Highly efficient UV detection in a metal–semiconductor–metal detector with epigraphene

Detectors of ultraviolet radiation (UV, λ = 10–400 nm) have proven to be useful in a number of applications, including military, industrial, environmental, and health sectors.^1–3 The pursuit of high-performance UV detectors remains attractive for applications such as non-line-of-sight (NLoS) communications^4–6 and UV astronomy.^7–9 Quantum noise limited UV photon counting receivers would enable high data transfer rates on short to medium NLoS ranges, exploiting the nearly zero UV background conditions on ground level.¹⁰ Astronomy in the UV range, on the other hand, can play a crucial role in understanding the chemical evolution of the universe and provide evidence for dark energy in cosmology studies.^11,12 To achieve the ultimate sensitivity in the UV range, these and other applications require solar (or visible) blind detectors, which should be insensitive to light with wavelength λ > 400 nm. A preferred route to achieve solar blindness in photovoltaic solid-state detectors is by using wide bandgap semiconductors (E_G$≳$3 eV) as the core of UV detection. A large E_G imposes constraints on photon energies E_ph for efficient absorption and electron–hole pair generation, removing the need to use optical filters in practical detectors.

Silicon carbide (SiC) is one of the most studied wide-bandgap materials (E_G = 3.26 eV for the 4H polytype at 300 K), and since the first observation of UV detection in SiC p–n junctions,¹³ it has become a preferred choice for UV photodetection, providing practical advantages such as operation in harsh environments and high temperatures.^13–16 Conventional SiC UV detectors are based on UV photons being absorbed in the bulk of SiC, producing electron–hole pairs that can be separated by an electric field, and thereby generating an external current that is proportional to the number of absorbed photons. The technology of SiC-based UV detectors has been optimized over time employing different architectures^17–19 with the goal of improving the responsivity and noise performance. The responsivity R quantifies the amount of photocurrent I_PH generated under illumination at a given optical power P_OPT, with R = I_PH/P_OPT = ηe/(hf) (A/W), where η is the external quantum efficiency, e is the electron charge, h is Planck's constant, and f is the frequency of the incoming radiation. The sensitivity of the detector is limited by noise, and the noise-equivalent-power, NEP = S_N/R (W/√Hz), provides a metric for the ultimate practically detectable signal, with S_N being the current noise spectral density S_N (A/√Hz). Taking the effective area of the detector A into account, the specific detectivity serves the purpose of enabling the performance comparison among different detector technologies, and it is defined as D = √A/NEP²⁰ in units of Jones (cm√Hz/W).

With the advent of epitaxial growth of graphene on SiC (epigraphene), UV detection with SiC has been demonstrated using the epigraphene layer as a transparent contact.^21,22 Graphene-SiC UV detectors have been demonstrated, which display responsivities of the order of 10 mA/W at λ = 254 nm, corresponding to rather modest external quantum efficiencies of η ∼ 4%.²¹ Greater responsivities have been achieved at the expense of adding complexity to the devices, e.g. in vertical p-n mesa junctions formed by epilayers of SiC.²² Yet, remaining challenges with epigraphene-based UV detectors include avoiding the formation of stacking faults in the SiC epilayers,²³ avoiding the need for large bias operation²⁴ and minimizing dark currents.²⁵ 

Here, we present a graphene-based UV detector with a peak external quantum efficiency of η ∼85% for wavelengths 250–280 nm, corresponding to nearly 100% internal quantum efficiency when accounting for reflectance losses in the material (reported experimental reflectance for 4H-SiC is in the range 15%–30%^26,27). The device is a metal–semiconductor–metal (MSM) structure, and the high performance is based on our capability to form a reproducible junction between the high-purity semi insulating (HPSI) 4H-SiC substrate and the graphene layer grown atop. Electrically, an MSM structure is formed by two Schottky diodes connected back-to-back [Fig. 1(a)]. Under an applied voltage, irrespective of the polarity of the bias, at least one of the diodes remains reversed-biased, so that a small saturation current I_o flows through the device. Upon illumination, the excess current produced by the photo-generated carriers, i.e. the photocurrent I_PH, can exceed I_o by many orders of magnitude, making it easily detectable. In our case, the saturation currents in the dark conditions can be as low as I_o ≈ 50 fA, and the optical power from a UV LED source can produce I_PH = 1 μA (ON/OFF ratio >10⁷). A practical advantage of MSM photodetectors is the simplicity in manufacturing, and in our case, the high performance of the devices has been observed on substrates just after the growth of epigraphene, by making contacts to SiC and graphene using wedge bonding, without any need for microfabrication. However, for careful characterization, the results presented in this work were collected on devices having a well-defined geometry achieved by standard e-beam lithography. In total, two graphene substrates, grown on SiC from different commercial suppliers, were studied in detail by quantifying the responsivity, effective area, noise, and speed of response.

Epigraphene was grown on the silicon-terminated face of high-purity semi-insulating (HPSI) 4H-SiC substrates (Cree Inc. and Norstel AB, bulk doping of the order of N_D ∼10¹⁴ cm⁻³ Ref. 28) by encasing them in a graphite crucible, in an inert argon atmosphere of 850 mbar, and heating using RF power. This growth method results in high quality monolayer epigraphene.^29,30 For UV detection, we have optimized the growth temperatures T >1850 °C using a design of experiments (DOE) methodology.³¹ Raman spectroscopy and transmission mode microscopy were used to quantify the content of bilayer graphene in the range of 10%–20% for the substrates used in our experiments³² (see supplementary material, S1). Device fabrication used standard electron beam lithography techniques. Epigraphene was patterned using oxygen plasma etching, and metal contacts were deposited using physical vapor deposition of 5 nm Ti and 80 nm Au.^33,34 Optical characterization of the substrates and spectral responsivity measurements were taken using an Agilent Cary 60 spectrophotometer in the range 180–1100 nm. Electrical characterization was performed under ambient illumination conditions at room temperature using a Variable Gain Sub Femto Ampère Current Amplifier DDPCA-300 and a Variable Gain High-Speed Current Amplifier DHPCA-100 from Femto Messtechnik GmbH. The samples were illuminated in two different ways: (1) by a deep UV (DUV) flood exposure unit from Bachur & Associates, equipped with a DUV 220 nm mirror, primarily reflecting radiation with λ = 220–260 nm, providing an optical power density up to p = 10 mW/cm² and (2) a continuous LED light source (λ = 250 nm) LLS-250 from Ocean Optics fiber-coupled to a ball lens to provide a light spot of diameter about D ∼1.5 mm, providing an optical power up to P = 6 μW (p = 0.3 mW/cm²). In all cases, the light intensity was calibrated using both a silicon photodiode power sensor S130VC from Thorlabs and an optical power meter OE-200-UV from Femto Messtechnik GmbH.

Figure 1(b) shows the schematic of the measurement setup for device A, which consists of a 5 × 5 mm² epigraphene square. Electrical contacts are placed around the epigraphene perimeter, with additional contacts also deposited directly onto the SiC substrate. A large area sample ensures that the UV light-spot falls entirely inside the epigraphene area during optical characterization, so that all the light, except for the reflection losses, is coupled to the device. Figure 1(c) shows the current–voltage (I–V) characteristic for this device both in the dark and under illumination with λ = 250 nm.

The dark current for this square geometry device is below pA in the bias range V_B = ±10 V, being as low as I_OFF = 52 fA for V_B = 10 V (epigraphene as anode). In the absence of light, the zero-bias differential resistance measured between epigraphene and the contacts on SiC is R₀ = dV/dI ∼20 TΩ.

Under illumination (P_opt = 6 μW), the current measured at V_B = 10 V is I_ON = 0.8 μA, yielding a current on-off ratio I_ON/I_OFF ∼1.54 × 10⁷. The shape of the I–V curve is that of an MSM photodetector with asymmetric Schottky contacts, where the epigraphene is the anode and Ti/Au is the cathode. In the light on case, the I–V curve clearly exhibits two distinct operating regions, referred to as photoconductive and saturation regimes.³⁵ The photoconductive regime is the low forward bias with a steep sloped region of the I–V curve, while the saturation at high bias is due to all photogenerated carriers being swept out by the electric field.

A quantitative analysis of the device responsivity to UV photons requires knowledge of the active area of the device, to quantify the optical power available for photocarrier generation. The effective area of the device was investigated by scanning the light-spot across the entire chip while keeping a constant bias V_B = 10 V. Figure 1(d) shows that the effective photo-detecting area unambiguously corresponds to the region where the epigraphene contact covers the surface of SiC.

Having established the effective area, the spectral responsivity of the device was studied over the wavelength range λ = 200–1000 nm. Figure 1(e) shows the responsivity peaking at λ = 250–280 nm, with a maximum measured value about R = 175 mA/W, corresponding to an external quantum efficiency η ∼85%. The responsivity is drastically reduced for λ >400 nm (see supplementary material, S2), owing to the large bandgap of 4H-SiC, and results in a visible rejection ratio of at least 1400, which we define as the ratio of the responsivity at λ = 250 nm to that at λ = 400 nm.

Of practical relevance is that the responsivity of the device remains large even at zero bias, making this a self-powered detector. Figure 2(a) shows the I–V response of the detector for different UV intensities. The linearity of photocurrent in the device extracted from these measurements is shown in Fig. 2(b) for two different biases. Typical for MSM detectors, a larger forward bias leads to a higher responsivity, due to an increase in size of the depletion width of the reverse-biased contact. The zero bias operation is possible, thanks to the asymmetric contacts, i.e., contacts with dissimilar work functions, which give rise to a built-in bias in the detector in the absence of externally applied voltage (for device B with symmetric Schottky barriers, see supplementary material, S3).

We explore the device performance in terms of speed and detectivity (Fig. 3). The rise and fall times (τ_r and τ_f, respectively), shown in Fig. 3(a), are characterized by a double exponent, with the fastest time being of the order of τ₁ ≈ 10 μs. The fast time constant is associated with the electron–hole generation process, and the slow one is associated with charge redistribution processes.³⁵ We note that the fastest rise time was limited by the speed of the optical chopper available in our setup, so that graphene based MSM can, in principle, be much faster than here described (supplementary material, S4).

As for detectivity, the small dark currents due to the large resistance in the dark state, R₀ = dV/dI ∼20 TΩ [Fig. 1(c)], give expectation for large detectivity values. A common way to estimate detectivity is to assume that the noise is of purely thermal origin, which allows for an estimate of detectivity, D ≈ eλη/hc(4k_BT/R₀A)^−1/2 (Refs. 17 and 20). If we apply this formula with λ = 250 nm, T = 300 K, η = 1, R₀ = 20 ΤΩ, and Α = 5 × 5 mm², the detectivity is D = 3.5 × 10¹⁵ Jones, which is among the highest theoretical estimates ever reported in literature for UV detectors.¹⁷ Yet, we can compare this estimate with experimental data. For this, Fig. 3(b) shows the noise characterization of the device. The current spectral noise density is flat, S_N = 0.7 pA/√Hz above f ∼10 Hz. At optimal responsivity of R = 168 mA/W, this leads to a NEP = S_N/R = 4.2 pW/√Hz, which, in turn, results in a detectivity of about D = √A/NEP = 1.2 × 10¹¹ Jones for device A with area A = 5 × 5 mm². This is the experimentally determined detectivity, limited by the amplifier noise, whereas theoretically one should achieve four orders of magnitude better detectivity.

Finally, we mention that the unprecedently high responsivity to UV light demonstrated here for the epigraphene system is observed only in graphene grown at high temperatures, T >1850 °C (supplementary material, S5). It is known that the recombination lifetime of photo-generated carriers in 4H-SiC can be increased by annealing at high temperatures (>1700 °C).^36–40 Thus, we suggest that the high temperature process needed to produce epigraphene with high crystalline quality is the cause for a reduced concentration of defects at the graphene-SiC substrate interface that could act as charge traps and/or recombination centers. A typical challenge of MSM detectors has been to reliably control the metal–semiconductor interface. For epigraphene, the graphene layer acting as the metal is produced at high temperature in quasi-thermodynamic conditions, making it highly reproducible. Epigraphene is particularly suitable to implement the fully planar device architecture of the MSM structure with a low processing effort, and makes it technologically attractive for the development of high-performance multi-pixel UV sensor arrays.

See the supplementary material for extra data on optical and Raman analysis of epigraphene, responsivity in the visible range, UV photo response of device B, speed of photo response, and comparison of UV detection in epigraphene grown at different temperatures.

This work was jointly supported by the Swedish Foundation for Strategic Research (Nos. GMT14-0077 and RMA15-0024), Chalmers Excellence Initiative Nano, and 2D TECH VINNOVA competence Center (Ref. No. 2019-00068). This work was performed in part at Myfab Chalmers.