping-free fabrication process, but exhibits excellent IR photo response [17–20]. However, light absorption of a single SWCNT is too weak to be useful for practical weak IR detection. In our earlier report, thin film BFBDs were fabricated using SWCNT submonolayer network, but the multiple CNT-CNT junctions in the network degraded the device performance [21]. More recently a method for growing CNT arrays on the monocrystal quartz substrate was developed, yielding perfectly assigned SWCNT arrays along certain crystal orientations [22–24]. The growth and transfer of these CNT arrays have been realized in wafer-scale, with the highest density of CNTs up to 60 tubes/μm, and a ratio of semiconducting CNTs of up to 95% [25,26]. It is expected that such parallel CNT arrays should lead to better device performance than that based on the network due to the absence of CNT-CNT junctions between CNTs in the parallel arrays. In this paper, we aim to demonstrate that photodiodes based on well-aligned CNT array are well suited for IR detection. In particular our experimental results show superior performance advantages of the CNT BFBD based IR detectors over that based on thermal effects, and a large photo photoresponsivity ~9.87 × 10⁻⁵ A/W and detectivity ~1.09 × 10⁷ cmHz^1/2/W.

2. Experimental

SWCNT arrays used in this work were grown by chemical vapor deposition (CVD) on monocrystal quartz and transferred to silicon wafer covered with 500 nm thermally grown SiO₂. The CNT density is about 2-3 tubes per micron after transfer. The CNT arrays were cut into stripes with a width W = 20 μm (perpendicular to CNTs) by using electron beam lithography and O₂ plasma etching, and the lengths of the stripes were adapted to the channel lengths. All electrodes were patterned by electron beam lithography and deposited by electron beam evaporation. All transport measurements were carried out using Keithley 4200 semiconductor analyzer at room temperature. Photovoltaic measurements were carried out using a laser with λ = 785 nm and the power density of the focused laser beam on the device may be varied from 0 to 1.57 kW/cm².

3. Results and discussion

Shown in Figs. 1(a) and 1(b) are scanning electron microscopy (SEM) images of different magnifications showing perfectly aligned CNT arrays as grown on quartz. These CNT arrays are synthesized by chemical vapor deposition (CVD) as described in early reports [24–26]. The CNTs on the quartz are more than 100 μm in length and 1-2 nm in diameter with a density of 2-3 tubes per micron. Due to the presence of both metallic (M) and semiconducting (S) nanotubes in the CNT arrays, it is necessary to remove the metallic pathway spanning the channel of the devices used for IR detection here. Since it is difficult to remove the metallic CNTs in the channel for the devices fabricated directly on quartz substrate [27], CNTs were transferred from quartz to Si/SiO₂ or other substrate, on which we can apply an appropriate back-gate to turn off semiconducting CNTs and breakdown metallic ones using a large bias current [21,28]. CNT arrays are transferred via poly (methyl methacrylate) (PMMA)-mediated approach, which has been used for transferring nanomaterials, such as CNT, zinc oxide, graphene with simple process and high repeatability [29]. Figure 1(c) is a representative SEM image showing the transferred CNT arrays on a Si/SiO₂ substrate, and Fig. 1(d) is a SEM image showing a typical photodiode fabricated by asymmetric contact on one end of the as-transferred CNT array channel with Sc and on other end with Pd. We note that the density of CNTs does not decrease significantly during the transfer and fabrication process.

 

Fig. 1 Fabrication of the asymmetrically contacted CNT thin film diode. (a) Low and (b) high magnification SEM images of the aligned SWCNTs grown on quartz. (c) Arrays of SWCNTs on Si/SiO₂ substrate after being transferred from quartz. (d) Typical SEM image of thin film diode contacted with Sc and Pd as source and drain respectively, with channel width W = 20 μm and channel length L = 1 μm.

Download Full Size | PDF

To fabricate BFBD devices, the as-transferred SWCNT arrays were cut into stripes prior to electrode depositions and unwanted nanotubes outside the device channel region are removed via electron beam lithography (EBL) and oxygen plasma etching process. Figure 2(a) illustrates the brief fabrication process of the asymmetrically contacted photodiode which is based on the as-made SWCNT array stripe. The process begins with the fabrication of a p-type FET. The two Pd electrodes with a thickness of 50 nm are patterned by EBL and deposited by electron beam evaporation to act as the p-type source and drain electrodes of the CNT FET, and the 500 nm silicon dioxide beneath the device is used as the back-gate dielectric. Figure 2(b) shows typical p-type transfer characteristics measured from a CNT FET (with L = 1.5 μm and W = 20 μm) before and after metallic nanotubes breakdown. The as-made FET device exhibited a very low current on/off ratio of about 2, but the ratio was improved to be more than 10⁵ after metallic CNT breakdown. The corresponding output characteristics show clearly linear dependence of the device current on the voltage (see Fig. 2(c)) at low bias, showing that ohmic contacts are formed between CNTs and Pd electrodes after breakdown process. The on-current density of this FET is about 4 μA / μm at bias V_ds = 2 V.

 

Fig. 2 Electrical properties of CNT thin film FET and diode devices. (a) Schematic diagrams showing the fabrication process of the diode. Starting from a p-type FET with Pd contacts, Sc extension is deposited on one of the Pd contact to transfer it into an n-type contact, followed by PMMA coating for passivation. (b) Transfer characteristics of a typical p-type CNT thin film FET (with W = 20 μm and L = 1.5 μm) before and after electrical breakdown, and corresponding (c) output characteristics after breakdown. (d) Transfer characteristics and (e) linear and logarithm (inserted Fig.) I-V characteristics of the diode device (W = 20 μm and L = 1 μm). The channel length is reduced after adding Sc extension and PMMA passivation to the initial p-type FET.

Download Full Size | PDF

To convert the p-type CNT FET into a diode, a 500 nm Sc extension into the channel is patterned on one of the Pd electrodes of the as-made p-type FET by EBL (Fig. 2(a)). The Sc extension acts as the n-type contact to the CNT array, and the channel length is reduced from original 1.5 μm (for the p-type FET) into 1 μm. In this new device, the CNT array channel is contacted asymmetrically by Pd and Sc. The Pd electrode is aligned with the valence band of the semiconducting CNT to allow high efficient holes injection into the channel without barrier, while the Sc electrode is aligned with the conduction band of the CNT and forms ohmic contact with the CNT for barrier-free electrons injection [17,20]. In order to increase the long-time stability of the as-fabricated device, the diode is covered with a PMMA coating as passivation layer. Figure 2(d) shows the transfer characteristic of the diode coated with a PMMA passivation layer, exhibiting a typical ambipolar characteristics of an ambipolar FET with an off-state at a gate voltage of approximately 0 V. This is a preferred feature for real PV applications, since the coated diode with off-state at zero gate voltage has the largest power conversion efficiency [18]. The two-terminal I-V characteristic of the diode is shown in Fig. 2(e), exhibiting excellent rectifying behavior. The rectification ratio (i.e., on-current at bias of 2 V divided by the absolute value at bias of −2 V) approaches 10⁴, which is much higher than that reported before for diodes based on aligned CNT arrays [30].

The uniformity of CNT diode devices is an important merit for large area IR detector applications. The characteristics of the devices based on individual semiconducting CNTs are usually significantly different from each other due to the variations on CNT diameters and chiralities. However, the uniformity of devices based on CNT arrays with sufficient number of tubes in the channel is much better than based on a single CNT because the characteristics for the arrays are determined by the average among all the nanotubes participating in the transport. In this work, FETs and diodes with the same device dimensions exhibit almost the same characteristics, including the current on/off ratio, on state current, threshold voltage and rectification ratio. As shown in Fig. 3 , ten FETs (each single FET with L = 1.5 μm and W = 20 μm) and diodes (with L = 1.0 μm and W = 20 μm) were fabricated and connected in parallel. The device current for both the parallel connected 10 FETs and 10 diodes with PMMA coating are about 10 times that of the single device (see Figs. 3(a) and 3(d)). The optic image of the device formed by 10 diodes connected in parallel is showed in Fig. 3(e). Since there no CNT-CNT junctions for the parallel aligned nanotubes in the channel, the electrical breakdown process is highly controllable resulting in a high devices yield. This is important for successfully fabricating large scale or parallel devices.

 

Fig. 3 Transfer and output characteristics of ten parallel CNT FETs and diodes. (a) Transfer characteristics of a device composed of ten parallel connected p-type CNT thin film FET (with W = 20 μm and L = 1.5 μm) before and after electrical breakdown, and corresponding (b) output characteristics of the device. (c)Transfer characteristics and (d) I-V characteristics of a device (W = 20 μm and L = 1 μm) composed of ten parallel connected CNT diodes. (e) Optical image showing ten parallel CNT diodes device.

Download Full Size | PDF

Figure 4(a) shows an optical image of a real CNT array diode. A built-in electric field is formed along the nanotubes due to the asymmetrically contacted n-type (Sc) and p-type (Pd) electrodes, which is essential for a photodetector to efficiently separate photogenerated electrons and holes [17,20]. Because the channel length (~1 μm) is of the same length scale as the thickness of the back gate dielectric (SiO₂), the band bending of the nanotubes extends through the device due to the weak coupling between the channel and bake-gate. Consequently, the built-in field is spread over the whole channel which is highly desirable for electron-hole separation [31]. Under illumination, the photogenerated electron-hole pairs are separated by the built-in field and collected by the source and drain electrodes, resulting in photovoltage and photocurrent as shown in Fig. 4(b). In our photoresponse measurements, the Sc electrode is always grounded and the bias is applied on the Pd electrode. When the bias V is equal to the open circuit voltage V_OC, the photocurrent and the dark current cancel each other and result in a zero net current. The I_SC and V_OC of the device are 10.94 nA and 0.24 V respectively under 785 W/cm² illumination, and increased to 21 nA and 0.25 V when the power density increased to 1570 W/cm². The incident power density dependent I_SC and V_OC are shown in Fig. 4(c).

 

Fig. 4 Photovoltaic and electronic characteristics of a typical CNT thin film diode. (a) Optical image showing a CNT diode with channel length L = 1 μm. (b) I-V characteristics of the diode measured in dark and under illuminations. (c) Experimental data and fitted results for open circuit voltage and short circuit current as a function of illumination power density. (d) Photoresponse of the CNT IR detector under different illumination densities.

Download Full Size | PDF

While I_SC increases linearly with increasing power density, V_OC increases logarithmically with increasing incident light intensity, exhibiting typical photodiode behaviors. When compared with diodes based on individual CNT, about 1/20 illumination power density is needed for thin film diode to yield similar photocurrent and photovoltage, indicating that there remain approximate 20 semiconducting tubes in the device channel after electric breakdown. The more nanotubes present in the channel the more carriers and thus photocurrent will be produced under the same illumination density. This makes the photocurrent signal of a thin film diode more detectable under lower power density. From Fig. 4(c), it can be readily estimated that IR photo responses are detectable at 157 mW/cm², which is at least one order of magnitude lower than that of the diode based on a single CNT. Figure 4(d) shows the repeated photoresponse of the diode under the same power densities. We can see that I_SC of the diode is stable and repeatable during long time measurement, indicating that this device is reliable for application in photovoltaic IR detection.

One of the basic metrics for evaluating photovoltaic detector is external quantum efficiency (EQE), defined as the number of carriers produced per photon, i.e., _{$\eta =\left(I_{\text{ph}}/q\right)\left(h\upsilon /P_{\text{in}}\right)$} [32], where I_ph is the photocurrent or I_SC at the zero bias condition, q is electron charge, h is the Planck constant, υ is the frequency of light, and P_in is the incident power. The EQE η of the CNT array diode is estimated to be about 0.0156% for I_ph = 3.1 pA under 157 mW/cm². The relatively low level of EQE is ascribed to the small effective absorption cross section of the CNTs in the channel. If P_in is estimated by considering the total area of the CNT sections perpendicular to the incident direction as the actual effective absorptive area, η of the diode then raises to 10.4%, coinciding with previous reports [2].

To access the sensibility of the CNT array based photovoltaic IR detector, two parameters are calculated: the current responsivity _{$R=I_{\text{ph}}/P_{\text{in}}$} and the detectivity _{$D^{\text{*}}=R\sqrt{A}/i_{\text{n}}$} [32,33]. Here A is the active area of the detector and i_n is the root-mean square current noise per bandwidth. In the zero bias condition, I_ph is equal to I_SC. The maximum R of our experiments is ~9.87 × 10⁻⁵ A/W, which raises to 6.58 × 10⁻² A/W based on the actual area of CNTs.

There are three contributions to the noise powers i_n² that limit detectivity D*: thermal noise, shot noise from dark current and flicker noise. On the basis of the dark current characteristics of the CNT array diode, the total noise is primarily thermal in origin under zero bias condition. The noise may be given as _{$i_{\text{n}}=\sqrt{4k_{\text{B}}T/R_{\text{D}}}$}, where k_B is the Boltzmann constant, T is temperature, and R_D is zero-bias differential resistance of the diode, at room temperature T = 300 K and R_D ~1 GΩ, i_n is estimated to be 4.07 × 10⁻¹⁵ AHz^-1/2. As a result, the room temperature detectivity D* is calculated to be ~1.09 × 10⁷ cmHz^1/2/W, which outperforms the maximum detectivity of CNT-based bolometers which is of the order of 10⁶ cmHz^1/2/W [9,10].

While the CNT array diode shows promise for IR detection, the device may be further improved significantly by increasing the density of semiconducting CNTs (d_CNT) in the channel. First, the more CNTs exist in the channel, the more efficiently the photons will be absorbed and the weaker irradiation is needed to generate sufficient signal for detection. Second, under the same illumination, the average internal quantum efficiency of each CNT is similar and then the external quantum efficiency of the diode increases linearly with d_CNT. This suggests that the responsivity of the detector will be proportional to the number of the CNTs. On the other hand, the resistance of the diode R_D will be reduced with increasing CNTs in the channel, and the thermal noise is inversely proportional to R_D^1/2, thus the noise will be increased linearly with the d_CNT^1/2. Therefore, the detectivity D* is proportional to the d_CNT^1/2 when the noise is dominated by thermal noise. The density of CNTs can be raised by controllable growth or multiple transfers [25,34,35]. Here we just discuss a possible ideal growth result based on existing reports. If the CNTs reach 60 tubes per micron and 95% of them are semiconducting as mentioned above, the amount of active CNTs in the channel is about 60 times of the present device. It is believable that the responsivity of CNT IR detector would then be improved one or two orders of magnitude up to ~10 mA/W and the detectivity would be nearly one order of magnitude larger and reach ~10⁸ cmHz^1/2/W.

In summary, this report demonstrates CNT arrays based high performance IR detector with a responsivity of 9.87 × 10⁻⁵ A/W and a detectivity > 10⁷ cmHz^1/2/W. The doping-free process utilized for fabricating the CNT array diodes has compatibility with Si based technology, allowing wafer-scale CNT IR detector be integrated with CMOS integrated circuits, including readout circuits. This type of IR detector preserves the potential advantage desired for cost-effective, lightweight, and compact IR photodetectors, with additional advantages on simple device fabrication and un-cooled IR detection at room temperature.