1. Introduction

In 1996 Professor Eric Mazur at Harvard University developed a laser treatment technique by irradiating a silicon wafer in vacuum filled with sulfur hexafluoride chalcogen-containing gas with femtosecond light pulses. Blackened silicon surface covered with array of self-organized nanoscale spikes were formed [1], hence marked the birth of black silicon (b-Si). However, we note that similar modification was discovered earlier as an unwanted side effect of reactive ion etching [2]. Nevertheless, this modified Si material has many remarkable properties, including the near-unity absorption of light in broad spectral ranges from the visible to the Near Infrared (NIR). It was demonstrated that the sulfur atoms, imbedded in the silicon surface helps to create a band of states in silicon bandgap, therefore allowing photon absorption to longer wavelengths. In 2003, Mazur and his team built the first b-Si photodetector for NIR applications [3,4], followed by the demonstration of a b-Si avalanche photodiode [5]. Subsequently, other surface modifications have been explored, including electrochemical etching and reduction [6], stain etching [7], metal-assisted chemical etching [8], and inductively coupled plasma–reactive ion etching (ICP-RIE) [9].

Black silicon is a needle-shaped, high aspect ratio nanostructure made on the surface of silicon. Its main feature is enhanced incident light absorption, which is due to the formation of an anti-reflective interface with the continuously changing refractive index that reduces Fresnel reflection [10]. When the depth of the graded layer and the nanostructure size match the wavelength of light, photons are completely absorbed, and the reflection is reduced to near zero. With ICP-RIE processing the surface modification is independent of local crystalline orientation and a uniform texturing effect can be achieved across the entire surface of the silicon wafer [11]. b-Si has attracted a lot of interest in the solar photovoltaic industry as the low reflectivity would greatly improve the efficiency of silicon solar cells, and significantly help to reach grid parity. However, attempts to integrate b-Si in solar cells resulted in low efficiency far from theoretical maximum or best commercial solar cells (e.g. Sunpower, HIT) due to the increased charge carrier surface recombination caused by the increased total area of the nanostructures and lack of effective passivation technique [12]. It was then discovered that black silicon surface could be passivated using a conformal aluminum oxide layer with high negative charge leading to the demonstration of 22.1% efficiency b-Si solar cell using interdigitated back contact structure [13].

Photodiodes, similarly, to solar cell, benefit from the reduced reflection and a near-ideal b-Si photodiode was demonstrated which exhibits a Quantum Efficiency (QE) >96% in the NIR and visible wavelengths [14]. The high QE was achieved by combining low reflectivity of b-Si, and by using the aforementioned aluminum oxide with high negative charge to form an induced junction [15], which eliminates the Auger recombination and minimizes charge carrier surface recombination. This recombination prevention is especially important in ultraviolet detection since ultraviolet photons are absorbed very close to the device surface and as a result QE>100% was achieved with wavelengths below 300 nm. This finding was verified by Physikalisch-Technische Bundesanstalt (PTB) while extending their measurement wavelength down to 200 nm [16]. However, an important region for direct scintillation light detection in noble gas and liquid detectors is below 200 nm; 175 nm, 150 nm, and 128 nm for liquid Xenon, Krypton, and Argon, respectively, where the QE of single-photon VUV sensitive photodetector is <25%. Therefore, if the high QE of the b-Si photodiodes is maintained also in the VUV with the possibility of an internal charge gain, significant improvements in single-photon detection could be achieved with much improved energy resolution in applications such as DarkSide (LAr), LUX (LXe), XENON (LXe), and nEXO (LXe), where the aforementioned liquid scintillators are used in dark matter search experiments [17–19].

In this paper we extend the investigation of these novel photodiodes into the VUV wavelengths. We present the QE measurements, response time, and leakage current as a function of the bias voltage of the b-Si photodiodes down to 170 nm (7.3 eV) at room temperature and at ∼87°K cryogenic temperature in vacuum. Finally, we compare the results from b-Si photodiodes to standard photodiodes and discuss further possibilities to extend the b-Si technology to photodetectors with internal gain.

2. Experimental

2.1. Device fabrication

The photodiodes were fabricated by ElFys, Inc. using process similar to the one previously described [14], hence only the main features are outlined here. Two different kind of n-type silicon starting materials were used: Float zone (FZ) 675-µm-thick >10 kΩ cm and magnetic Czochralski (MCZ) 500-1000 Ω cm. The minority-carrier lifetime in both starting materials is >2 ms. The active areas of the photodiodes were 5 × 5 mm² and 10 × 10 mm² where b-Si was etched at cryogenic temperature using ICP–RIE. The b-Si was passivated using 20 nm thick atomic layer deposited (ALD) alumina (Al₂O₃), which also forms the induced junction at the same time. The b-Si region was surrounded by a boron-implanted area with aluminum on top to realize a p-type ohmic anode contact to the inversion layer. Similarly, the n-type cathode contact was made by aluminum covered phosphorus implantation on the backside. A scanning electron microscope image of the b-Si nanostructured active area is shown in Fig. 1(a).

 

Fig. 1. (a) Scanning electron microscope image of the b-Si nanostructured active area. (b) A 5 × 5 mm² black silicon photodiode (model # Pin25sH) mounted on a PC board. (c) A sample mounted on a copper cold finger illuminated at the center by VUV photons.

Download Full Size | PDF

The b-Si photodiodes were glued to custom printed circuit boards (PCB) with conductive epoxy for the cathode contacting and wire bonding was made for the anode contacting, see Fig. 1(b). The PCB was then mounted on a copper cold finger and b-Si photodiode was illuminated at the center by VUV photons, see Fig. 1(c). The presented test results are from two different batches of b-Si photodiodes, Pin25sH (5 × 5 mm²) using FZ substrate and Pin25sM/Pin100sM (5 × 5 mm² and 10 × 10 mm²) using MCZ substrate. The main differences between the batches arise from the different substrate doping concentrations. The very low doping level in the -sH series leads into wide depletion region which is seen as low capacitance but at the same time, the dark current generation is higher. The higher doping level in the -sM series yields a significantly lower dark current but increased capacitance.

2.2. Device characterization

All b-Si photodetector characterization was performed at the Brookhaven National Laboratory. For 170-260 nm VUV spectral response measurements, the b-Si photodiode was mounted on a cold finger placed at the exit port of a vacuum monochromator (Acton Research, f=0.5 m) equipped with an in-vacuum glass shutter for background subtraction by blocking VUV light. All measurements were done first at room temperatures in ∼10⁻⁴ Torr vacuum condition followed by cryogenic cooling with LN₂ to the steady state temperature of ∼87°K. For photon illumination, we used a fiber-coupled, dry nitrogen purged cw VUV white light source (Energetiq, LDLS EQ-99XFC) coupled directly to the entrance port of a vacuum monochromator using a 600 µm in-vacuum solarization-resistant UV fiber feedthrough. Wavelength is spectrally selected by the monochromator in ∼10 nm bandwidth resulting in 0.03-1 nW (depending on wavelength) of unpolarized VUV photons delivered to the b-Si photodiode at normal incidence angle. Figure 1(c) shows a representative illumination light spot on the b-Si surface. The minimum wavelength was limited by the quartz light bulb of the VUV white light source and the transmission of the UV fiber feedthrough. The absolute VUV optical power was calibrated against a 10 × 10 mm² NIST traceable vacuum photodiode placed at the same exit port of the monochromator. To ensure the stability and repeatability of the measurement, the light source was dry nitrogen purged at 20 psig for 3 hours prior to data collection, and calibrations were performed before and after every set of measurements. To confirm the reproducibility, all measurements were repeated on different days. We found no sample degradation after many thermal cycles. In addition, using a different setup at ambient condition, we confirmed the near unity QE of the b-Si in the VIS/NIR wavelengths (240 nm to 1 µm), where 2 to 10 µW (depending on the wavelength) were used. The VIS/NIR optical power was calibrated against a Newport 1930C power meter with a certified 918-UV-enhanced head, while the NIR against a Thorlabs PM200 power meter with a S122C NIR sensitive Ge-sensor head, respectively.

The photocurrent versus voltage (IV) curves were measured by an electrometer (Keysight B2987A Electrometer/High Resistance Meter, >200 TΩ input impedance) with a programmable voltage bias to the anode. Temperature dependence of the IV was measured with bias voltage scanned from zero to -50 volt with 10 mV increments and a settling time of 0.1 to 10 seconds per data point depending the stability of the photocurrent and the measurement range. The photocurrents were recorded in logarithmic scale from ∼50 fA (instrument resolution limit) to 2 mA (protection limit) over 10 decades of dynamic range.

The temporal responses of the b-Si photodiodes were measured at room and 87°K cryogenic temperature under various bias voltages from 0 to -20 volt. A bias-T (Mini Circuits ZX85-12G-S+, 0.2 MHz to 12 GHz) was used to provide the voltage bias and to send the signal pulse without amplification terminated with 50 Ω at the oscilloscope (Agilent Infiniium DSO91304A, 13 GHz, 40 GS/s). We employed three sub-nanosecond pulsed lasers emitting 980 nm (Thorlabs VCSEL-980 electrical pulse drive at 2.5 volt, 1 ns pulse width, 100 Hz), 532 nm (PicoQuant LDH-P-FA-530XL, 70 ps, 100 Hz), and 266 nm (Continuum Leopard Nd:YAG 4th harmonic, 60 ps, 10 Hz) wavelengths fiber-coupled to the vacuum fiber feedthrough at the entrance of the vacuum monochromator. Light intensity was constant and maintained well below saturation on the b-Si photodiode at all temperatures.

3. Results and discussions

At a given wavelength, the dimensionless QE of a photodiode is defined as a ratio between the number of electrons or holes extracted to the number of incident photons, QE=$\frac{{\textrm{number of electrons}}}{{\textrm{number of photons}}}$. Accordingly, the responsivity (R) of a photodiode is defined as the ratio of the photocurrent to the incident light power at a given wavelength, R=$\frac{{\textrm{photocurrent}\; ({\textrm{Amp}} )}}{{\textrm{incident light power }({\textrm{Watt}} )}}$. They are related by, QE=$\frac{\textrm{h}{\nu }}{\textrm{q}} $. = 1240$\frac{\textrm{R}}{{{\lambda }\; ({\textrm{nm}} )}}$, where hν, q and λ being the photon energy, the electron charge, and the wavelength of light in nanometer, respectively.

In general, the quantum efficiency varies with the wavelength of the incident light, the applied reverse bias, and temperature. The temperature variations are mostly due to decrease or increase of the band gap with temperature. Typically, in the conventional silicon photodiode the spectral QE varies significantly with wavelength and it could increase slightly with bias due to improved charge collection efficiency. b-Si photodiode is distinctively different with a nearly flat spectral response in the visible wavelengths due to nearly zero reflection. Also, the responsivity is practically independent of bias voltage due to long bulk diffusion length and wide depletion region even without bias. The most notable benefit from biasing would be faster response speed due to higher drift velocity, lower capacitance and smaller region where charge carriers need to move slowly by diffusion.

3.1. Spectral responses

The complete spectral dependence on the quantum efficiency of the -sH series b-Si photodiode at zero bias from 0.8 eV to 7.3 eV is shown in Fig. 2. There is a good overlap at the UV wavelengths among the 2 sets of data from the VUV and VIS/NIR measurements. Much like in the case of conventional silicon photodiodes, the QE below the 1.1 eV intrinsic bandgap of silicon is nearly zero. But the QE of b-Si remains relatively flat at near unity from near-IR to ∼3.5 eV, becomes greater than 100% above 4 eV (below ∼310 nm), and increases near linearly with photon energy. At room temperature, QE of ∼150% at 7.3 eV was achieved (corresponding responsivity of 0.2 A/W at 170 nm). These results were obtained consistently from all -sH series b-Si photodiodes. At 87°K cryogenic temperature in vacuum, the QE dropped by <10% at photon energies <4 eV. However, above 4 eV the QE reduction is more significant but remains nearly unity (∼100%). At the vicinity of 6.2 eV, the QE dropped to ∼85%. The measurement uncertainty of the QE is <3% in NIR/UV wavelength range and increases up to ∼8% in the VUV range. Representative error bars for one sample are shown at the data points of 7.1 eV for both temperatures. The small QE variation is likely due to sample difference than statistical uncertainty.

 

Fig. 2. (a) The measured quantum efficiency versus the incident photon energy for a b-Si photodiode (Pin25sH, 5 × 5 mm²) from 0.8 eV to 7.3 eV (1550 nm to 170 nm) at room temperature (red) and at 87°K (blue). Open circles and squares are the data of 2 samples among two sets of results from the VUV and VIS/NIR measurements, lines are connected for clarity. Representative error bar for one sample is shown at the data points of 7.1 eV for both temperatures but in reverse color for clarity.

Download Full Size | PDF

The gradual drop in the spectral QE with deceasing temperature was independently studied using a highly stable UV-LED source at 270 nm; the ∼15% QE drop from room to 87°K is unambiguously confirmed, see Fig. 3(d). Some degradation in QE is expected because the bandgap energy increases with decreasing temperature, which should affect the impact ionization probability. The phenomenon of QE dropping at low temperature resemble that observed in conventional Si photodiodes [20]. Other possible explanations include change in reflectance of b-Si, increased absorbance of aluminum oxide for ultraviolet light, change in surface recombination velocity or change of recombination in nanostructures at lower temperature. Surface recombination is known to have negative temperature coefficient [21] which makes it unlikely reason for change in QE. Field enhanced recombination in induced junction could be stronger than in p-n diode due to higher electric field. This effect leads to positive temperature dependence [22], which in principle could partly explain the observed temperature behavior. It is also interesting to note that some experimental results suggest that electron-hole pair creation energy has maximum between 200-250 nm [23], which could explain larger than excepted change in QE at low temperature around 6.2 eV and its spectral dependence.

 

Fig. 3. (a) Stability of the b-Si photodiode response to the 175 nm VUV illumination at room temperature, 1.4%. (b) Photo-response with bias voltages after subtraction of the corresponding dark current at room temperature and 87°K. (c) Linearity of the b-Si photodiode at room temperature and (d) the drop of the photocurrent with decreasing temperature under 270 nm UV photon illumination.

Download Full Size | PDF

We note that the bandgap of silicon estimated from the cut-off wavelength in Fig. 2 increased from 1.1 eV at room temperature to ∼1.2 eV at 87°K, in agreement with the known temperature dependence coefficient of the energy bandgap 4.73 × 10⁻⁴ eV/^°K (∼0.1 eV increase of bandgap over ∼208^°K drop in temperature) [24].

A set of VUV spectral measurements typically completed within 15 minutes. We quantified the stability of the photocurrent and the VUV white light source at 175 nm to be ∼1.4% over the same duration of measurement time, shown in Fig. 3(a). We also confirmed that biasing the b-Si photodiode has negligible effect on the QE in either warm or cold conditions, see Fig. 3(b). A typical VUV optical power illuminating on the b-Si is 20 to 500 pW, while the photocurrent is >20 pA, significantly higher than the ∼50 fA noise current of the instrument. We note that the dark current inevitably increased with bias but was subtracted out in each photocurrent measurements shown in Figs. 3(b) and 3(d). The linearity of the b-Si photodiode response is shown in Fig. 3(c) where the photocurrent is plotted against the incident optical power at 270 nm in logarithm scale. The maximum optical power incident on b-Si was around few mW in the UV range. The large dynamic range outperforms the first induced junction detectors is encouraging [25,26].

By definition 100% QE means one electron-hole pair is generated per incident photon. Above-unity internal QE response for UV, VUV and soft X-ray photons in silicon photodiodes is well known and can be explained by secondary ionization and carrier multiplication [27–31]. However, in the VUV photon energy range, QE is generally low because of large reflection and high absorption close to the silicon surface. Here we show for the first time that a nearly ideal QE is achieved in the VUV photon energy range using a b-Si photodiode when three effects are combined: (1) the b-Si nanostructure efficiently suppresses light reflection, (2) the ALD of alumina minimizes surface recombination, and (3) the induced junction minimizes the Auger recombination. These design elements also ensure the nearly 100% QE achieved at lower photon energies without invoking the carrier multiplication process. If any of the three effects is not properly implemented, it would have lower QE especially in the UV [32]. The carrier multiplication requires the photon energy to be at least two times the bandgap and its probability increases linearly with photon energy, leading to an average pair creation energy of about 3.66 eV in silicon at high photon energies [33]. This threshold qualitatively agrees with our observation. The observed increase in bandgap when the detector is cooled down would also at least partly explain why lower QE is obtained at 87° K. The change in average pair creation energy should increase even more than the bandgap because there will be less excess energy transferred to the charge carriers after photon absorption and additionally the initiation of impact ionization also requires more energy.

3.2. Dark current characteristics

The spectral response shown in Fig. 2 was measured at zero-volt bias on the -sH series b-Si photodiode. At room temperature it had a dark current of ∼0.25 nA/cm² which increased to ∼2.3 nA/cm² at 10 mV. The complete IV characteristic on both series of b-Si photodiodes at warm and cold temperatures is shown in Fig. 4(a). Other than a general increase of dark current with bias voltage, which resembles a typical silicon photodiode, the -sM series has more than 2-order of magnitude lower dark current than the -sH series. Nevertheless, cryogenic cooling to 87°K significantly lowered the dark current by 4-order of magnitude on both series. The higher bulk doping in the -sM series showed a distinctively lower dark current, dropping to the instrument limit of ∼50 fA below 225^°K. The temperature dependence of the dark current at zero- and 0.1-volt bias was measured as the b-Si photodiode was slowly warmed up from 87^°K to room temperature in vacuum at a rate of ∼0.8°K/minute, see Fig. 4(b). The dark current increased at the rate of ∼1 decade/20^°K on all series of the b-Si photodiodes. The slope of the Arrhenius plot, log(I) vs. 1/kT, shown in the inset of Fig. 4(b), suggested an activation energy of ∼1 eV across all series of b-Si photodiodes.

 

Fig. 4. (a) Increase of dark current with bias voltage at room and cryogenic temperatures and (b) the temperature dependence of dark current at 0 and -0.1-volt bias for the -sH and -sM series b-Si photodiodes, respectively.

Download Full Size | PDF

3.3. Temporal response

Temporal responses of the b-Si photodiode at 295°K and 87°K to sub-nanosecond light pulses at 980 nm, 532 nm, and 266 nm wavelengths under different bias voltages from 0 to -20 volt are shown in Fig. 5. A pin silicon photodiode was used to verify the short sub-nanosecond light pulses on each wavelength measurements. The responses of the b-Si photodiodes to the light pulse were recorded directly on an oscilloscope without any additional electronic amplifier. There is no significant difference in the response speed (T_rise & T_fall) at room temperature on all wavelengths for each series of the b-Si photodiodes. A slower time response (both T_rise & T_fall) was observed on the -sH (C=0.32 nF/cm² at 0V bias) than the -sM (C=1.4 nF/cm² at 0V bias) series b-Si photodiodes for all wavelengths. In cold both the rise and fall times are faster at higher bias for all wavelengths on both series compared to room temperature.

 

Fig. 5. Time response of the b-Si photodiode at 295°K (left) and 87°K (right) temperatures to sub-nanosecond light pulse of irradiation at (a) 980 nm, (b) 532 nm, and (c) 266 nm wavelengths under different bias voltages from 0 to -20 volt. The instrument response from a pin silicon photodiode (Thorlabs DET10A, T_rise<1ns) at each wavelength measurements are depicted in black dotted signal pulse traces. The colored solid (dash) signal traces are for the -25sH (-25sM) series b-Si photodiodes, respectively, at the corresponding bias voltages. The amplitudes of all signal traces are normalized to the highest bias voltage trace in cold of the corresponding wavelength. All -25sM series signal traces are time shifted for clarity.

Download Full Size | PDF

The time response of b-Si photodiodes is limited by electron drift across the depletion layer, RC time constant, and dielectric relaxation time. At 532 nm and 266 nm wavelengths, the absorption depths are less than 1.3 µm [34]; this means that all light is absorbed close to the surface and the hole drift time is negligible compared to the drift time of electrons that need to drift across almost the entire depletion layer. The calculated pulse widths for a step response (electron mobility at 1400 cm²/Vs [35]) are ∼50 ns and ∼7 ns for -sH and -sM series, respectively, at -20 V bias [36], which are in good agreement with our measurements.

The mobility of electrons increases approximately by a factor of 10 in cold [35] which explains the faster response observed in cold compared to room temperature. Because of the wider depletion region in the -sH series, the increased drift velocity contributes more and larger change is seen in the response time. While the time response of b-Si photodiode is faster in cold and at the higher bias, the collected charge (integrated area of the charge pulse signal) remains nearly the same, supporting our findings in Section IIIA that QE does not drop significantly in cold.

4. Conclusion and outlook

We demonstrated that the QE of b-Si photodiode in UV/VUV exceeded unity and increased linearly with photon energy higher than ∼4 eV (or wavelength below 200 nm). Such a record-high performance was achieved using a nanostructured silicon photodiode with an induced junction. We hypothesized that the observed QE above 100% is due to multiple carrier generation by impact ionization taking place in the b-Si nanostructures. The high charge density present in the Al₂O₃ could also provide charge carriers collection enhancement due to excellent surface passivation [37]. The new technology and the benchmark results presented here demonstrate that b-Si photodiodes are particularly sensitive to UV and VUV wavelengths. This might have significant impact on various scientific programs where low light level photon detections are needed across a broad spectral region. However, detailed analysis on the electric field and electrostatic potential distributions inside the b-Si needle-like nanostructure are needed and was attempted to further understand the above unity QE of b-Si photodiode [16,32]. While additional experiments and model simulations are needed to clarify the temperature dependency of QE on the spectral response of b-Si, particularly in the VUV region ∼6.3 eV where a distinctively larger QE drop was observed.

It is interesting to compare the QE of various silicon photodiode technologies to b-Si over a wide spectral range, see Fig. 6(a). A typical NIST calibrated silicon photodiode (NIR/VIS/UV and AXUV-100G vacuum UV) exhibits a characteristic lower QE in the 2.5 to 6 eV region; while it already exceeds 100% above ∼7.5 eV and appears to increase linearly followed by a distinctive drop around 10 eV. On the contrary, the QE of b-Si photodiode is remarkably high at nearly 100% for all photon energies and increases linearly but at a slightly lower rate beyond 4 eV. Nevertheless, both have the same cut-off energy at the conventional silicon band gap of ∼1.1 eV. Basic silicon and b-Si photodiodes have unity charge gain at the p-n junction; while state-of-the art silicon photomultiplier (SiPM) have an internal charge gain >10⁶. In contrast, the QE of VUV sensitive SiPM is at most ∼40% in the VUV, see Fig. 6(a), where the QE of a VUV sensitive SiPM is plotted (QE=$\frac{{\textrm{PDE}}}{{\textrm{fill factor}}}$, PDE∼23.5% and fill factor=0.6 for a 50 µm pixel VUV4 Hamamatsu SiPM). The realization of a b-SiPM with internal charge gain >10⁶ and near unity QE would greatly benefit many scientific fields.

 

Fig. 6. (a) Quantum efficiency of b-Si photodiode compared to state-of-the-art photodiodes and SiPM. (b) The dark current per unit area of b-Si photodiode (ELFYS, -sM series, 10 × 10 mm²) compared to a VUV sensitive SiPM (Hamamatsu S13371-6342, 50 µm pixel, 4x(6 × 6) mm²). The breakdown voltages of the SiPM are 52 volt and 42.5 volt at 295°K and 87°K, respectively.

Download Full Size | PDF

In its simplest form, SiPM is realized by inducing a deeper depletion across a p-n junction with high E-field in the depletion layers until breakdown occurs. When the strength of the E-field is sufficiently high, the mean kinetic energy of carriers can exceed the silicon band gap energy and ionize the lattice atoms upon impact and release an additional electron-hole pair. This impact ionization effect constitutes a carrier multiplication phenomenon by which the number of drifting carriers increases rapidly, resulting in an avalanche breakdown. In contrast, the VUV response of b-Si photodiode utilizes multiple carrier generation by impact ionization but in the absence of an avalanche gain region. It is challenging to design and fabricate a b-SiPM without significantly increasing the leakage current until breakdown. Previous attempt to fabricate an avalanche junction on b-Si APD resulted in a QE of 58% at 1064 nm with an internal gain of ∼450 suggesting that improvement in charge collection is possible [6,38–39]. We noted that the dark current of -sM series b-Si photodiode is sufficiently low at room temperature and is remarkably low in cold, in concert with modern VUV sensitive SiPM, see Fig. 6(b). The extremely low dark current characteristics of b-Si may pave the way toward the realization of a b-SiPM photodetector