Report

All-printed thin-film transistors from networks of liquid-exfoliated nanosheets

See allHide authors and affiliations

Science  07 Apr 2017:
Vol. 356, Issue 6333, pp. 69-73
DOI: 10.1126/science.aal4062

Printing nanosheet-network transistors

Two-dimensional (2D) materials such as graphene and metal chalcogenides such as tungsten diselenide (WSe2) are attractive for use in low-cost thin-film transistors (TFTs) because they have high charge-carrier mobility. Kelly et al. printed TFTs from networks of exfoliated dispersions of 2D materials with graphene contacts, WSe2 as the semiconductor, and a boron nitride separator. Electrolytic gating with ionic liquids enabled higher operating currents than achieved with comparable organic TFTs.

Science, this issue p. 69

Abstract

All-printed transistors consisting of interconnected networks of various types of two-dimensional nanosheets are an important goal in nanoscience. Using electrolytic gating, we demonstrate all-printed, vertically stacked transistors with graphene source, drain, and gate electrodes, a transition metal dichalcogenide channel, and a boron nitride (BN) separator, all formed from nanosheet networks. The BN network contains an ionic liquid within its porous interior that allows electrolytic gating in a solid-like structure. Nanosheet network channels display on:off ratios of up to 600, transconductances exceeding 5 millisiemens, and mobilities of >0.1 square centimeters per volt per second. Unusually, the on-currents scaled with network thickness and volumetric capacitance. In contrast to other devices with comparable mobility, large capacitances, while hindering switching speeds, allow these devices to carry higher currents at relatively low drive voltages.

The development of printed electronics is becoming increasingly important, with much research focusing on developing new materials. A number of material sets have been studied, including organics (1), inorganic nanoparticles (2) and nanotube/nanowire networks (3). High operating voltages (up to 50 V), low mobility (<10 cm2 V–1 s–1), and poor current injection remain challenges for organic thin-film transistors (TFTs) (1, 4). Networks of inorganic nanoparticles or nanotubes have demonstrated mobilities and on:off ratios of >10 cm2 V–1 s–1 and >106, respectively (2, 3), but may incur problems of scalability and integration.

We argue that fabricating TFTs from printed networks of two-dimensional nanosheets will yield several advantages over organics and nanoparticles. Nanosheets can be conducting (e.g., graphene or MXenes), semiconducting (e.g., transition metal chalcogenides or phosphorene) or insulating (e.g., boron nitride or silicates), and so include all of the building blocks of electronics (5). A wide variety of nanosheet inks can be produced cheaply by liquid-phase exfoliation (5) and can easily be printed into networks (6). Both conducting and semiconducting nanosheets have high intrinsic mobilities (7), with network mobility limited by junction resistances, a constraint that should be addressable via junction engineering (8). Nanosheet inks (5) could be used to print both in-plane and stacked heterostructured nanosheet networks (6, 9) consisting of conducting, semiconducting, and insulating regions corresponding to electrodes, channel, and dielectric, respectively.

Here, we describe the fabrication and characterization of working nanosheet network transistors (10). We prepared suspensions of nanosheets of MoS2, MoSe2, WS2, and WSe2 by liquid-phase exfoliation in N-methyl 2-pyrrolidone (11). To reduce spatial variations in local band gap, we performed size selection to remove the thinnest of the variable–band gap (12) nanosheets before solvent exchange to isopropanol (Fig. 1A). Transmission electron microscopy (TEM) analysis showed that all dispersions contained two-dimensional (2D) nanosheets (Fig. 1B). Optical absorption spectra (measured with an integrating sphere to remove scattering) (13) confirmed that all materials were semiconductors, with optical gaps between 1.3 and 1.8 eV (Fig. 1C). Atomic force microscopy (AFM) measurements (Fig. 1, D and E) showed that typically >85% of nanosheets had N > 5 layers and therefore had bulk-like electronic structure (14). For all materials, the mean nanosheet lengths and thicknesses were in the range of were 330 to 380 nm and 13 to 17 layers, respectively.

Fig. 1 Basic characterization of nanosheets and nanosheet networks.

(A) Photo of dispersions of MoS2, MoSe2, WS2, and WSe2 (C ~ 0.2 mg/ml). (B) Typical TEM image of liquid-exfoliated WSe2 nanosheets. (C) Optical absorption spectra (extinction minus scattering) measured on nanosheet dispersions (C ~ 0.005 mg/ml). (D) Plot of nanosheet length l versus thickness (layer number N) for all materials. The horizontal line approximately separates thinner nanosheets with N-dependent band gap from thicker ones with bulk-like band gap. Inset: Typical AFM image. (E) Mean nanosheet thickness versus mean nanosheet length. (F) Typical scanning electron microscopy (SEM) images of a sprayed network of WSe2 nanosheets. (G) Raman spectra measured on networks of all four materials. (H) Measured network density plotted versus nanosheet density; the resultant porosity values P are indicated. (I) Conductivity versus inverse temperature for all four materials (no ionic liquid). Color codes for MoSe2, WSe2, MoS2, and WS2 are the same in all graphs. In (E) and (H), error bars denote SE and measurement error, respectively.

Initially, the nanosheet dispersions were sprayed onto flexible alumina-coated polyethylene-terephthalate (PET) substrates to form porous nanosheet networks (PNNs). Such networks appeared uniform over length scales greater than ~10 μm but displayed considerable local disorder (Fig. 1F). Raman spectroscopy confirmed the nanosheet type, and density measurements revealed relatively large porosities of 43 to 63% (Fig. 1, G and H). We measured the electrical conductivity of PNNs of all four materials as a function of temperature (Fig. 1I); the conductivity fell with decreasing temperature in all cases, consistent with thermal activation (15), probably associated with activated inter-nanosheet hopping.

Although transistors made from carbon nanotube networks are well known (3), attempts to produce dielectrically gated TFTs made from nanosheet networks have resulted in poor switching (16), probably as a result of electrostatic screening effects. Here, we exploit the inherent porosity of nanosheet networks to fabricate electrochemical TFTs, where the PNNs are gated electrolytically. In an electrolytically gated TFT (Fig. 2A) (17), the gate dielectric is replaced with a liquid electrolyte, in this case the ionic liquid (IL) 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (EMIm TFSI), placed as a droplet on top of the PNN. Applying a gate voltage drives ions to the electrolyte–active material interface, which in turn draws charge from the external circuit to create a double layer. This charge accumulation mechanism results in strong drain-current modulation (17).

Fig. 2 Characterization of porous nanosheet network thin-film transistors (TFTs).

(A) Schematic of a TFT gated using the ionic liquid 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (EMIm TFSI). Electrodes were lithographically patterned gold with channel length and width L = 120 μm and w = 16 mm. (B) Examples of transfer characteristics for TFTs fabricated from WS2 and MoSe2; PNN thicknesses t are indicated. (C and D) Mean threshold voltages (C) and maximum observed on:off ratios (D) for all four materials. In (C), observed p- or n-type behavior is indicated. (E) Measured transconductance plotted versus network thickness. The lines are linear fits; the resultant μNetCV values (in units of F m−1 V−1 s−1) are indicated. (F) Examples of cyclic voltammograms measured for MoS2 PNNs deposited on ITO-coated glass, using EMIm TFSI as the electrolyte. (G) Areal capacitance, extracted from curves such as those in (F), plotted versus PNN thickness. The lines are linear fits; the resultant CV values are indicated. (H) Network mobility μNet plotted versus nanosheet mobility μNS (as measured by THz spectroscopy). Inset: μNSNet ratio plotted versus μNS. (I) Mobility plotted versus areal capacitance for TFTs from previously reported data (10) and compared to our thickest PNNs. The dashed lines show contours of constant μ(C/A) (in units of F V−1 s−1). DG, dielectrically gated; EG, electrochemically gated.

Here, we demonstrate PNN-based TFTs with a liquid electrolyte penetrating the internal free volume, which enables switching throughout the network so that thick networks can be switched as effectively as thin ones. Shown in Fig. 2B are example transfer curves for electrolytically gated TFTs fabricated from MoSe2 and WS2 PNNs. In each case (10), we observed transistor action with both p- and n-type behavior (depending on the material), low threshold voltages Vt between –1.3 and 1.0 V, large on-currents exceeding 1 mA, and maximum on:off ratios of ~600 (Fig. 2, C and D). The on:off ratios were limited by high off-currents because of doping associated with the IL, which was present even with zero gate bias, possibly as a result of ion-selective binding. In addition, some hysteresis was observed, as is typical for electrolytically gated TFTs (17).

We found that the transconductance gm = (∂Ids/Vg)Max (where Ids is the source-drain current and Vg is the gate voltage) increased linearly with PNN thickness for all materials (Fig. 2E), reaching values as high as 6 mS; this is competitive with the highest values reported to date (18). Such thickness dependence is not typical of TFTs because switching is usually an interfacial effect such that gm = μCAVdsw/L, where μ is the channel mobility, CA is the areal dielectric capacitance, Vds is the source-drain voltage, and w and L are the channel width and length, respectively (19). However, in an electrolytically gated porous network, ions can adsorb onto the internal surface, thus switching the entire network volume. Under these circumstances (10, 20) the transconductance is given byEmbedded Image(1)where μNet is the network mobility, CV is the volumetric capacitance, and t is the network thickness. This is important as it means both Ids and gm can be tuned via the network thickness and volumetric capacitance as well as the mobility. Unlike polymer-based electrochemical TFTs, here CV can be tuned via the nanosheet thickness (21). Additionally, this equation implies that μNetCV, which can be found from the gm versus t data in Fig. 2E, is a figure of merit for PNNs.

Equation 1 assumes that the network areal capacitance scales with thickness as C/A = CVt. Cyclic voltammetry measurements (Fig. 2F) for PNNs of different thickness confirmed this, giving CV ~ 1 F/cm3 (Fig. 2G) (21). Combining these CV values with the μNetCV data yielded mobilities between 0.08 and 0.22 cm2 V−1 s−1. For comparison, we measured the intrinsic nanosheet mobilities μNS by optical-pump terahertz probe (THz) spectroscopy (22), finding values between 47 and 91 cm2 V−1 s−1, with μNet plotted versus μNS in Fig. 2H. The ratio μNSNet increases with increasing μNS, consistent with internanosheet junction-limited transport (10), Mobility improvements could be achieved by increasing flake size and network connectivity and reducing the intersheet resistance.

Although limited by junctions, these network mobilities are much larger than those of early (undoped) conjugated polymers (23) and are competitive with many recent organic TFTs (Fig. 2I) (24). We can benchmark our PNN-TFTs with reported TFTs by comparing the product μ(C/A), as we have done in Fig. 2I. Here, the best-performing TFTs are at the upper right, and the dashed lines represent constant contours of μ(C/A). By this measure, our thickest PNN-TFTs are competitive with benchmark TFTs because high values of C/A compensate for the relatively low mobilities.

Although high capacitances facilitate transport, they also increase switching time. Switching time can be approximated as the RC time constant of the network/electrolyte combination, which has terms associated with electronic and ionic transport (10, 25):Embedded Image(2)where σNet and σIL are the network and IL conductivities and d is the length scale associated with ion transport. By applying a square-wave gate voltage and measuring the source-drain current Ids (Fig. 3, A and B), we found τ = 75 to 100 ms for WSe2 TFTs with different channel lengths L between 80 and 240 μm gated using a droplet of IL (Fig. 3C). The dependence on L is consistent with Eq. 2 and shows these TFTs to be limited by ion transport. This implies that switching speed can be increased by either reducing CVt (at the cost of diminished Ids and gm) or by decreasing d/σIL, perhaps by using very thin solid electrolytes.

Fig. 3 Characterization of switching speed of WSe2 TFTs.

Network thickness t ~ 400 nm and channel length L = 120 μm, unless otherwise stated. (A and B) Square-wave gate voltage (Vg, frequency 1 Hz) and resultant source-drain current (Ids) as a function of time for an ionic liquid (IL) gate. (C) Time constant (from rise time) versus L2 for an IL gate. Error bars denote SD. (D and E) Vg and Ids versus time for an IL/polymer gel gate. (F and G) Photographs showing (F) a PNN spray-deposited under an interdigitated gold electrode array (s, source; d, drain) and (G) a similar device with a spray-deposited boron nitride PNN (white, ~2 μm thick, marked BN) upon which a gold top gate electrode (g) was evaporated. The IL was inserted into the porous volume of the device by drop-casting. (H) SEM of the BN network. (I and J) Vg and Ids versus time for a BN-contained IL gate. In (B), (E), and (J), the time constant 〈τ〉 associated with the current rise is shown. (K) Source-drain current on:off ratio plotted versus gate voltage on-off switching frequency.

However, solid electrolytes based on polymers or gels usually have low ionic mobilities, further reducing the switching speed (17, 26). This effect is illustrated in Fig. 3, D and E, where the use of an IL/polymer-based gel increased switching time by a factor of 10 relative to the liquid IL. We resolved this problem by spraying a PNN of BN nanosheets (porosity 60%) (10) on top of our active layer, not as a dielectric but as an electrochemical separator between the active layer and an evaporated gold top gate (Fig. 3, F to H). Drops of IL placed on the resultant heterostructure then filled the entire porous free volume. This gave a solid-like structure containing highly mobile ions, allowing fast, liquid-like gating (Fig. 3, I and J). Comparing the effective on:off ratios measured at different switching frequencies for a liquid IL, an IL gel, and an IL contained within the network showed the latter system to perform almost as well as pure IL and considerably better than the IL gel (Fig. 3K).

This BN separator allows us to develop a solid-like, vertically stacked, all-printed, all-nanosheet, electrolytically gated TFT (Fig. 4A). We produced this device by inkjet-printing interdigitated graphene electrodes (~400 nm thick) with a printed WSe2 (chosen for its relatively high conductivity) active channel (t ~ 1 μm, L = 200 μm, and w = 15.6 mm). A BN PNN (~8 μm thick) was sprayed on top of the channel, followed by an inkjet-printed graphene top gate (~400 nm thick; Fig. 4, B to D). The IL was drop-cast onto the network and rapidly wicked into the porous volume (Fig. 4E). This all-printed PNN-TFT performed reasonably well with on:off ratios of >25 and gm = 22 μS (Fig. 4F).

Fig. 4 All-printed, all-nanosheet TFT.

(A) Schematic showing all-printed TFT structure. The source, drain, and gate electrodes are inkjet-printed networks of graphene nanosheets; the channel is an inkjet-printed network of WSe2 nanosheets. The gate electrode is separated from the channel by a spray-cast BN nanosheet network. The entire porous volume of the structure is filled with an ionic liquid to facilitate electrolytic gating. (B) Photographs of the printing steps. From left to right: Graphene source (s) and drain (d) electrode (t ~ 400 nm); the WSe2 channel (t ~ 1 μm, L = 200 μm, w = 16 mm); the BN separator (t ~ 8 μm); and finally the graphene gate (g, t ~ 400 nm). (C) A flexible array of printed TFTs. (D) Cross-sectional SEM image showing WSe2 channel and BN separator. (E) Magnified image of BN network showing porosity (P = 60%). (F) Transfer curves for a printed TFT with a WSe2 active channel after cycling the gate voltage 1, 10, 25, and 50 times.

Although we have demonstrated nanosheet network transistors, much work remains. For example, it will be important to improve control of network morphology and connectivity with the aim of substantially enhancing the network mobility. In addition, optimizing the ionic liquid to improve the on:off ratio and switching speed will be essential. With such improvements, we believe nanosheet network transistors can challenge their organic and nanotube-based counterparts on both performance and ease of fabrication.

Supplementary Materials

www.sciencemag.org/content/356/6333/69/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S17

Tables S1 and S2

References (2757)

References and Notes

  1. See supplementary materials.
  2. Acknowledgments: Supported by the Foundation for Fundamental Research on Matter (FOM), part of the Netherlands Organization for Scientific Research (NWO) (L.D.A.S.); FOM grant 67595 (A.K.); Toyota Motor Europe (J.L.); Science Foundation Ireland grants SFI/11/PI/1087 and SFI/12/RC/2278; European Commission grant 696656 (Graphene Flagship); and European Research Council grant FUTURE-PRINT. The invention described in this report was the subject of a patent filing at the European Patent Office in March 2017 (inventors, J.N.C., A.G.K., T.H., and G.S.D.; assignee, Trinity College Dublin).
View Abstract

Stay Connected to Science

Navigate This Article