We report the performance of a bipolar epitaxial graphene (EG)/p-SiC/n+-SiC UV phototransistor fabricated with a Schottky (EG)/SiC junction grown using a SiF4 precursor. The phototransistor showed responsivity as high as 25 A/W at 250 nm in the Schottky emitter (SE) mode. The Schottky collector (SC) mode showed a responsivity of 17 A/W at 270 nm with a visible rejection (270 nm:400 nm)>103. The fastest response was seen in the SC-mode, with 10 ms turn-on and 47 ms turn-off, with a noise equivalent power of 2.3 fW at 20 Hz and a specific detectivity of 4.4 × 1013 Jones. The high responsivity is due to internal gain from bipolar action. We observe additional avalanche gain from the device periphery in the SC-mode by scanning photocurrent microscopy but not in the SE-mode. This high-performance visible-blind photodetector is attractive for advanced applications such as flame detection.
Ultraviolet (UV) detectors find many applications such as in flame sensors, space communication, biological detection, and missile plume sensing.1 Commercially available Si UV detectors2 require optical filters to block visible light and suffer from low UV responsivity. Due to their wide band-gap, 4H-SiC (3.26 eV) and III–V semiconductor materials such as AlGaN and GaN are excellent candidates for visible (λcut-off = 400 nm) and solar blind (λcut-off = 290 nm) UV detectors. Particularly, the use of SiC based photodetectors can be advantageous owing to their radiation hardness3 and potential for integration with SiC electronics4 for readout.
In all these devices, above bandgap UV-light is absorbed, leading to electron-hole (e-h) pair creation. These e-h pairs are then separated by a built-in and/or applied electric field, leading to photocurrent. Simple devices such as p-i-n or Schottky photodiodes offer high speed but low responsivity due to unity photocurrent gain, even at 100% quantum efficiency.5–7 Photocurrent gain can be achieved using bipolar action,8 avalanche multiplication,9 or persistent photoconductivity.10 Persistent photoconductivity (such as in FETs11) gives high responsivity but large dark currents, leading to large noise equivalent power (NEP),12 as well as potentially slow response.13 Avalanche photodiodes (APDs) offer high gain (>104) and fast response with low NEP but require precise engineering of the breakdown and often require large applied voltages.9 Bipolar phototransistors offer a compromise between these two extreme cases, with reasonable gains >102 and faster response than a photoconductor. In an npn phototransistor, the light is absorbed in the base, and the e-h pairs are separated at the base-collector junction, leading to the accumulation of holes in the base. This lowers the base potential, leading to the injection of minority electrons and bipolar gain.
To maximize responsivity, R, defined as the number of amperes of photocurrent per Watt of incident optical power, the absorption of light must be maximized in the base. In addition to having a base sufficiently wide to absorb all incoming light, reflections at the electrodes must be minimized. A key disadvantage of many device structures is the low UV responsivity incurred due to reflection/absorption losses coming from the top metal contacts.
There has been interest in the epitaxial graphene (EG)/SiC material system due to the tunable native Schottky junction14 and potential for SiC integrated circuit manufacturing on 6-in. wafers. In particular, EG is a UV transparent material, which enables large area windows for high responsivity (R) UV-detection. EG/SiC based UV detectors with improved responsivities were demonstrated by reducing the reflection/absorption losses at the detector surface.15 The EG/SiC Schottky emitter phototransistor (SEPT) reported by our group16 showed high R = 7 A/W at 365 nm due to minority electron injection from EG into p-SiC leading to high bipolar gain. However, this device suffered from large dark current due to the lack of base/collector junction mesa isolation and poor visible rejection <102 at 400 nm.17 Moreover, this device did not show appreciable gain in the Schottky collector (SC) mode.17 These characteristics are impractical for post-Si detection applications.
In this work, we use a bipolar junction transistor (BJT) device similar to that reported in Ref. 16 having an EG/p-SiC/n+-SiC structure with a thinner SiC base (13 μm) to improve the base transit factor and hence collection of the minority carrier injected by bipolar action.16 EG was grown using a new SiC homoepitaxy-compatible18 SiF4 gas precursor. The device showed significantly improved performance in both Schottky emitter (SE) and SC-modes. We report comprehensive characterization of the EG/SiC phototransistor including spectral responsivity, speed, noise equivalent power (NEP), and detectivity and show that it compares well with other visible-blind devices reported in the literature.
For the phototransistor device fabrication, the 13 μm thick p-SiC base epilayer is grown on an 8° offcut n+-4H-SiC (0001) substrate by a CVD reactor using dichlorosilane (DCS) and propane in hydrogen ambient at 300 Torr and 1600 °C at a C/Si ratio of 1.9,19 giving a growth rate of ∼26 μm/h for 30 min, producing a 13 μm thick film as determined by Fourier transform infrared reflectance (FTIR). The resultant doping of the epilayer, due to site-competition epitaxy,20 was found to be p-type 3.7 × 1014 cm−3 by the Hg-probe capacitance-voltage (C-V) measurement. This thickness was based on our previous work,16 where a diffusion length of ∼10 μm was measured in the 30 μm base.17 Thus, to improve the base transit factor and hence the current gain, a thinner layer was used, although this always comes at the expense of lower light absorption for long wavelengths (∼30 μm for λ = 365 nm, as in Refs. 16 and 17). To achieve reasonable absorption in the range of 250–400 nm,21 while maintaining adequate current gain, the 10 μm base thickness range was chosen, with the resultant 13 μm base obtained for our standard 30 min growth.
The EG top electrode layer is then grown on the SiC base at 1600 °C and 300 Torr, in the same reactor, using the SiF4 precursor in Argon for 10 min using a chemically accelerated Si-removal process developed at our lab.18 From FTIR and X-ray photoelectron spectroscopy,22 the thickness of the EG is estimated to be ≈15 monolayers for these growth conditions. Circular graphene regions of diameter 250 μm are defined for the device, using photolithography followed by O2 plasma reactive-ion etching (RIE).
Figure 1 shows the schematic and band diagrams of the vertical bipolar phototransistor device using EG/p-SiC/n+-SiC layers. The device is operated in two different modes, namely, Schottky emitter (SE) and Schottky collector (SC) based on the polarity of the bias applied at the top EG layer. The device is said to operate in the SE mode, when the EG/p-SiC junction is forward biased and the p-SiC/n+-SiC junction is reverse biased [Fig. 1(b)]. On the other hand, the device is said to operate in the SC mode when the EG/p-SiC junction is reverse biased and the p-SiC/n+-SiC junction is forward biased [Fig. 1(c)]. When the device is illuminated with light from top, e-h pairs will be created due to photon absorption in SiC. These e-h pairs will be separated by the electric fields in the base-collector depletion regions in SE and SC modes. We note that the Schottky barrier height for the EG to n-SiC is 0.8 eV, as measured by C-V on EG/n-SiC Schottky test structures,23 and is higher than the 0.5 eV barrier height reported for thermally grown EG/SiC junctions.14,24
The current-voltage (Ic-VCE) characteristics of the device are measured in the dark and under light for both SC [Fig. 2(a)] and SE [Fig. 2(b)] modes, by illuminating the device with light of various wavelengths using a monochromatic light source (10 nm bandpass), from which the action spectra are reconstructed. Dark currents of 230 pA and 670 nA are observed in SC and SE modes at 20 V. The significantly larger dark current in the SE-mode is due to the lack of mesa isolation at the 10 μm deep backside SiC p-n junction, which is 1 cm2, compared to the ∼4.9 × 10−4 cm2 area of the graphene/SiC Schottky top junction (Fig. 1), leading to a corresponding increase in the leakage area. However, the dark current in the SE-mode devices, 670 nA, was still 3 orders of magnitude lower than 100 μA observed in our previous devices.16 We attribute this decrease to the significant optimization of our SiC epitaxy which has led to defect reduction in our epilayers.23,25,26 Mesa-isolation of the base-collector SiC p-n junction should significantly reduce the dark current in the SE-mode to values comparable to SiC p-n diodes (Table I), which are among the lowest of any wide bandgap UV detectors.
Device type . | Voltage (V) . | Dark current . | Responsivity at 270 nm (A/W) . | Visible rejection (270:400 nm) . | Response time . | NEP (W) . | Specific detectivity, D* (Jones) . |
---|---|---|---|---|---|---|---|
EG/SiC BJT (SC mode) | 20 | 230 pA | 17 | 5.6 × 103 | 10 ms (On) 47 ms (Off) | 2.3 × 10−15 | 4.4 × 1013 (BW = 20 Hz) |
10 | 100 pA | 2.2 | 2.2 × 102 | NA | 1.1 × 10−14 | 8.6 × 1012 (BW = 20 Hz) | |
5 | 55 pA | 1.1 | 4 × 102 | NA | 1.7 × 10−14 | 5.8 × 1012 (BW = 20 Hz) | |
EG/SiC BJT (SE mode) | 20 | 670 nA | 25 | 12.3 | 46 ms (On) 730 ms (Off) | 3.3 × 10−12 | 9.5 × 109 (BW = 1 Hz) |
SiC pin diode5 | 20 | <0.5 pA | 0.13 | >104 | NA | NA | NA |
SiC Schottky36 | 5 | 8 pA | ∼ 0.09 | >103 | NA | NA | NA |
SiC APD37 | 144 | 5 pA | 93 (at 280 nm) | NA | NA | 4.4 × 10−15 | 6.4 × 1013 (BW = 1 kHz) |
GaN Schottky33 | 1.5 | 34 nA | ∼ 0.07 | >102 | 150 ns | 3.7 × 10−9 | NA |
AlGaN Schottky7 | 1.35 | 7 nA | ∼ 0.07 | NA | 1.6 μs | 6.6 × 10−9 | NA |
AlGaN photodiode38 | 0.5 | 0.1 pA | ∼ 0.01 | 2.6 × 103 | <0.4 μs | NA | NA |
Device type . | Voltage (V) . | Dark current . | Responsivity at 270 nm (A/W) . | Visible rejection (270:400 nm) . | Response time . | NEP (W) . | Specific detectivity, D* (Jones) . |
---|---|---|---|---|---|---|---|
EG/SiC BJT (SC mode) | 20 | 230 pA | 17 | 5.6 × 103 | 10 ms (On) 47 ms (Off) | 2.3 × 10−15 | 4.4 × 1013 (BW = 20 Hz) |
10 | 100 pA | 2.2 | 2.2 × 102 | NA | 1.1 × 10−14 | 8.6 × 1012 (BW = 20 Hz) | |
5 | 55 pA | 1.1 | 4 × 102 | NA | 1.7 × 10−14 | 5.8 × 1012 (BW = 20 Hz) | |
EG/SiC BJT (SE mode) | 20 | 670 nA | 25 | 12.3 | 46 ms (On) 730 ms (Off) | 3.3 × 10−12 | 9.5 × 109 (BW = 1 Hz) |
SiC pin diode5 | 20 | <0.5 pA | 0.13 | >104 | NA | NA | NA |
SiC Schottky36 | 5 | 8 pA | ∼ 0.09 | >103 | NA | NA | NA |
SiC APD37 | 144 | 5 pA | 93 (at 280 nm) | NA | NA | 4.4 × 10−15 | 6.4 × 1013 (BW = 1 kHz) |
GaN Schottky33 | 1.5 | 34 nA | ∼ 0.07 | >102 | 150 ns | 3.7 × 10−9 | NA |
AlGaN Schottky7 | 1.35 | 7 nA | ∼ 0.07 | NA | 1.6 μs | 6.6 × 10−9 | NA |
AlGaN photodiode38 | 0.5 | 0.1 pA | ∼ 0.01 | 2.6 × 103 | <0.4 μs | NA | NA |
Clear bipolar phototransistor action is seen in both SC and SE-modes.8 In the SE-mode, the current increases starting at VCE = 2 V, in agreement with the 2.4 eV p EG/SiC Schottky barrier estimated above (Fig. 1). For the SC-mode, the bipolar behavior is seen until VCE ∼ 10 V beyond which the photocurrent increases sharply due to avalanche effects from the electric field concentration at the reverse-biased EG/SiC Schottky barrier periphery.27 In the SE-mode, there is no periphery due to the lack of mesa-isolation, so avalanche breakdown at the device periphery is not seen. This assertion is clearly supported by scanning photocurrent microscopy (SPCM) maps at 444 nm (Ref. 17) (Fig. 2), where “hot-spots” are seen in the SC-mode at the periphery that increase in prominence at higher VCE, in concurrence with the sharp current increase in the photocurrent, whereas similar features are not observed in the SE-mode. The uniform ring/halo at the device edge in the SE-mode is due to non-specular scattering at that edge, leading to greater photocurrent in the non-mesa isolated SiC p-n junction collector. It is evident, however, that the hot-spots seen in the SC-mode are not visible in the SE-mode, showing that the sharp increase in current in the SC-mode at VCE > 10 V is due to avalanche from the device periphery. This avalanche effect may be minimized by using field-plate techniques,27 although careful engineering may allow it to be exploited as in avalanche photodiodes.28
In the SE-mode Ic-VCE curves under illumination (Fig. 2), a small hump is seen near ∼0.7 V, which we attribute to the presence of a Schottky barrier height from the edge, in addition to the larger one from the bulk (Fig. 1). As the emitter-base junction turns on, the influence of the parasitic smaller barrier is eventually overwhelmed by the bulk owing to the much larger area associated with the higher barrier. This could be due to independent contributions from bulk and periphery of the graphene contact, in analogy to SC-mode results where edge and bulk clearly give independent contributions.
Larger absolute photocurrents are seen in the SE-mode, again due to the lack of mesa-isolation in the SE-mode, since the circular spot size of the light from the spectrometer is ∼1 cm2. This means that in the SE-mode, the e-h pair collection area is ≫4.9 × 10−4 cm2 of the graphene/SiC junction, whereas it will be comparable to this area in the SC-mode, leading to the apparent large difference in photocurrents, despite the similar absolute responsivities, R(λ), reported in Fig. 3. R(λ) is defined as the ratio of the observed photocurrent (difference of current under illumination and in the dark) to the optical power incident on the device. The dependence of R on wavelength (λ) was measured under wide area illumination by comparison to a calibrated Si photodiode. To account for the difference in the collection area discussed above, the absolute responsivity, R, was calibrated to measurements performed at 365 nm with illumination through a microscope focused to an area < the device area. The R(λ) values are higher than expected from 100% quantum efficiency (dashed line in Fig. 3) for above bandgap (∼390 nm for SiC) light illumination, indicating current gain in both SE and SC-modes. A peak R (250 nm) = 25 A/W is observed in the SE-mode which corresponds to current gain g > 120, as given by29
where R is the measured responsivity (in A/W), λ is the incident light wavelength, h is Planck's constant, c is the light velocity, and q is the electron charge, where we assume a quantum efficiency, η = 1, to estimate a lower bound on g in the final expression in (1). In the SC-mode, a peak R (270 nm) = 17 A/W is measured, corresponding to g > 78 although as discussed above, this is due to a combination of bipolar gain and avalanche gain from the device periphery at VCE = 20 V. At VCE < 10 V, avalanche gain from the periphery is effectively suppressed, and R is also reduced, leading to bipolar current gain, g ∼ 10. In the SC-mode, the short absorption lengths in SiC21 (≈1 μm at 270 nm) for short wavelength photons result in lower R due to the recombination of the photogenerated carriers5 at the EG/SiC Schottky collector junction. In a long-base bipolar device, where we assume a minority carrier injection efficiency of ∼1 and that g is limited by base transit, we estimate the recombination time, τrec, from30
where WQNR is the quasi-neutral region width at a given voltage from the difference in the base-width and the depletion region at the collector side, and Dn = 23 cm2/V s is the diffusivity of electrons in SiC.31 This leads to τrec ∼ 20 ns in both SE and SC-modes. In the SC-mode, the recombination velocity, S, at the EG/SiC interface is estimated from WQNR/τrec 105 cm/s at VCE = 10 V, which is in excellent agreement with that estimated for sub-bandgap illumination previously.17
The UV-visible rejection ratio, R (270 nm)/R (400 nm), is better in the SC-mode ∼5.6 × 103 compared to ∼12.3 in the SE-mode. We attribute the poor visible rejection in the SE-mode to absorption of sub-bandgap light by donor acceptor pairs (DAP)17 present in the highly doped n+-SiC substrate but not in the low-doped p-SiC, since the collection region in this mode spans the n+-substrate unlike the SC-mode. We exclude the contribution of stacking faults to the sub-bandgap response in the SE-mode shown in our previous devices,17 as the SPCM maps show no evidence of this (Fig. 2). Our R(λ) and visible rejection values are compared with those of other wide bandgap photodetectors in Table I, where our device compares well. We also note that our high responsivity is achieved at a relatively low bias voltage of 20 V compared to >100 V for the avalanche photodiodes.
The response times and dark noise spectrum were measured using a virtual-ground transimpedance current pre-amplifier to convert the dark and photo-induced currents to a voltage. This output voltage signal was visualized on an oscilloscope in the time domain for the response times and was converted into the frequency domain using the Fast Fourier Transform (FFT) function for the noise power measurements in the dark. The response times were measured at 320 nm (largest absolute photocurrent) by mechanically chopping the light and fitting the resultant output with exponential decay/growth, where the time constant is taken as the response time. In the SC-mode, ON/OFF response times of 10 ms/47 ms are extracted, whereas in the SE-mode, ON/OFF response times of 46 ms/730 ms are extracted. These values are significantly slower than τrec = 20 ns estimated above, likely due to RC delays, as also seen at low collector currents in Si bipolar phototransistors.32 This assertion is supported by the fact that the response time is slower in the SE-mode, where the collector is a much larger area, leading to a large collector capacitance compared to the SC-mode. Mesa-isolation of the base-collector SiC p-n junction in the SE-mode is expected to improve the response time. Faster response times should also be observable at higher light levels.
The measured noise power density Sn(f) (in A2/Hz) in the SE mode at VCE = 20 V is shown in the inset of Fig. 3 along with the noise floor at this preamplifier gain setting. The 1/f flicker noise is clearly observed and reaches the measurement limit at frequencies >1 kHz. We do not observe a crossover to a shot-noise limited regime in our measurements. In the SC-mode, on the other hand, the dark current was only 230 pA. The noise power density was measured to be <10−28 A2/Hz, the noise threshold at the corresponding pre-amplifier setting, and a 1/f regime was unmeasurable with our setup. Thus, we estimate the dark noise spectral density for the shot noise using Sn(f) = 2eIc(dark)33 for the SC-mode devices. At VCE = 20 V, a white noise density of Sn(f) ≈ 6 × 10−29 A2/Hz is used for estimation of the noise equivalent power (NEP) below. From the measured current noise spectrum, Sn(f) (Fig. 3 inset), NEP, the minimum optical power that can be detected by at the given noise level, is extracted. For a bandwidth of B, the total noise current is given by
where we assume that the value at 1 Hz is assumed to be flat until 0 Hz in the first integral on the RHS as is commonly done.7,34,35 The noise equivalent power (NEP) is then estimated using the following equation:
where R is the responsivity (in A/W) of the device. From the NEP, the specific detectivity (D*) is
where A is the device area in cm2, and B is the bandwidth in Hz. Given that the response time is 47 ms in the SC-mode, we assume a bandwidth of 20 Hz, giving NEP = 2.3 fW and D* = 4.4 × 1013 Jones. Similarly, since the response time in the SE-mode is 730 ms, we assume a 1 Hz bandwidth, giving NEP = 3.3 × 10−12 W and D* = 9.5 × 109 Jones. These values compare favorably with other visible-blind detectors (Table I). Mesa-isolation of the base-collector SiC p-n junction would reduce the area to ∼4.9 × 10−4 A/cm2, leading to a reduction of dark current to <270 pA from 670 nA assuming the same leakage current density, pushing the NEP to <2 fW. The response time should also improve as discussed above, with a corresponding increase in B and consequently D*.
In summary, the device performance of the visible-blind EG/SiC phototransistor is compared in SE and SC modes. UV-visible rejection >103 is achieved, with responsivity ∼17–25 A/W at low voltages VCE = 20 V. This responsivity is shown to arise from bipolar gain and avalanche from the device periphery in the SC-mode. The response times of these devices are relatively slow, 10 ms–730 ms, compared to low responsivity photodiodes. Finally, NEP = 2.3fW and D* = 4.4 × 1013 Jones indicate excellent performance of these devices in both SE and SC modes and are comparable to or better than the state-of-the-art. The overall performance of the phototransistor in this letter can be improved further by mesa-isolation of the SiC p-n junction base layer, making them an attractive candidate for visible-blind UV detection applications.
This research work was partially supported by the National Science Foundation (NSF), ECCS Award No. 1711322 under the supervision of program director Dr. Dimitri Palvidis. B.G.B. was additionally supported by an NSF IGERT fellowship under Award No. 1250052. The authors would also like to acknowledge the support from the University of South Carolina through an ASPIRE Grant.