50 MHz WSe₂ Photodetectors from Roll-to-Roll Exfoliated Films

Abstract

We investigate the time-resolved photodetection characteristics of multilayer WSe₂ films fabricated via roll-to-roll mechanical exfoliation. By systematically studying the bias and irradiance dependence of the photocurrent dynamics, and by employing spatially resolved photocurrent mapping, we provide insights into the mechanisms governing the device response. We demonstrate a pronounced irradiance dependence of the response times and reveal decay times of 9.8 ns as well as 3 dB electrical bandwidths exceeding 50 MHz, rendering these devices on par with high-quality, manually exfoliated WSe₂ photodetectors. These results establish roll-to-roll exfoliation as a powerful and scalable fabrication technique capable of producing high-speed photodetectors.

In the search for advanced photodetector materials, few have received as much attention as two-dimensional materials, particularly transition-metal dichalcogenides (TMDCs). They benefit from outstanding optical properties, such as a tunable bandgap , and high charge carrier mobilities, rendering them suitable materials for photodetection in high-speed optoelectronic applications. The main method to obtain high-performing TMDC photodetectors is via mechanical exfoliation, which can be used to produce single crystals with well-defined layer numbers. With this method, remarkable response times down to the nanosecond regime have been achieved. However, the exfoliation of two-dimensional materials is a poorly reproducible process, requiring manual selection of suitable flakes and challenging transfer steps. These requirements make the process hard to automate and pose substantial obstacles to scaling for industrial production. Consequently, alternative methods of the fabrication of varying TMDCs have been investigated, the most well-known methods being chemical vapor deposition (CVD), − pulsed laser deposition (PLD) − and liquid phase exfoliation (LPE) which, however, usually exhibit a higher density of defects compared to mechanical exfoliation. − Defects are usually detrimental to device speed, reducing charge carrier mobility , and introducing deep trap states, preventing fast excitonic recombination and prolonging the photocurrent decay. − Therefore, extrinsic response times of photodetectors obtained from scalable methods are typically slower, ranging from seconds to milliseconds, with only few devices reaching microseconds, which is still insufficient for high-speed optoelectronic applications like data transfer. Recently, an all-dry “roll-to-roll” mechanical exfoliation method has been reported to produce large area TMDC films with transfer yields of up to 75%. Using this approach, WSe₂ films have demonstrated mobilities of up to 3.4 cm²/(V s) and responsivities as high as 117 A/W. , Preliminary photoresponse measurements showed that photodetection was faster than the 20 ms detection limit of the employed setup.

Here, we present an in-depth study of the time-resolved photoresponse of roll-to-roll exfoliated, multilayered WSe₂ photodetectors. We analyze the bias- and irradiance-dependence of their photocurrent dynamics and apply photocurrent mapping to rationalize our results. We show that the large area WSe₂ photodetectors obtained by roll-to-roll exfoliation exhibit response times <10 ns and electrical bandwidths >50 MHz, rendering them competitive with manually exfoliated WSe₂.

Figure a displays a roll-to-roll exfoliated WSe₂ nanosheet network on a SiO₂/Si substrate with lithographically defined Cr/Au top contacts. Profilometry shows an average WSe₂ nanosheet film thickness of ≈100 nm. The Cr/Au (3/60 nm) electrodes were deposited by e-beam evaporation. Additional micrographs, film profilometry and Raman spectra of the material are provided in the Supporting Information (SI) (Figures S1–S3). The channel length and width of this two-terminal WSe₂ device are 20 and 140 μm, respectively. Consistent with ref ., the WSe₂ nanosheet film consists of ≈1–10 μm size flakes with >80% coverage.

1.

(a) Optical microscopy image of a roll-to-roll exfoliated WSe₂ film (5 transfers from roll-to-roll) with Cr/Au electrodes evaporated on top. (b) Source-Drain sweeps of the device shown in (a), with a 20 μm × 1 mm channel, in the dark (red) and under approximately 2 × 10^–2 W/cm² white light illumination (blue). (c) Irradiance dependence of the On/Off ratio (blue) and detectivity (orange) of a 20 μm × 1 mm channel device under 635 nm CW illumination and 1 V bias.

In Figure b, we examine the current vs voltage (I-V) characteristics of the device shown in Figure a. We find mostly symmetric, back-to-back diode-like transport characteristics in the dark and under white light illumination (irradiance ≈ 2 × 10^–2 W/cm²). We further report the On/Off ratio and specific detectivity under 635 nm excitation and 1 V bias, visualized in Figure c. The detectivity is obtained via

$$D^{*}=R·\frac{\sqrt{A}}{\sqrt{2·q·I_{\mathrm{dark}}}}$$ 1

with the responsivity $R=\frac{I_{\mathrm{light}}-I_{\mathrm{dark}}}{P_{\mathrm{opt}}}$ , area A, the elemental charge q, the current in the dark or under illumination I _dark/I _light, and the optical power P _opt. To a first-order approximation, we assume I _dark to be the dominant noise contribution. This method however can lead to an overestimation of the detectivity, if a different noise type is predominant. To address this problem, a consensus statement by several authors of the photodetector community has proposed to report the specific detectivity in a standardized manner, using

$$D^{*}=D\sqrt{A·B}$$ 2

Herein A is the illuminated channel area and B is 1.57 × f _3 dB if a first-order low-pass filter characteristic is assumed. The detectivity D is defined by D = 1/NEP, where NEP is the noise equivalent power, at which the signal-to-noise-ratio (SNR) is P _signal/P _noise = 1. By extrapolation from the performed measurements, we estimate a NEP of 53 nW, resulting in a D* of 1.5 × 10⁷ Jones. For further information on the methodology, we refer to the Supporting Information (Figure S4).

Varying the irradiance from 10^–4 to 10² W/cm² results in the common increase of the On/Off ratio at the expense of a lower detectivity, reflecting the general trend toward photosaturation. For details regarding the data acquisition, irradiance-dependence of photocurrent, responsivity and external quantum yield (EQE) as well as spectrally resolved responsivity, the reader is referred to the Supporting Information (Figures S5–S7).

In Figure a we display the time-resolved photoresponse measurements with 635 nm light pulses, revealing bias-dependent fall times between 10 and 22 μs. The complete data set, including rise- and fall flanks, is provided in the Supporting Information (Figure S8a). In Figure b, higher irradiance yields faster photoresponse, changing by approximately 2 orders of magnitude over the same range of irradiance. The irradiance-dependent measurements are consistent with earlier reports showing that the response time in TMDC photodetectors is typically RC-limited, as we observe shorter fall times at higher irradiances. For irradiances ≥10 kW/cm², we need to increase the repetition rate to 100 kHz to fully resolve the response dynamics. Comparative photocurrent measurements are provided in the Supporting Information (Figure S8b). The saturation of the photocurrent was verified for a comparative device at repetition rates of 100 Hz up to 500 kHz (see Figure S9 and Table S1). A detailed summary of the results depicted in Figure b is provided in the Supporting Information (Tables S2 and S3).

2.

(a) Transient photocurrent response (at the fall segment) of a 25 μm × 1 mm channel device under 635 nm square-pulse excitation (1 kHz repetition rate, irradiance ≈2 × 10^–3 W/cm²) at different bias voltages. The 10–90% thresholds used to extract the fall time are indicated by dashed lines. (b) Irradiance dependence of the rise (green) and fall times (blue) for a 40 μm × 1 mm channel at 3 V bias under 635 nm excitation and 10 kHz repetition rate, with an inset showing additional measurements conducted at a 100 kHz. The dashed horizontal line at ∼100 ns indicates the instrumental resolution limit at 10 kHz. Measurements in (a) and (b) were performed on different optical setups (probe station vs confocal microscope), resulting in differing power densities. Experimental details are provided in the SI.

No correlation between channel length and photoresponse has been observed for measurements with channel lengths between 5 and 80 μm. An overview of the temperature- and channel-length-dependent measurements can be found in the Supporting Information (Figures S10 and S11).

To further investigate the apparent lack of channel-length dependence and the counterintuitive increase of the fall time with increasing bias (Figure a), we perform photocurrent mapping (Figure ). A linear-scale representation of the absolute photocurrents is provided in the Supporting Information (Figure S12). For small biases (±0.1 V), we find that the majority of the photocurrent is generated in close proximity to the metal contacts. We note a significant photocurrent even at zero bias, suggesting a Schottky-type contact (in good agreement with the observed back-to-back diode-like I–V characteristics). While certain hotspots and an uneven distribution of generated photocarriers along the metal/WSe₂ interface are evident, there is no obvious correlation with the flake-to-flake boundaries in the film which we display in terms of its scattering image from Figure a. At higher fields (±1 V), the photocurrent generation extends over the entire channel. Under such high field conditions, a significant number of photocarriers generated far away from the metal contacts can be efficiently collected, diffusing across multiple flake-to-flake boundaries and long percolative paths. We suggest that this may be the origin of the longer fall times at large biases in Figure a. Additional photocurrent mapping performed on a second device is presented in the Supporting Information (Figure S13), demonstrating comparable field-dependent behavior.

3.

Scattering image of the channel depicted in Figure a with overlaid photocurrent maps for several bias-voltages. To enable uniform scaling across all subfigures, a log­(1 + I _Photo/pA) function is applied to the current values. Measurements were performed with a wavelength of 532 nm (P _opt,532 nm ≈ 1.4 mW at 78 MHz repetition rate) chopped to a 700 Hz frequency.

To assess the potential of roll-to-roll exfoliated WSe₂ for fast photonic data transfer, we analyze the fall time after a ∼ 50 ps laser impulse in Figure a. From the time-dependent photocurrent I(t) recorded during this measurement, we obtain the power spectrum via Fourier-transformation according to P(ω) = |FFT­(I(t))|² which affords the electrical bandwidth at 3 dB attenuation (f _3 dB). An exemplifying measurement exhibits a fall time of 9.8 ns and yields a 3 dB cutoff frequency of f _3 dB = 53.5 MHz. Measurements demonstrating a complete decay of the impulse photoresponse for pulse repetition rates between 150 kHz and 2 MHz are provided in Figure S14 and Table S4. A schematic of the experimental setup (Figure S15), along with a table summarizing the individual electrical components and their respective speed limitations, is provided in Table S5. The reproducibility of the high-speed switching was verified for five additional devices, and a statistical overview of the results is depicted in Figures S16 and S17. For all of these data sets, we refer to the Supporting Information.

4.

Time-resolved photocurrent measurements of a 30 μm × 1 mm channel. (a) Power spectrum of the device, obtained via Fourier transformation of the impulse photoresponse shown in the inset. The measurement was performed with a 636 nm impulse laser at 1 MHz repetition rate under a 1 V bias. The 3 dB bandwidth (53.5 MHz) is indicated by the dashed lines. The average irradiance is ≈2 × 10³ W/cm². (b) Square-pulse measurement conducted with a 635 nm square pulse illumination (665 μW focused to approximately 1 μm spot diameter, irradiance ≈8 × 10⁴ W/cm²) at a repetition rate of 500 kHz and 3 V bias voltage. The 10% and 90% thresholds are shown with dashed lines.

Discussion

We contextualize our results by comparison with other large-scale fabrication approaches, such as CVD and LPE, as well as manually exfoliated (small-area) devices in Figure . This overview highlights that roll-to-roll mechanically exfoliated WSe₂ affords response times that are competitive even with manually exfoliated devices (made from single-crystal flakes), however with the unique advantage of rapid, semiautomatic, and scalable device production up to the wafer scale. We acknowledge that this method produces bulk-like multilayers, while manual exfoliation often focuses on mono/few-layer devices to take advantage of a direct bandgap, lower out-of-plane drift and reduced interlayer diffusion. In practice, however, bulk TMDCs have also been shown to offer distinct advantages for fast photodetection, such as a reduced influence of surface defects, higher absorption rates and higher charge carrier mobilities. , This is reflected in the responsivity, detectivity, response times, and f _3 dB values reported here, which compare favorably with the large body of TMDC-photodetector literature. − ,,−

5.

Comparison of responsivity and response time between roll-to-roll exfoliated devices and other large-scale fabrication methods for TMDCs. Herein, the slowest reported rise- or fall time was taken as the response time, as this was considered the limiting factor for the device. In this representation, devices from different TMDC materials are included. A detailed literature review, including details on the materials, is presented in the Supporting Information (Table S6). − ,,− The responsivity values for this work (roll-to roll) were calculated from Figure a (*) and Figure b (**). The applied irradiances differ considerably (2 × 10^–3 W/cm² and 8 × 10⁴ W/cm², for Figure a and Figure b, respectively), demonstrating a trade-off between the measured device speed and responsivity.

An inherent draw-back of roll-to-roll exfoliation is the inhomogeneous, interconnected flake network nature of the TMDC film, whichas we suggestinvokes the negative correlation between bias voltage and response time. In a homogeneous photodetector, increasing the bias voltage typically results in faster rise- and fall times. Future fabrication refinements should therefore focus on even more homogeneous films with larger grain sizes. Furthermore, the apparent Schottky-nature of the metal/WSe₂ interface revealed by I-V characteristics (Figure b) and photocurrent mapping (Figure ) could be improved.

The pronounced correlation of response time with irradiance (Figure b) is often attributed to a decrease in photoresistance at higher irradiances, which reduces the RC time and leads to a trade-off between response speed and responsivity or photogain. This highlights the importance of providing intensity/irradiance-dependent data for meaningful comparisons between different materials. While the large irradiances used here to achieve the exceptionally fast response times in Figure b may be impractical for certain applications, we stress that this is only necessary to maintain the full photocurrent intensity. At lower irradiances, a significant portion of the photocurrent still decays at these fast time scales (cf. Figure S18), in addition to a slow component, for instance due to RC limitation. Consequently, the fast response times are largely retained at the expense of a photocurrent signal reduced to approximately 70%, which may still be sufficient for some applications.

Our measurements show a sublinear correlation between the response times and responsivity whitin the applied irradiances, observable in Figure c) and in the SI (Table S2). This behavior is consistent with earlier reports, and attributable to faster recombination rates at higher exciton densities − as well as a variety of further effects. ,, Our measurements also operate at high optical powers, where photogating, an explanation for high responsivities at low optical irradiances, is suppressed and response are speeds governed mainly by RC-time reductions. ,

In conclusion, we report decay times of 9.8 ns and 3 dB-bandwidths of ≈50 MHz for WSe₂ photodetectors obtained by roll-to-roll mechanical exfoliated, all-dry transfer. We observe a strong irradiance dependence of the photocurrent dynamics and an inverse correlation between the applied bias and the speed of photodetection. We attribute the irradiance dependence to a decrease in photo resistance, and the bias dependence to an inhomogeneous film composition that leads to localized hotspots of charge carrier generation. Our results show that roll-to-roll exfoliation is a promising, automatization-suitable method for fabricating large-area TMDC photodetectors with high-speed photoresponse, which can be further optimized to address the intrinsic trade-off between responsivity/detectivity and speed.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.6c01579.

• Methods and materials, optical micrographs, film profilometry, Raman characterization, irradiance-dependent photocurrent measurements, photocurrent spectroscopy, bias- and irradiance dependent time-resolved measurements, comparative time-resolved data at different laser repetition rates, temperature-dependent rise- and fall times, channel-length dependent rise- and fall times, additional material regarding scanning photocurrent microscopy, tabular literature review regarding time-resolved properties of large-scale TMDC photodetectors (PDF)

P.M. and W.R. contributed equally. W.R. and P.M.: optical and optoelectronic measurements, optical lithography, original draft, and visualization. Y.S.: roll-to-roll exfoliation and optical lithography, writing review. M.S.: funding acquisition and project administration. M.S. and A.C.-G.: writing review, editing, and supervision. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.