*Review* **Ammonia Gas Sensors: Comparison of Solid-State and Optical Methods**

#### **Zbigniew Bielecki 1,\*, Tadeusz Stacewicz 2, Janusz Smulko <sup>3</sup> and Jacek Wojtas <sup>1</sup>**


Received: 19 June 2020; Accepted: 24 July 2020; Published: 25 July 2020

**Abstract:** High precision and fast measurement of gas concentrations is important for both understanding and monitoring various phenomena, from industrial and environmental to medical and scientific applications. This article deals with the recent progress in ammonia detection using in-situ solid-state and optical methods. Due to the continuous progress in material engineering and optoelectronic technologies, these methods are among the most perceptive because of their advantages in a specific application. We present the basics of each technique, their performance limits, and the possibility of further development. The practical implementations of representative examples are described in detail. Finally, we present a performance comparison of selected practical application, accumulating data reported over the preceding decade, and conclude from this comparison.

**Keywords:** ammonia detection; NH3; MOX sensors; polymer sensors; laser absorption spectroscopy; CRDS; CEAS; MUPASS; PAS

#### **1. Introduction**

Ammonia is a highly toxic chemical substance, common in biological processes, and applied in technical installation processes (cooling systems, chemical industry, and motor vehicles). The American Conference of Industrial Hygienists has set a limit to ammonia concentration in air of 25 ppm for long-term exposure (8 h) and 35 ppm for short-term ones (15 min) [1]. In medicine, the concentration of ammonia in the breath between 2500 and 5000 ppb is directly related to organ dysfunction and diabetes [2]. Therefore, the design of novel techniques and sensors which allow accurate and fast in-situ detection of trace ammonia concentration is highly desirable. Such sensors should satisfy the specific requirements: high sensitivity, enhanced selectivity, short response time, reversibility, high reliability, low energy consumption, low cost, safety, broad range of measurement at various operation temperatures, etc.

The issue of ammonia detection by gas sensors has attracted many researchers. A recent review paper about ammonia sensing focused on chemical mechanisms of gas sensing by solid-state or electrochemical sensors, with limited details about optical methods [3]. In our studies, we focus on optical methods, developed in our research teams. Moreover, another way of solid-state sensors modulation by UV irradiation was proposed and discussed. This method was advanced in the research teams preparing this review. These problems have not been considered in such a way in other papers about gas sensing.

Very advanced approaches like gas chromatography–mass spectrometry (GC-MS) and selective ion flow tube–mass spectrometry (SIFT-MS) are accurate for NH3 measurement, but their use is complicated and require qualified staff, laborious samples preparation procedures, and time consuming measurements, as well as not-compact and expensive instruments which are difficult to maintain. Many applications require faster and easier tools. Therefore, solid and optical detection methods are rapidly developing (Figure 1). In this paper, much attention was paid on the optical sensors.

**Figure 1.** Ammonia detection technology.

We would like to underline that we preselected the considered ammonia gas sensors and there are other promising technologies which can be effectively used for ammonia sensing because of low production costs or measurement methods. Good examples are biosensors utilizing bacteria cultures or nanotechnology [4–8].

#### **2. Solid-State Ammonia Sensors**

Solid-state ammonia sensors can be divided into two groups considered in our paper: metal oxide-based sensors and conducting polymer sensors [3].

Metal oxide-based sensors (MOX) belong to the most investigated groups. Their main features offer simplicity, comprehensive detection action, miniature dimensions, flexibility in fabrication, long life expectancy, low cost, and serviceableness for alarm warning applications. MOX sensors change DC (static) resistance when exposed to ambient gas or humidity. Any change of MOX sensor resistance can trigger an alarm when toxic gas appears in an ambient atmosphere.

There is a variety of sensitive materials and methods of their preparation for NH3 detection. Metal oxides, like SnO2, ZnO, WO3, TiO2, and MoO3, are most widely utilized for this purpose. They are classified into *n*-type and *p*-type [3]. Usually, the *n*-type semiconducting metal oxides are used for gas sensors due to their higher sensitivity [9]. The sensor comprises of grains of various diameter with a large active surface area. Smaller grains enable better sensing properties due to the larger ratio of active surface to the sensor volume. Atoms of oxygen are bound to the grains. When these atoms are displaced by other species present in ambient atmosphere, a potential barrier between the grains changes. As a result, the DC (static) resistance between the sensor's terminals changes.

Although the metal oxide sensors have attractive properties, their main disadvantage is a low selectivity in detecting of one particular component in a gas mixture [10]. Moreover, MOX sensors operate at elevated temperatures (up to a few hundreds of Celsius degrees), which requires additional energy for heating. Careful selection of ingredients, accelerating adsorption–desorption processes due to catalytic properties, helps to lower operating temperature and to enhance gas selectivity, pointing at selected gases. The sensor exhibits limited specificity for NH3 and can be affected by acetone, ethanol, hydrogen, methane, nitric oxide, and nitrogen dioxide [11]. Artificial neural networks, conductance scanning at periodically varied temperature, as well as principle component or support vector machine analysis help to develop selective sensor systems, but only to some extend [12,13]. The algorithms support to determine even particular components of the gas mixtures. The average detection limit of these sensors ranges from 25 ppb to 100 ppm (Table 1). The selectivity can be further improved by noble metals doping [9,10] or selecting appropriate operating temperature [14]. Unfortunately, even these actions give limited results and require additional advances.

Another approach improving gas sensitivity and selectivity by MOX sensors utilize low-frequency noise (flicker noise) which can be more sensitive to the ambient atmosphere than DC resistance only [15,16]. This idea was proposed more than two decades ago and is still developed. DC resistance gives a single value due to a change of potential barrier at the presence of ambient gases. At the same time, the potential barrier fluctuates slowly, and these fluctuations are observed as flicker noise of the resistance fluctuations. Therefore, low-frequency noise measured as a power spectral density can be more informative than the value of DC resistance only. We confirmed experimentally that a single MOX sensor could detect two toxic gases, NH3 and H2S, by applying low-cost measurement set-up and flicker noise measurements [17]. MOX sensors are made of porous materials, which generate quite intense flicker noise components. Low-frequency noise is observed up to a few kHz at least and, therefore, can be measured using common electronic circuits (e.g., low-noise operational amplifiers and A/D converters sampling the signals up to tens of kHz only).

Some materials used for MOX sensors are photocatalytic (e.g., WO3, TiO2, Au nanoparticles functionalized with organic ligands). These materials can be modulated by UV-light irradiation (e.g., applying low-cost UV LEDs of different emitted wavelengths) [18]. UV light generates ions O2 −, which are weakly bound to the surface of the grains and therefore enhance gas sensitivity. We confirmed experimentally that this modulation improves sensitivity at low gas concentrations and can be utilized for the detection of selected organic vapors (e.g., formaldehyde, NO2 [18]).

The MOX sensors exhibit a temporal drift and ageing of their sensitivity. This detrimental effect can be reduced by algorithms of signal processing or by measuring the derivative parameters (e.g., change of DC resistance only) at relatively short time intervals [19,20]. Such effects are induced by the structure and technology of the MOX sensors, which comprise of grains of different size and morphology. Some ambient gases can be stably adsorbed by the grains and induce the ageing and drift of sensitivity. These effects can be reduced by pulse heating or intense UV irradiation, used for sensor cleansing [21].

Improvement in gas sensing stability can be reached by applying the materials of very repeatable structures. Two-dimensional materials (MoS2, WS2, phosphorene, graphene, carbon nanotubes) are of high interest for gas sensing due to their unique properties (high ratio of the sensing area to volume) and repeatable morphology (e.g., graphene layers) [22]. The sensors provide an opportunity to detect even a single gas molecule adsorbed by the active surface. An electronic device, using single-layer graphene for the gate of the field-effect transistor, can detect low gas concentrations and give repeatable results. Recent experimental studies confirmed that 1/*f* noise in such device has a component, called Lorentzian, of the frequency characteristic for the ambient gas (e.g., C4H8O, CH3OH, C2H3N, CHCl3) [23]. Moreover, the experiments gave repeatable results for different samples of electronic devices. We expect that such sensors should detect NH3 molecules at very low concentrations in a similar way as the organic gases mentioned above. It was experimentally confirmed that the oxidized graphene sensor is suitable for NH3 detection [24].

These impressive results were achieved for the structures, which are very repeatable and can be spoiled only by some limited impurities situated on the two-dimensional layers. Unfortunately, their production is expensive and requires specialized equipment. We may expect that low-cost MOX gas sensors might be produced using a mixture of graphene flakes decorated with the nanoparticles of the selected material for gas sensing. These sensors might be less sensitive than the presented electronic device. However, the morphology of repeatable two-dimensional structures should enhance their sensing properties in conjunction with very low costs and simplified technology of production. Further, gas sensing improvement can be achieved by UV-light irradiation because graphene is a photocatalytic material.

Flexible gas sensing materials are of great interest for emerging applications in wearable electronic devices for portable health monitoring applications. Detection of NH3 is one of the hot topics in this area because ammonia may cause severe harm to the human body and is the most common air contaminant emitted from various sources (e.g., at the decomposition of protein products). The gas sensing materials should be transparent, mechanically durable, and operate at room temperature. New structures of low-cost ammonia gas sensors were proposed recently [25–28]. These structures comprise different materials, including two-dimensional reduced graphene oxide [26], one-dimensional nanostructures (nanowires) [27], or polymers [25,28]. The sensors are chemo-resistive and, therefore, can be used in wearable or handheld portable applications. Moreover, their resistance can be monitored by utilizing triboelectric charging in self-powered wearable applications.

The second group of solid-state ammonia sensors consists of conducting polymer sensors. They represent important class of functional organic materials for next-generation sensors. Their features of high surface area, small dimensions, and unique properties have been used for various sensor constructions. Many remarkable examples have been reported over the past decade. The enhanced sensitivity of conducting polymer nanomaterials toward various chemical/biological species and external stimuli made them ideal candidates for incorporation into the design of the sensors. However, the selectivity and stability can be further improved.

Advances in nanotechnology allow the fabrication of various conducting polymer nanomaterials [29]. Among the conducting polymers, polyaniline (PANI), polypyrrole (PPy), polythiophene (PTh), and poly (butyl acrylate) (PBuA) or poly (vinylidene fluoride) (PVDF) are the most frequently used in the ammonia sensor as the active layer [30–32]. Gas sensors with conducting polymer are based on amperometric, conductometric, colorimetric, gravimetric, and potentiometric measuring techniques. However, their selectivity and stability should be also improved.

Some recently reported solid-state ammonia sensors are summarized in Table 1.


**Table 1.** Parameters of some solid-state ammonia sensors.

Where: GO—graphene oxide, RGO—reduced graphene oxide, RT—room temperature, PANI—polyaniline, PPy—polypyrrole, SWNT—single-walled carbon nanotubes.

Colorimetric sensors utilize optical measurements of the sample (porous matrix) interacting with the ambient gas. Therefore, we present these gas sensors in the group of solid-state sensors. They are used for the measurement of NH3 in blood, urine, and wastewater. Colorimetric gas sensors are based on the change in color of a chemochromic reagent incorporated in a porous matrix such as porphyrin-based or pH indicator-based films. To detect this change in color, three basic components are needed: a light source, the chemochromic substance, and the light sensor. Liu et al. [43] has developed a solid-state, portable, and automated device capable to measure total ammonia amount in liquids, including the biological samples (e.g., urine). The idea of operation of the colorimetric sensor is shown in Figure 2. A horizontal gas flow channel passing through the sensing chamber, a red LED light source, and four photodiodes (a sensing-reference pair and a sensing-reference backup pair) were applied. The target gas is exposed to the sensor which then exhibits a color change proportional to the NH3 concentration. The photodiodes convert the color change to electronic signals. Such a sensor is of

high sensitivity, short response time, and fast reversibility for NH3 gas concentrations ranging from 2 ppm to 1000 ppm.

**Figure 2.** The schematic of the colorimetric optoelectronic ammonia sensor [43].

#### **3. Optical Methods**

Most of the optical approaches to ammonia detection are based on the effects of absorption and luminescence, more rarely on the refraction or the light reflection.

In an optical absorption technique, the measured gas is contained in the sensor chamber. The radiation passing through the chamber can be absorbed by gas molecules. Measuring the light absorption at the specific wavelength (λ), with possibly no other gas species absorbing in this spectral range, the concentration (*N*) of ammonia can be determined. The concentration estimation follows the Beer–Lambert absorption law

$$
\sigma(\lambda)L = \sigma(\lambda)NL = \ln\left(\frac{I\_0}{I}\right) \tag{1}
$$

where α(λ) is the absorption coefficient defined as the logarithmic ratio of the incident (*I*0) and the transmitted (*I*) light intensities, *L* denotes the light path length in the chamber, and σ(λ) denotes the absorption cross section. Ammonia exhibits UV and IR absorption bands around 135 nm, 155 nm, 195 nm, 1.5 μm, 2 μm, 3 μm, 4 μm, 6 μm, around 11 μm, and 16 μm (Figure 3). The cross-section (σ), defining the system sensitivity, is the highest in the UV region. It is about 10-times larger than at the 11 μm region (e.g., ~10−<sup>18</sup> cm2) and even 105-times larger than in other NIR ranges [44].

**Figure 3.** Absorption cross-section of ammonia for different temperatures in UV [45,46] (**a**). Absorption cross-section of selected molecules existing in standard conditions (1 atm, 288.2 K) calculated on the basis of the HITRAN database in IR range (**b**).

Unfortunately, in the UV and NIR range, there is a problem with interferences by water vapor and N2O or NO2 molecules, which occur at high concentrations in air. Their absorption bands might overlap with NH3 spectra disturbing the measurements. Therefore, the selection of proper spectral range is crucial for successful optical detection.

Nondispersive infrared (NDIR) sensing belongs to simplest approaches of NH3 detection. Figure 4 shows a schematic diagram of such a gas sensor. It usually consists of a broadband source (cheaper and smaller black body emitters or IR LEDs), absorption cell, optical filters, and detectors. Radiation from the broadband source passes through the chamber and two filters. The first filter covers the absorption band of the target gas (named active channel), while the other covers a non-absorbed spectral range (the reference channel). Transmission bands of the filters should not overlap with absorption bands of the other gases present in the chamber. Thus, absorption in the active channel is proportional to NH3 concentration. The light transmitted through the reference channel is not attenuated. Therefore, it is used to compensate instabilities of the light source. The sensitivity of NDIR is influenced by the intensity of the source, the optical waveguide and detector parameters. The disadvantage of this technique is the low precision of the detecting small signal changes at an eventual large background, which results in low selectivity and a high detection limit.

**Figure 4.** Design and principle of operation of the nondispersive infrared (NDIR) sensor.

There are various methods to improve the performance of NDIR sensors. For example, Max-IR Labs utilizes the NDIR technique together with fiber-optic evanescent wave spectroscopy (FEWS) for ammonia detection [47]. The IR radiation is transmitted through a silver-halide (AgClxBr1-x) optical waveguide without cladding and the detection performed by means of the evanescent field (Figure 5). The maximum of the peak due to ammonia absorption was observed at 1450 cm−1. Such a sensor allows NH3 detection beyond a 1 ppm limit.

**Figure 5.** Diagram of the sensor based on NDIR and fiber-optic evanescent wave spectroscopy (FEWS) principles.

Laser spectroscopy is the best choice for trace gas analysis among optical approaches. It is characterized by high sensitivity and selectivity. Ammonia can be detected using tunable laser absorption spectroscopy (TLAS), multi-pass optical cell (MUPASS), cavity ring down spectroscopy (CRDS), cavity-enhanced absorption spectroscopy (CEAS), and photoacoustic (PAS) approach.

TLAS is a very interesting technique for ammonia concentration measuring using tunable diode lasers. The advantage of TLAS over other techniques consists in its ability to achieve low detection limits by using optical path extension techniques and improving the signal-to-noise ratio (SNR).

A single-pass TLAS system, operating in direct absorption measurement mode, consists of a tunable laser, transmitting optics, sample cell, receiving optics, photodetector, and a signal detection circuit (Figure 6). The optics matches the laser and photodetector to the gas inside the sample cell. A thermoelectric controller (TEC) is used to set the laser-operating temperature to a value where the desired wavelength can be reached due to injection current tuning. The injection current is usually scanned periodically with a ramp signal, which leads to laser wavelength scanning. The amplitude of the scan should cover the absorption transition of interest. Laser radiation passes through the absorption cell and the transmitted signal is measured using the photodetector. In absence of the absorption, the detector signal represents laser power changes vs. the current. When absorption occurs, a dip in transmission is observed. Ratio of the signals registered at the line center corresponding to the case without absorption (*I*0) and with absorption (*I*1) can be used to calculate the gas concentration according to the formula (1).

**Figure 6.** Block diagram and principle of operation of the typical tunable laser absorption spectroscopy (TLAS) sensor.

Based on TLAS, different components such as NH3, CO, O2, CH4, H2O, CO2, and HCl can be detected with high selectivity and sensitivity. TLAS has been employed for various applications, including industrial process monitoring and its control, environmental monitoring, combustion and flow analyses, trace species measurements, and so on. However, the sensitivity of this technique is usually limited to the absorption coefficient value of ~10<sup>−</sup>2–10−<sup>3</sup> cm−<sup>1</sup> (Equation (1)).

Wavelength modulation spectroscopy (WMS) is a useful technique providing SNR improvement, which is used in high sensitivity applications [48]. Small modulation (with frequency *f* ≈ 1–20 kHz) is added to the laser current scan, which is provided in a similar way as in TLAS (Figure 7). The transmitted signal, detected by the photodiode, is fed to a lock-in amplifier. As the laser is scanned across an absorption line profile, the transmitted on-absorption signal changes at the frequency of 2*f*, while the transmitted off-absorption signal changes at the frequency of *f* (Figure 8).

**Figure 7.** The principle of operation of the sensor using a combination of TLAS and wavelength modulation spectroscopy (WMS) techniques.

**Figure 8.** Idea of first and second harmonic detection with tunable diode lasers.

Application of high frequency minimizes 1/*f* noise. Therefore, setting the lock-in band around second harmonic ensures that the sensor becomes more suitable for high sensitivity applications compared to the standard TLAS approach. Typical sensitivity limits the absorption coefficient achievable with WMS and is about 10−<sup>4</sup> cm−1. Better limits, about 10−5–10−<sup>7</sup> cm−1, can be obtained for balanced detection-based WMS [49]. This method involves an electrical circuit that subtracts the photocurrents of two detectors: one of them registers the reference laser intensity while the other measures the signal passing through the absorption cell. That provides opportunity to reject common mode laser noise.

Small absorptions, which occur due to low densities of molecules in the samples or due to weak line strengths, are usually compensated by extending the optical path with multi-pass cells (White or Herriott) or cavity-enhanced methods. They enable the optical path to be extended. If instead of a single-pass chamber we apply the multi-pass cell (Figure 9), the sensitivity of the sensor would be improved by the factor of the optical path lengthening. A key parameter of multi-pass cells is the ratio of path length to volume. Moreover, this system provides an opportunity for the simultaneous application of the WMS approach and *2f* detection.

**Figure 9.** Scheme of a multi-pass experimental system for multi-pass optical cell (MUPASS)– WMS spectroscopy.

Unlike conventional configurations, which involve at least a pair of mirrors separated by exactly defined distances, circular multi-pass cells have been developed (Figure 10) [50]. Thanks to the single piece, the cell is especially robust against thermal expansion. Minimizing the cell size is desired for the development of fast and portable gas sensors.

**Figure 10.** The circular multi-pass—small volume cell.

Cavity ring-down spectroscopy (CRDS) exploits the sample cells in the form of optical cavities (resonators) built with mirrors of very high reflectivity. A simplified scheme of such experiment is presented in Figure 11. Laser pulse injected into the cavity through one of the mirrors is then reflected many times among them. Its wavelength must be tuned to the NH3 spectral line in order to measure the absorption coefficient of ammonia contained inside. The light transmitted through the exit mirror is monitored by a detector. Analysis of the output signal by the acquisition system provides opportunity to determine the Q-factor of the cavity which is limited due to diffraction and mirror losses as well as due to the light absorption or scattering inside the cell. As far as the Q-factor is inversely proportional to the signal decay time, its value is found due to analysis of the photoreceiver signal by the acquisition system. The majority of the approaches exploit a two-step procedure consisting of the decay time measurement: first, when the cavity is empty (τ0), and then, when the cavity is filled with the tested gas (τ*A*). These decay times depend on the mirror reflectivity, resonator length, and extinction factor (absorption and scattering of light in the cavity) [51]. Comparing both decay times, the absorber concentration can be found.

$$N = \frac{1}{c\sigma(\lambda)} \left(\frac{1}{\tau\_A} - \frac{1}{\tau\_0}\right) \tag{2}$$

where *c* is the light speed and σ(λ) is the absorption cross-section.

**Figure 11.** Idea of the cavity ring down spectroscopy (CRDS).

There are various approaches to CRDS with pulsed lasers or with continuously operating AM modulated ones. Using these techniques, the absorption coefficients α = σ*N* < 10−<sup>9</sup> cm−<sup>1</sup> can be observed. The detection limit is mainly related to the resonator quality (determined by τ0), but also by the precision of decay time measurement. The advantage of CRDS over other absorption spectroscopy approaches consists not only in dominant sensitivity. Their superiority also results from minimized impacts of light source intensity fluctuations or detector sensitivity changes on the measurement results.

This method is extremely sensitive but requires that the laser frequency is precisely matched to the cavity mode. In this way, it is possible to obtain a high Q*-*factor of the resonator and efficient storage of optical radiation. On the other hand, small mechanical instabilities cause changes in the cavity mode frequency and significant output signal fluctuations, which is the main disadvantage of this method [52]. It can be minimalized by use of cavity length stabilization [53,54], or by application of the cavity with dense mode structure, called cavity-enhanced absorption spectroscopy (CEAS) [55].

In CEAS, the off-axis arrangement of the laser and cavity is applied. Similarly, to the conventional CRDS system, the light is repeatedly reflected by the mirrors. However, the reflected beams, which correspond to different trips, are spatially separated inside the resonator. The use of highly reflecting mirrors provides huge extension of the effective optical path. As result, either a dense mode structure of low finesse occurs, or the mode structure does not establish at all. In this way, sharp resonances of the cavity are avoided, so the system is much less sensitive to mechanical instabilities. The CEAS sensors attain the detection limit of about 10−<sup>9</sup> cm−<sup>1</sup> [56]. Detailed information about CRDS and CEAS techniques is presented in the authors' papers [57–59].

Among the so-called in-situ sensors, photoacoustic spectroscopy (PAS) belongs to the most popular one. In PAS, conversion of laser light energy into an acoustic wave is applied. The gas sample, placed in photoacoustic chamber, is irradiated by optical radiation, AM modulated with acoustic frequency. The laser wavelength is matched to the absorption line of a molecule of interest. If the absorber is present in the cell, a portion of optical radiation is converted onto heat energy. Then, a local and periodic growth of temperature and pressure occurs (Figure 12). The resulting acoustic wave is detected at modulation frequency by a very sensitive microphone placed in the chamber. The PAS signal, which is proportional to the absorber concentration (N), is given by

$$A(T,\lambda) \propto P\_o \text{Na}(T,\lambda) L \frac{Q}{fV} \eta = P\_o \text{Na}(T,\lambda) \frac{Q}{fA} \eta \tag{3}$$

where *Po* denotes average laser power, *Q* is the quality factor of the resonant cell, *f* is the frequency of modulation, *V* is the gas volume, *A* is the cross-sectional area, and η is the system efficiency factor (e.g., microphone efficiency and loss factors).

**Figure 12.** Idea of photoacoustic spectroscopy.

The microphone output signal is recorded using a low-noise preamplifier and a lock-in voltmeter. The strongest effect is achieved when the modulation frequency is matched to the resonance of the photoacoustic chamber. In order to increase the signal, various constructions of photoacoustic cells (acoustic resonators with higher Q-factor) are applied.

The improvement of this technique, which allows to achieve very high detection sensitivity, is the quartz-enhanced photoacoustic spectroscopy (QEPAS). The basic principle of operation is similar, but here the laser beam is focused between the U-shaped prongs of a resonant piezo-quartz fork [60]. This type of acoustic transducer is sensitive to signals generated by asymmetric prongs oscillations caused by acoustic wave, induced by the AM-modulated laser radiation. External sources of interference do not provide output signals because they cause symmetrical oscillations. Similarly to the conventional PAS system, in QEPAS set-ups, the measurements are performed with a wavelength modulation technique using the 2*f* detection [61]. Combined with the high Q-factor of the quartz fork, it gives the opportunity to build ultra-compact sensors that reach detection limits comparable to that of CEAS. In case of ammonia, it corresponds to a few ppb or even sub-ppb level.

Recently reported parameters of some optical ammonia sensors are given in Table 2.


**Table 2.** Parameters of some optical ammonia sensors.

QCL—quantum cascade laser, PAS—photoacoustic spectroscopy, CRDS—cavity ring down spectroscopy, QEPAS—quartz-enhanced photoacoustic spectroscopy, WMS—wavelength modulation spectroscopy, R—reflectivity of mirrors, L—optical cavity length, EC—external cavity, p—pressure. Other methods of gas detection were described in our earlier studies [68,69].

Among many applications, the ammonia sensors for medical purpose are an important part of such detectors. The normal concentration of ammonia in the breath of a healthy man is in the range of 0.25–2.9 ppm [70]. An excessive concentration might suggest renal failure, *Helicobacter Pyroli*, diabetes, and oral cavity disease [71,72]. The main problem in the construction of ammonia sensors for breath analysis does not consist in extreme sensitivity but in high selectivity. More than 3000 various constituents were already detected in exhaled air [73]. Their absorption spectra might overlap the spectral fingerprint of the ammonia and disturb the measurement. Water vapor and carbon dioxide are the main interferences since their concentrations can reach up to 5% in breath, and thus it exceeds the ammonia density by many orders of magnitude. Carbon monoxide and methane should also be taken into account.

As it was mentioned above, the highest sensitivities of ammonia detection using laser absorption spectroscopy can be achieved in UV and in a range of 10–11 μm, due to the largest values of the absorption cross-sections. However, the detection at the wavelengths used in telecommunication 1.4–1.6 μm is more convenient because of a large variety of relatively cheap laser sources, optical elements, and photodetectors. The measurements of ammonia around 1.51 μm with the detection limit of 4 ppm was already demonstrated using the CEAS technique [74]. We found that suitable wavelengths for ammonia detection are also the lines at 1.5270005 μm and 1.5270409 μm (Figure 13). The single-mode laser wavelength was controlled with the HighPrecision lambdameter (model WS6) ensuring precision in a short time (1 min) of 0.0005 nm. The NH3 absorption coefficient (at 2 ppm) reaches here about 3.5·10−<sup>6</sup> cm<sup>−</sup>1, and experimental techniques like MUPASS–WMS can be effectively used. In this range, carbon dioxide and water vapor interference is about eight times weaker.

**Figure 13.** Absorption spectra of ammonia, H2O, and CO2 around 1.527 μm [75].

Our ammonia sensor used the MUPASS–WMS approach (Figure 9). A single-mode diode laser (Toptica, model DL100, 20 mW) was applied as a light source. Output signals were integrated over 1 min. Results of sensor investigation are presented in Figure 14. It provides proper results for NH3 concentrations of up to about 1 ppm. For lower values, the deviation from a linear characteristic is observed. This is probably a result of poor regulation of reference concentration inside the sensor due to NH3 deposit on the walls of the system, which causes a systematic error of measured data. In order to avoid the concentration measurement disturbed by NH3 molecules adsorbed on the walls (and then desorbed), the sensor was kept at a temperature of 50 ◦C.

Nevertheless, due to the proper choice of spectral lines, we achieved a good immunity of the detection against H2O and CO2 at the concentrations which might occur in breath (Figure 13). The detection limit of 1 ppm was better than in other experiments preformed at a spectral range of 1.5 μm [74]. Our multi-pass sensor is suitable for rough monitoring of ammonia in the exhaled air and for detection of morbid states (>1.5 ppm).

**Figure 14.** Results of ammonia measurement with MUPASS–WMS system [75].

#### **4. Conclusions**

Based on the analysis, it can be concluded that the further development of in-situ ammonia detection technologies will focus on the improvement of sophisticated laboratory equipment, e.g., the mass spectrometers and lasers. One can expect the development of sensors dedicated to a specific application. This article focuses on the second group of the sensors in which two technologies with high potential for application and development have been distinguished: solid-state ammonia sensors and laser absorption spectroscopy. Metrological and operational parameters of such sensors largely depend on the application. Basically, they must be characterized by fast operation, the largest measuring range and the highest measurement precision, high reliability, selectivity, as well as small dimensions, and low cost. Sensors belonging to the former group can detect the ammonia with a concentration of several ppm within tens of seconds. Their advantage consists in very small dimensions and very low cost. Materials engineering and nanotechnology are key technologies for the development of this sensor group. We underlined that recent advances in materials and measurement methods enhance gas detection by such low-cost gas sensors in portable applications. We are conscious that our research presents a limited number of NH3 gas sensors. This issue is a hot topic in the gas sensing area, and other attractive sensors were developed recently (e.g., capacitive sensors [76], colorimetric sensing materials [77], or advanced MOX sensors [78]). The latter group of sensors can detect up to four orders of lower ammonia in fractions of a second. However, their costs are much higher, mainly due to the lasers and high-quality optical components. The further improvement of these sensors is much more dependent on the development of lasers [79] than photodetectors [80], because they are already at a very high level. It mainly concerns a progress in lasers for mid-infrared radiation, since one can expect the highest sensitivity and selectivity in this spectral range. However, miniaturization, reliability, and reduction of production costs are still a current challenge for all optical sensor components. It should be also mentioned that MOX sensors can utilize optical methods in some sense by applying UV light to improve the sensitivity and selectivity of gas detection. Therefore, both types of gas sensors start to interlace MOX and optical sensors technologies to reduce costs and popularize their applications. We believe that different types of gas sensors find complementary areas of new applications.

**Author Contributions:** Z.B.: conceptualization, writing—original draft, visualization. T.S.: writing—review & editing, investigation. J.S.: writing—review & editing, investigation. J.W.: formal analysis, investigation, writing—review & editing. All authors have read and agreed to the published version of the manuscript.

**Funding:** The publication was supported by The Polish National Centre for Research and Development as part of the "Sense" project, ID 347510.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Trends in Performance Limits of the HOT Infrared Photodetectors**

**Antoni Rogalski 1, Piotr Martyniuk 1,\*, Małgorzata Kopytko <sup>1</sup> and Weida Hu <sup>2</sup>**


Chinese Academy of Sciences, 500 Yu Tian Road, Shanghai 200083, China; wdhu@mail.sitp.ac.cn

**\*** Correspondence: piotr.martyniuk@wat.edu.pl; Tel.: +48-26-183-92-15

**Abstract:** The cryogenic cooling of infrared (IR) photon detectors optimized for the mid- (MWIR, 3–5 μm) and long wavelength (LWIR, 8–14 μm) range is required to reach high performance. This is a major obstacle for more extensive use of IR technology. Focal plane arrays (FPAs) based on thermal detectors are presently used in staring thermal imagers operating at room temperature. However, their performance is modest; thermal detectors exhibit slow response, and the multispectral detection is difficult to reach. Initial efforts to develop high operating temperature (HOT) photodetectors were focused on HgCdTe photoconductors and photoelectromagnetic detectors. The technological efforts have been lately directed on advanced heterojunction photovoltaic HgCdTe detectors. This paper presents the several approaches to increase the photon-detectors room-temperature performance. Various kinds of materials are considered: HgCdTe, type-II AIIIBV superlattices, two-dimensional materials and colloidal quantum dots.

**Keywords:** HOT IR detectors; HgCdTe; P-i-N; BLIP condition; 2D material photodetectors; colloidal quantum dot photodetectors

#### **1. Introduction**

HgCdTe takes the dominant position in infrared (IR) detector technology. This material has triggered the rapid development of the three "detector generations" considered for military and civilian applications and briefly described in the caption of Figure 1. IR detector technology combined with fabrication of epitaxial heterostructure [by molecular beam epitaxy (MBE) and metalorganic chemical vapor deposition (MOCVD)] and photolithographic processes revolutionized the semiconductor industry, thus enabling the design and fabrication of complex focal plane arrays (FPAs). Further their development will relate to implementation of fourth generation staring systems, which the main features are to be: high resolution (pixels > 108), multi-band detection, three-dimensional readout integration circuits (3D ROIC), and other integration functions such as polarization/phase sensitivity, better radiation/pixel coupling or avalanche multiplication. The first three generations of imaging systems primarily rely on planar FPAs. Several approaches to circumvent these limitations, including bonding the detectors to flexible or curved molds, have been proposed [1]. Evolution of fourth generation is inspired by the most famous visual systems, which are the biological eyes. Solution based on the Petzval-matched curvature allows the reduction of field curvature aberration. In addition, it combines such advantages as simplified lens system, electronic eye systems and wide field-of-view (FOV) [2,3]. The colloidal quantum dot (CQD) [4] and 2D layered material [5] photodetectors fabricated on flexible substrates exhibit the potential to circumvent technical challenges in the development of the fourth generation IR systems.

**Citation:** Rogalski, A.; Martyniuk, P.; Kopytko, M.; Hu, W. Trends in Performance Limits of the HOT Infrared Photodetectors. *Appl. Sci.* **2021**, *11*, 501. https://doi.org/ 10.3390/app11020501

Received: 30 October 2020 Accepted: 14 December 2020 Published: 6 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Figure 1.** The history of IR detectors and systems development. Four generation systems for military and civilian applications can be considered: first generation (scanning systems), second generation (staring systems—electronically scanned), third generation (staring systems with large number of pixels and two-color functionality), and fourth generation (staring systems with very large number of pixels, multi-color functionality, 3D ROIC, and other on-chip functions; e.g., better radiation/pixel coupling, avalanche multiplication in pixels, polarization/phase sensitivity.

The need for cooling considerably limits more widespread use of IR technology. There are significant attempts to decrease system size, weight, and power consumption (SWaP) to limit IR system's cost and to increase the operating temperature. The invention of microbolometer array was a milestone step in development of IR cameras operating at 300 K. However, microbolometers belong to the class of thermal detectors with limited response time—typically in millisecond range and could not be used in the multiband applications. To omit this limitation, further efforts are directed to increase operating temperature of photon detectors.

Initial efforts in development of the high operating temperatures (HOT) photodetectors were focused on HgCdTe photoconductors and photoelectromagnetic detectors [6]. Many concepts have been implemented and tested to improve the performance of IR photodetectors operating at near 300 K and compiled in References [7–10]. In addition to photoresistors and photodiodes, three other types of IR detectors can operate at near 300 K: magnetic concentration detectors, photoelectromagnetic (or PEM) detectors and Dember effect detectors. The HgCdTe non-equilibrium devices such as the Auger suppressed excluded photoconductors and extracted photodiodes require significant bias what creates excessive 1/f noise.

Up till now, mainly HgCdTe and Sb-based III-V ternary alloys including barrier detectors with type-II superlattices (T2SLs: InAs/GaSb and InAs/InAsSb) have been considered for HOT IR photodetectors. The recently published monograph covers this topic for III-V material systems [11]. In the past decade considerable progress in development of interband quantum cascade infrared photodetectors (IB QCIP) based on T2SLs, 2D material [12] and CQD photodetectors brought their performance close to commercial ones [13].

In 1999 Elliott et al. claimed that there is no fundamental obstacle to reach 300 K operation of photon detectors with background-limited performance even in reduced fields of view [14]. In this paper we attempt to reconsider the performance of different material systems for the HOT detection operation in IR spectral range. Theoretical estimates are collated with experimental data for different photodetectors.

#### **2. Trends in Development of Infrared HOT Photodetectors**

As is shown by Piotrowski and Rogalski [15], the IR detectors performance is limited by statistical character of generation-recombination processes in the material. Thermal processes in the device material limit the detectivity, *D*\* of an optimized IR detector. It can be given by the following equation

$$D^\* = k \frac{\lambda}{\hbar c} \left(\frac{a}{G}\right)^{1/2},\tag{1}$$

where λ—Wavelength, *h*—Planck's constant, *c*—Light speed, *α*—Absorption coefficient, *G*— Thermal generation in the active detector's region, *k*—Coefficient depending on radiation coupling to the detector. *α*/*G* is the absorption coefficient to the thermal generation rate ratio and can be considered as the fundamental figure of merit of any material used for IR detectors (*α*/*G* ratio could be used to evaluate any material). Among different bulk materials, the narrow gap semiconductors are more suitable for the HOT photodetectors than competitive technologies, such as extrinsic devices, Schottky barrier photodiodes, quantum dot infrared photodetectors (QDIP) and quantum well infrared photodetectors (QWIP) [16]. The high performance of intrinsic photodetectors results from high density of states in the conduction and valence bands (contributing to high IR absorption), and long carrier lifetime (contributing to low thermal generation).

The goal of IR detector technology is the fabrication of HOT photodetector characterized by the dark current lower than the background flux current and 1/f noise negligibly lower than to the background flux shot noise [17,18].

In several papers, it was shown that the detector size, d, and F-number (f/#) are the main IR systems parameters [19,20]. Since they depend on *F*λ/*d* (λ—Wavelength), both influence the detection/identification range, as well as the noise equivalent difference temperature (NEDT) [20]

$$Range = \frac{D\Lambda x}{M\lambda} \left(\frac{F\lambda}{d}\right),\tag{2}$$

$$NEDT \approx \frac{2}{C\lambda \left(\eta \Phi\_B^{2\pi} \tau\_{int}\right)} \left(\frac{F\lambda}{d}\right) \,\tag{3}$$

where *D*—Aperture, *M*—Needed number of pixels to identify a target Δ*x*, *C*—Scene contrast, η—Quantum efficiency (QE), Φ2*<sup>π</sup> <sup>B</sup>* —Background flux into a 2*π* FOV, *τint*—Integration time. According to the relations (2) and (3), the *F*λ/*d* parameter could be used for IR system optimization. For the f/1 optics, the smallest practicable detector size should be ~2 μm for the MWIR and ~5 μm for the LWIR, respectively [21]. With more realistic f/1.2 optics, the smallest practicable detector size is ~3 μm and ~6 μm for the MWIR and LWIR, respectively.

Kinch claims that the IR system ultimate cost reduction could only be reached by the 300 K operation of depletion-current limited arrays with pixel densities that are fully consistent with background and diffraction-limited performance due to the system optics [20]. The depletion-current limited P-i-N photodiodes demand long Shockley-Read-Hall (SRH) carrier lifetime, marked as τSRH, to meet the requirements of a low dark current. The long HgCdTe SRH lifetime makes this material a great candidate for 300 K condition [20].

#### **3. The Ultimate HgCdTe Photodiode Performance**

In 2007, the Teledyne Technologies published an empirical expression, called "Rule 07", for estimation of the P-on-n HgCdTe photodiodes dark current versus normalized wavelength-temperature product (λcT) [22]. This metric is closely related to Auger 1 diffusion-limited photodiode with n-type active region doping concentration ~10<sup>15</sup> cm<sup>−</sup>3. In the past decade, the Rule 07 has become very popular as a reference level for the other technologies (especially to III-V barrier and T2SLs devices). However, at present stage of technology, the fully depleted background limited HgCdTe photodiodes can reach the level of 300 K dark current considerably lower than predicted by the Rule 07. The discussion below explains exactly this statement.

#### *3.1. SRH Carrier Lifetime*

The SRH generation-recombination mechanism determines the carrier lifetimes in both lightly doped n- and p-type HgCdTe in which SRH centres are related to residual impurities and native defects. Kinch et al. in 2005 [23] published that the experimental carrier lifetimes for n-type LWIR HgCdTe range from 2 up to 20 μs at 77 K irrespective of doping <1015 cm−3. The MWIR carrier lifetime are substantially longer assuming 2 up to 60 μs. However, several papers published in the last decade have shown τSRH significantly higher in low temperature range and low doping concentrations, above 200 μs up to even 50 ms versus cut-off wavelength [20] see Table 1. The range of low doping that can be reproducibly obtained in Teledyne Technologies HgCdTe epilayers grown by MBE is ~10<sup>13</sup> cm−3. Gravrand et al. [24] published that for most tested MWIR photodiodes from CEA Leti and Lynred by Sofradir & Ulis, the estimated SRH carrier lifetimes [from direct measurements (photoconductive or photoluminescence decay) and indirect estimates from current-voltage (I-V) characteristics], are in the range between 10 and 100 μs. Those values are lower than the earlier assessed by US research groups [25]. However, they were estimated for devices with higher doping level in absorber >1014 cm−3. However, from just published announcement results, Teledyne Technologies confirmed fabrication of depletion layer limited P-i-N HgCdTe photodiodes with SRH recombination centers exhibiting carrier lifetimes in the range 0.5–10 ms [25].

**Table 1.** Summary of the SRH carrier lifetimes determined based on current-voltage characteristics (data after reference [20]).


All SRH lifetimes estimated for HgCdTe are usually carried out for temperatures below 300 K. Their extrapolation to 300 K to predict the photodiode operation behavior is questionable. In our estimates we assume τSRH = 1 ms, which is supported by experimental data reached by Leonardo DRS and Teledyne Technologies research groups.

Figure 2a shows a schematic P-i-N detector energy band profile for a reverse voltage. The active region consists of an undoped i-layer (often called as ν region exhibiting low n-doping) sandwiched between wider bandgap cap (P) and buffer (N) region (see Figure 2b) [26]. Very low doping in the absorber (below 5 × 1013 cm<sup>−</sup>3) is required to allow full depletion at zero or low reverse voltage. The surrounded wide gap contact layers are designed to reduce the dark current generation from these regions and to prevent tunneling current under reverse bias. Moreover, fully depleted absorber surrounded by wide bandgap regions theoretically reduces 1/f and burst noise. As previously mentioned,

the fully depleted P-i-N architecture is compatible with the small pixel size, meeting the requirements of low crosstalk thanks to the built-in electric field [20,26].

**Figure 2.** P-i-N detector: (**a**) energy band profile under reverse bias, (**b**) heterojunction architecture (adapted after reference [26]).

In P-i-N design, the choice of absorber thickness should be a trade-off between the response time and QE (or responsivity). To reach short response times, the absorber thickness should be thin and fully depleted. For high QE the absorption region should be thick enough to effectively collect photogenerated carriers. However, to enhance QE while maintaining high response time, an external resonant microcavity was demonstrated [8]. In this design, absorber is placed inside a cavity so that more photons can be absorbed even in low detection volume.

#### *3.2. Dark Current Density*

In general, for the fully depleted P-i-N photodiode, the current is built by diffusion from N and P regions (that depends on SRH and Auger generations) and depletion current only determined by SRH generation in the space charge region. Influence of radiative recombination is still debatable but is not considered as a limiting factor of the small pixel HgCdTe photodiodes. Moreover, due to the photon recycling effect, the radiative recombination contribution can be significantly reduced [27]. By that reason in our discussion the radiative recombination is omitted.

The diffusion current of P-i-N HgCdTe photodiode structure arises from the thermal generation of carriers in thick, non-depleted absorber and is dependent on the Auger and SRH generation in n-type semiconductor [20]

$$J\_{diff} = \frac{qn\_i^2 t\_{diff}}{n} \left(\frac{1}{\tau\_{A1}} + \frac{1}{\tau\_{SRH}}\right),\tag{4}$$

where *q*—Electron charge, *ni*—Intrinsic carrier concentration, *n*—Electron concentration, *tdif*—Diffusion region thickness, *τA*1—Auger 1 lifetime, and *τSRH*—SRH lifetime. Auger 1 lifetime relates to the hole, electron, and intrinsic carrier concentrations, and *τA*<sup>1</sup> is given by equation

$$\pi\_{A1} = \frac{2\pi\_{A1}^i n\_i^2}{n(n+p)'} \tag{5}$$

where *p*—Hole concentration and *τA*1—Intrinsic Auger 1 lifetime. For a low temperature operation or a non-equilibrium active volume, when the majority carrier concentration is held equal to the majority carrier doping level [and intrinsically generated majority carriers are excluded (p << n ≈ Ndop)], Equation (5) becomes

$$
\pi\_{A1} = \frac{2\pi\_{A1}^i n\_i^2}{n^2}.\tag{6}
$$

The shortest SRH lifetime occurs through centers located approximately at the intrinsic energy level in the semiconductor bandgap. Then, for the field-free region in an n volume (*n* >> *p*), *τSRH* is given by

$$
\pi\_{SRH} = \frac{\pi\_{no}n\_i + \pi\_{po}(n + n\_i)}{n},
\tag{7}
$$

where *τno* and *τpo*—specific SRH lifetimes. At low temperatures, where n > *ni*, we have *τSRH* ≈ *τpo*. At high temperatures where n ≈ ni, we have *τSRH* ≈ *τno* + *τpo*. For a nonequilibrium active volume, *τSRH* ≈ (*τno* + *τpo*)*ni*/*n* exhibits a temperature dependence given by ni.

The second component is the depletion current arising from the portion of the absorber that becomes depleted. The depletion current density can be assessed by the relation

$$J\_{dep} = \frac{qn\_it\_{dep}}{\tau\_{no} + \tau\_{po}} \,\prime \tag{8}$$

where *tdep*—Depletion region thickness.

The P-i-N HOT detector is characterized by useful properties at reverse voltage. Figure 3 shows the calculated reverse voltage required to completely deplete a 5-μm-thick absorber for selected doping level. For the Rule 7 with doping range about 10<sup>15</sup> cm−3, a 5-μm-thick absorber can be fully depleted by applying a relatively high reverse bias between 10 V and 30 V. On the other hand, for the doping level reached presently at Teledyne Technologies (~10<sup>13</sup> cm−3), the 5-μm-thick active layer can be fully depleted for reverse bias from zero up to 0.4 V.

**Figure 3.** Calculated reverse voltage versus doping concentration required to deplete a 5-μm-thick MWIR HgCdTe absorber. Inset: absorber depletion thickness versus reverse bias and selected doping concentration.

If P-i-N photodiode operates under reverse bias, Auger suppression effect should be considered. This effect is important in HOT condition when ni >> Ndop. In nonequilibrium, large number of intrinsic carriers can be swept-out of the absorber region. It is expected that this impact is larger for lower n-doping levels since ni is proportionately higher. At very low level of n-type doping (about 1013 cm<sup>−</sup>3) the P-i-N photodiode ultimate performance is influenced by SRH recombination and neither Auger recombination nor Auger suppression.

As is shown in Figure 4, for the sufficiently long SRH carrier lifetime in HgCdTe, the internal photodiode current is limited, and the performance is contributed by the background radiation. Its influence is shown for four background temperatures: 300, 200, 100 and 50 K. Lee et al. suggested to replace Rule 07 by Law 19 corresponding exactly to the background limited curve for room temperature [25]. The internal photodiode current may be several orders of magnitude below Rule 07 versus given cut-off wavelength and operating temperature. It can be also seen that Rule 07 coincides well with theoretically predicted curve for the Auger-suppressed p-on-n photodiode with absorber doping concentration Nd = 1015 cm<sup>−</sup>3.

**Figure 4.** Current density of p-on-n HgCdTe photodiodes versus 1/(λcT) product (adapted after reference [25]). Experimental data is gathered for Teledyne Technologies and alternative technologies [22,25,28–31].

The experimental data for p-on-n HgCdTe photodiodes (Teledyne Technologies) [25] and for III-V barrier detectors (Raytheon Technologies [28] and SCD SemiConductor Devices [29]) operating at about 80 K, and 300 K IB QCIP [30] are presented in Figure 4. It is easy to notice that experimental data for III-V barrier detectors are slightly worse than the p-on-n HgCdTe photodiodes, but III-V IB QCIPs operating at 300 K are even better in LWIR. Figure 4 shows also representative data for both InSb (λ<sup>c</sup> = 5.3 μm, T = 78 K) and InGaAs (λ<sup>c</sup> = 1.7 and 3.6 μm, T = 300 K) photodiodes. InSb detector is characterized by several orders higher dark current density than HgCdTe one, however for optimal InGaAs detectors the dark current density is close to HgCdTe data [31].

The theoretical simulations presented in Figure 4 indicate that the background limited performance (BLIP) has the most impact on detector's current density for small 1/(λcT) products; in other words for photodiodes operating in LWIR and HOT conditions. HgCdTe

photodiodes operating at low temperature become generation-recombination limited due to the SRH centers influence the lifetime in the millisecond range.

Figure 5 shows the current density calculated using Rule 07 (determined for diffusion limited P-on-n photodiodes) and Law 19 (which exactly equals to the background radiation current density) versus temperature for short-wave infrared (SWIR: λ<sup>c</sup> = 3 μm), MWIR (5 μm), and LWIR (10 μm) absorber.

**Figure 5.** Calculated current density versus temperature using Law 19 and Rule 07 for SWIR (λ<sup>c</sup> = 3 μm), MWIR (λ<sup>c</sup> = 5 μm), and LWIR (λ<sup>c</sup> = 10 μm) HgCdTe absorber.

If the fully depleted P-i-N detector is influenced by the background current, a certain minimal value of SRH lifetime is required. The SRH lifetime calculations were made under condition where depletion dark current equals the background radiation current

$$J\_{dep} = J\_{BLIP}.\tag{9}$$

It was assumed that the 5-μm thick absorber is fully depleted.

The SRH lifetime at which the fully depleted P-i-N photodiode reaches the BLIP limit is presented in Figure 6. As shown, the SRH lifetime required to reach BLIP limit decreases versus temperature (nevertheless fully depleted P-i-N photodiodes are particularly interesting in HOT conditions). What more, for LWIR detectors, it is possible to reach BLIP for shorter carrier lifetimes. At 300 K and 5-μm fully depleted thick absorber, these carrier lifetimes are 15 ms for SWIR, 150 μs for MWIR and 28 μs for LWIR, respectively.

The Teledyne Technologies experimentally measured SRH lifetimes for 10-μm cut-off HgCdTe are higher than 100 ms (extracted at 30 K) [26]. Despite the fact that at 300 K the carrier lifetimes are likely to be at least 10 times lower (what results from a high thermal velocity increasing the carrier capture probability by the recombination centre), those low SRH lifetimes enable to reach BLIP limit. This prediction is supported by theoretical simulation presented in reference [32].

**Figure 6.** The SRH lifetime versus temperature where fully depleted P-i-N HgCdTe detector depletion dark current equals the background radiation current. The calculations were carried out for SWIR (3 μm), MWIR (5 μm), and LWIR (10 μm) absorbers.

#### *3.3. Detectivity*

The detector's detectivity, D\* is related to the current responsivity, *Ri* [see Equation (13)] and noise current, *in*, and can be given by relation

$$D^\* = \frac{R\_i}{i\_n}.\tag{10}$$

For the non-equilibrium devices, the *in* value can be calculated assuming thermal Johnson-Nyquist and shot noises contribution by the following expression

$$i\_{\rm li} = \sqrt{\frac{4k\_B T}{R\_0 A} + 2qI\_{dark\prime}}\tag{11}$$

where *kB*—Boltzmann constant, *R*0*A*—Dynamic resistance area product and *Jdark*—dark current density.

The performance of P-i-N MWIR HgCdTe photodiode (λ<sup>c</sup> = 5 μm) is presented in Figure 7.

**Figure 7.** MWIR P-i-N HgCdTe photodiode performance with <sup>τ</sup>SRH = 1 ms and absorber doping 5 <sup>×</sup> <sup>10</sup><sup>13</sup> cm<sup>−</sup>3: (**a**) diffusion and depletion current components versus temperature, (**b**) detectivity versus temperature. The thickness of active region: t=5 μm and consists of tdif = 2 μm and tdep = 3 μm. The experimental data is taken from different sources as marked. PV— Photodiode, CQD—Colloidal quantum dot, IB QCIP—Interband quantum cascade infrared photodetector. Experimental data is taken from [30,33,34].

As is shown in Figure 7a, the Teledyne Judson experimentally measured current densities, at the bias −0.3 V, are close to BLIP (f/3) curve and they are located less than one order of magnitude above this limit [34]. The current density at 300 K is even lower than predicted by Rule 07. The measured current densities presented by VIGO System are close to one order of magnitude higher, however in this case they were measured at lower reverse bias, −0.1 V, with less effective Auger suppression [33]. It is interesting to notice, that the performance of IB QCIP based on T2SLs InAs/GaSb coincides well with upper experimental data for HgCdTe photodiodes at 300 K [30].

Figure 7a shows the diffusion and depletion dark current components versus temperature assuming 1 ms SRH carrier lifetime, 5 <sup>μ</sup>m thick absorber and doping 5 × <sup>10</sup><sup>13</sup> cm<sup>−</sup>3. The diffusion component associated with Auger 1 mechanism is eliminated because of the absence of majority carriers due to exclusion and extraction effects [35,36]. The background radiation calculated assuming f/3 optics has decisive influence on dark current. It should be mentioned here that the background flux current is determined by the net flux through the optics (limited by f/#) plus the flux from the cold shield. This effect is shown by increased BLIP (f/3) influence on dark current at temperature >220 K.

Figure 7b shows calculated detectivity versus temperature for MWIR P-i-N HgCdTe photodiode assuming identical parameters as taken in calculations presented in Figure 7a (λ<sup>c</sup> = 5 <sup>μ</sup>m, <sup>τ</sup>SRH = 1 ms, t = 5 <sup>μ</sup>m, Ndop = 5 × <sup>10</sup><sup>13</sup> cm−3). The current responsivity was estimated assuming QE = 1 (however typical QE reaches reasonable value ~0.7). As is shown, for MWIR photodiode with 5-μm cut-off wavelength and low doping in active region, detectivity, D\* is limited by background and is about one order of magnitude higher than predicted by the Rule 07. The experimental data given for HgCdTe photodiodes in Teledyne Judson and VIGO System catalogues are more than one order of magnitude below background flux limitation for the f/3 optics.

#### **4. Interband Quantum Cascade Infrared Photodetectors (IB QCIPs)**

A low diffusion length, weak absorption and finally low dynamic resistance limit the performance of conventional p-n LWIR HgCdTe HOT detectors with doping concentrations in absorbers > 1016 cm<sup>−</sup>3. The QE is limited since the absorption depth of LWIR (λ > 5 μm) is much longer than the diffusion length allowing charge carriers photogenerated at distance shorter than the diffusion length to be collected by the contacts. For example, estimates show that for 10.6-μm detector the absorption depth is ~12 μm while the ambipolar diffusion length is less than 2 μm. In consequence, the QE is reduced to ~15% [9].

To overcome above problems, the multiple heterojunction devices based on thin elements connected in series were proposed, where a proper example is a detector with junctions perpendicular to the substrate, introduced in 1995. The multi-heterojunction device shown in Figure 8a contains backside illuminated n+-p-P detectors connected in series. The advantages of such design are a high voltage responsivity, a short response time while on the other hand the response depends on polarization of incident radiation and is nonuniform across the active area.

**Figure 8.** Backside illuminated multiple HgCdTe heterojunction devices: (**a**) junction's planes perpendicular to the surface, and (**b**) 4-cells stacked multiple detector (after reference [8]).

More promising design is the stacked tunnel junctions connected in series as shown in Figure 8b being similar to multi-junction solar cells. Potentially, this device can reach both good QE, high differential resistance, and fast response. As presented, each cell consists of lightly p-type doped absorber and N+/P+ wide-bandgap highly doped contact layers. The heterojunction contacts collect the photogenerated carriers absorbed in every active layer. However, practical problem is related to the resistance of the adjacent N+ and P<sup>+</sup> layers.

In the last decade, several designs of the multi-stage IR devices have been developed. They are based on III-V semiconductors and might be now divided into two classes: mentioned earlier interband (IB) ambipolar QCIPs and intersubband (IS) unipolar QCIPs. The first study on IS QCIPs began about two decades ago [11] as the photodetectors were developed from the quantum cascade lasers (QCLs). However, currently IB QCIPs show the higher performance in comparison with IS QCIPs due to the relativity much longer carrier lifetime. The IB QCIPs saturation current density is reported to be almost two orders of magnitude lower than for IS QCIPs [30].

Schematic illustration of IB QCIP is presented in Figure 9 where every single active layer is sandwiched between the relaxation and tunneling layers forming a cascade stage. The thickness of the single active layer should be thinner than the diffusion length to effectively collect all photogenerated carriers. The diffusion length restriction in traditional thick absorber detectors is bypassed by using the discrete absorber design imposing recombination of the photogenerated carriers in the next stage within short transport distance. The single thin absorbers are connected in series and the total thickness of the active layer can be even thicker than the absorption depth. The photocurrent is determined by carriers generated in the single absorber (one stage) and is independent of the number of stages meaning that the photons absorbed in following stages do not increase the net photocurrent but only provide the current continuity through the device. The noises suppression for shorter individual absorbers is the advantage of QCIP design. The QCIP detectivity, D\* is influenced by the Johnson and shot noises that is proportional to <sup>√</sup>*<sup>N</sup>* according to the relation [37]

$$d\_n = \sqrt{\frac{4k\_B T}{N R\_0 A} + \frac{2qI\_{dark}}{N}},\tag{12}$$

where *N*—Number of periods and both dynamic resistance, dark current correspond to one QCIP's period. The optimal number of periods is related to the thickness of the single

absorber, d, and the absorption coefficient and could be expressed as *N* = (2αd)−<sup>1</sup> in the first order approximation.

**Figure 9.** IB QCIP based on T2SL InAs/GaSb active, GaSb/AlSb tunneling and InAs/AlSb relaxation layers (after reference [38]).

The IB QCIPs (with T2SLs absorbers) MBE growth process is challenging where many interfaces and strained thin layers are deposited in structures. Nevertheless, the significant progress has been reached for T2SL based detectors particularly for the LWIR and HOT conditions. They exhibit the capabilities of the IB optical transitions with the exceptional carrier transport properties of the QCIP architectures.

Currently, IB QCIPs has two types of configuration: current-matched (designed to have an equal photocurrent in every single stage) and non-current-matched [39]. Hinkey and Yang described the IB QCIP structure with equal absorbers offering the potential for significant responsivity improvement assuming αL ≤ 0.2 (αL—Product of the absorption coefficient and the diffusion length) [40]. From a technological point of view, the noncurrent-matched IB QCIPs (identical absorber thickness in all stages) are simpler to design and grow in comparison to the current-matched ones. The disadvantage of non-currentmatched structure is the limited responsivity due to the significant light suppression along the detector's structure. The high electrical gain, lately observed at HOT conditions in these structures, could at least partially compensate in terms of responsivity reduction [41,42].

Despite the development of other technologies, HgCdTe is still the most broadly used adjustable gap semiconductor for IR detectors, to include uncooled operation, and stands as a reference for alternative technologies. Figure 10 demonstrates that T2SL InAs/GaSb IB QCIPs bipolar devices (dashed lines) are proper candidate for HOT conditions. The assessed Johnson-noise limited detectivities under unbiased conditions for IB QCIPs with T2SL InAs/GaSb absorbers (based on the measured R0A product and responsivity) are comparable with commercially available HgCdTe devices. The performance of both types of detectors is comparable only in SWIR range and IB QCIPs outperform uncooled HgCdTe detectors with a similar LWIR cut-off wavelength. Another advantage of IB QCIPs is related to the III-V semiconductors strong covalent bonding allowing for operation at temperatures close to 400 ◦C being not possible for HgCdTe.

**Figure 10.** Room-temperature spectral detectivity curves of the commercially available photodetectors [PV Si and InGaAs, PC PbS and PbSe, HgCdTe photodiodes (solid lines reference [33])]. The spectral detectivity curves of new emerging T2SL IB QCIPs are marked by dashed lines (reference 38). Also the experimental data for different types of 2D material photodetectors are included. Experimental data is taken from [43–48]. PC—Photoconductor, PV—Photodiode.

In Figure 10, the representative experimental data for 2D material single photodetectors operating in IR spectral range are also marked. It can be seen that in MWIR the performance of black phosphorous-arsenic (bPAs) photodetectors outperforms commercially available uncooled HgCdTe photodiodes, while in LWIR, the detectivity, D\* of the transition metal dichalcogenide (TMD) photodetectors (PdSe2/MoS2 heterostructure) is the best. More detailed comments about these results are included in Section 5.2.

Figure 11 compares the peak detectivity, D\* for HgCdTe photodiodes [33] and InAs/GaSb T2Sls IB QCIPs [38] operating at 300 K with bPAs and TMD photodetectors. In MWIR the performance of bPAs devices is higher than commercially available uncooled HgCdTe photodiodes, while in LWIR, the detectivity, D\* of the TMD photodetectors is the best. The HgCdTe and IB QCIPs response time at 300 K, typically in the order of nanoseconds, is significantly shorter than for 2D material photodetectors.

Figure 12 gathers the highest detectivity, D\* values published in literature for different types of single element photodetectors operating at room temperature. This fact should be clearly emphasized since detectivity, D\* data marked for commercial photodetectors is typical for pixels of IR FPAs. Figure 12 also presents the fundamental indicator for future trend in development of HOT IR photodetectors. At present stage of HgCdTe technology, the semiempirical Rule 07 is found not to fulfil primary expectations. It is shown that the detectivity, D\* of low-doping P-i-N HgCdTe (5 × 1013 cm−3) photodiodes, operating at 300 K in spectral band above 3 μm, is limited by background radiation (with detectivity, D\* level above 10<sup>10</sup> Jones, not limited by detector itself) and can be improved more than one order of magnitude in comparison with predicted by Rule 07. Between different material systems used in fabrication of HOT LWIR photodetectors, only HgCdTe can fulfill required expectations: low doping concentration—10<sup>13</sup> cm−<sup>3</sup> and high SRH carrier lifetime above 1 ms. In this context, 2D material photodetectors and CQD photodetectors cannot compete with HgCdTe photodiodes. The above assessments provide further inspiration for reaching low-cost and high performance MWIR and LWIR HgCdTe FPAs operating in HOT conditions. The performance of T2SL IB QCIPs is close to HgCdTe photodiodes

and quantum cascade photodetectors can operate in temperature > 300 K; however, their disadvantage is a challenging technology and higher fabrication cost.

**Figure 11.** HgCdTe photodiodes, T2SLs InAs/GaSb IB QCIPs and representative 2D material photodetectors peak detectivity, D\* comparison for 300 K. The measured data for HgCdTe photodiodes according to the VIGO System catalogue [33]. Data for IB QCIPs extracted from selected papers [38]. Data for selected 2D materials is taken from [33,38,43,46–48].

**Figure 12.** Detectivity, D\* versus wavelength for the commercially available room-temperature IR photodetectors (PV Si and Ge, PV InGaAs, PC PbS and PbSe, PV HgCdTe). There is also included experimental data for IB QCIP T2SLs, different type of 2D material and CQD photodetectors taken from literature as marked. The theoretical curves are calculated for the P-i-N HOT HgCdTe devices assuming <sup>τ</sup>SRH = 1 ms, the absorber doping 5 <sup>×</sup> <sup>10</sup><sup>13</sup> cm−<sup>3</sup> and the active region thickness t = 5 <sup>μ</sup>m. Experimental data is taken from [13,33,34,38,43,46–51]. PC—Photoconductor, PV—Photodiode.

#### **5. 2D Material Infrared Photodetectors**

Graphene and other two-dimensional (2D) materials, due to uncommon electronic and optical properties, are considered to be promising candidates for IR photodetectors [52]. The further development of graphene-based photodetectors is a consequence of the high dark current the gapless material significantly limits the sensitivity. The discovery of new 2D materials with direct energy gaps in a wide spectral range (from the visible to the IR) has set a new direction for detector's design and fabrication. Although the technology readiness, the 2D materials are still at low level of development, the detectors' manufacturability and reproducibility have been challenging (this topic is widely studied in research laboratories around the globe).

Nicolosi et al. [53] distinguished the different types of 2D materials and refined them into different families (see Figure 13) covering a broad range of electrical and optical properties:


**Figure 13.** Energy bandgap of the selected layered semiconductors versus wavelength. The energy bandgap exhibits the dependence on the layers number, strain level and chemical doping. FIR— Far infrared; LWIR—Long wavelength infrared; MWIR—Mid wavelength infrared; SWIR—Short wavelength infrared; NIR—Near infrared; UV—Ultraviolet.

2D materials have their roots in layered van der Waals (vdW) solids. Atomic layers are built by in-plane atoms connected by ionic or tight covalent bonds along 2D directions. Each layer is bonded with another by a weak vdW interactions along out-of-plane direction. Such design causes that many of 2D materials could be mechanically exfoliated from bulk single crystals. What more, due to a weak bond between layers, a mixing of different 2D materials together is also possible with the flexibility of the heterostructures.

Energy band profiles of the layered materials differ from their bulk counterparts. Since the material gets thinner from the bulk to the monolayer, e.g., for TMDs the band structure transits from smaller indirect transition to a larger direct due to quantum confinement effects, thus, the bandgap (operating wavelength) can be adjusted by the layers number. Moreover, large strains occurring in these materials strongly affect their optical

and electronic properties. A high absorption coefficient of TMDs (typically 104–106 cm<sup>−</sup>1) results from dipole transitions between localized d-states and excitonic coupling of such transitions. Thanks to that, >95% of the sunlight is absorbed in sub-micrometer thickness TMD films, while carrier mobility is low (typically less than 250 cm2/Vs). Despite the fact that the mobility can be improved by increasing the number of TMDs layers, this disadvantage is difficult to circumvent. The carrier density depends on the doping levels and the number of recombination centers, and typically reaches 1012 cm−<sup>2</sup> [54].

In comparison to graphene, TMDs, like molybdenum disulfide (MoS2), tungsten disulfide (WS2) and molybdenum diselenide (MoSe2) is characterized by higher absorption in the visible and NIR ranges. As presented in Figure 13, the 2D bandgap profile is so different what allows to cover a very broad range from the UV to IR. For current status of technology, only graphene-based, black phosphorus-arsenic (bPAs), noble metal dichalcogenides, and bismuthene (like Bi2Te3 and Bi2Se3) are treated as a main player in IR and also THz regions. Since 2D TMDs are limited to UV-NIR, bP can be adjusted to below 0.3 eV by As doping. Due to high mobility, up to 3000 cm2/Vs, bP is a proper candidate for high sensitivity and fast speed photodetection. Recently published paper indicates that low bandgap 2D noble metal dichalcogenides could be novel platform for 300 K LWIR detectors [47].

From the practical applications point of view, the most important aspect is the stability of the material determining the reliability and lifetime of the device. This is a main disadvantage of most 2D materials. Due to only one or several-atoms thick of the active detector layer, 2D materials are susceptible to ambient environment (especially bP degrades quickly under 300 K conditions). The role of different ambient species has remained debatable [55]. The layered bP devices are still in development stage with many unsolved issues and ideas [56]. On the contrary, the air- stability properties have been demonstrated for noble metal dichalcogenides [47].

#### *5.1. 2D Material Photodetectors: Current Responsivity Versus Response Time*

The detector's current responsivity is given by equation

$$R\_i = \frac{\lambda \eta}{hc} q \text{g}\_{\prime} \tag{13}$$

and is determined by the QE (*η*) and photoelectrical gain, *g*. The QE is given by the number of electron-hole pairs generated per incident photon and shows how the detector is coupled to the impinging radiation. The second parameter, the photoelectrical gain describes the number of carriers reaching contacts per one generated pair and shows how well the generated carriers are used to increase photodetector current responsivity. Other symbols of Equation (13) mean: λ—Wavelength, *h*—Planck constant, *q*—Electron charge, and *c*—Light velocity.

In general, the photoelectrical gain is given by

$$\mathbf{g} = \frac{\mathbf{\tilde{r}}}{\mathbf{t}\_t \mathbf{\tilde{r}}}.\tag{14}$$

where *τ*—Carrier lifetime and *tt*—Transit time of electrons between the device electrodes. If the drift length, Ld = vdτ, is less or greater than the distance between electrodes, l, the photoelectrical gain can be less or greater than unity. The value of Ld > l shows that a one carrier swept out by electrode is replaced directly by an equivalent carrier injected by the opposite contact. In this way, a carrier will circulate until it recombines. For the photodiode, the photoelectric gain usually =1, due to separation of minority carriers by the electrical field of depletion region. However, in a hybrid combination of 2D material photodetectors, the photogeneration and carrier transport occur in a separate regions: one for effective light absorption, and the second - to provide fast charge reticulation. In this way, high gain close to 10<sup>8</sup> electrons per one photon, and significant responsivities for SWIR photodetectors have been demonstrated [51].

The simple architecture of hybrid phototransistor, very popular in the 2D material photodetectors design with the fast transfer channel for carriers, is presented in Figure 14. 2D materials with atomic layer thickness are more vulnerable to local electric fields than conventional bulk materials and the photogating effect can strongly modulate the channel conductivity by external gate voltage, Vg. Improvement in the optical gain is particularly important since the QE is suppressed because of the weak absorption in 2D materials. This effect is especially seen in LWIR region, where the light absorption is weak. In the case of hybrid detector shown in Figure 14a, the holes are injected into transporting channel, whereas the electrons remain in the photoactive layer. The injected charges can reticulate even several thousand times before recombination, giving contribution to the gain under illumination. The photocarrier lifetime is enhanced through both the bandgap profile and defect engineering, and at the same time the trapping mechanisms limit the response time of photodetector even to several seconds. The trade-off between improvement in responsivity and response time must be considered during optimization process.

**Figure 14.** Photogating effect in 2D material photodetectors: (**a**) the operation of hybrid phototransistor, (**b**) closed channel under illumination, (**c**) photoconductive gain, and (**d**) I-Vg trace under illumination.

The photocurrent versus photogating effect can be given by [57]

$$I\_{\rm pl} = \mathcal{g}\_{\rm m} \Delta V\_{\mathcal{S}'} \tag{15}$$

where *gm*—Transconductance, Δ*Vg*—Equivalent photoinduced bias. Figure 14d indicates a shift of the *Ids*(*Vg*) trace after the light illumination. Generally, both positive and negative photoconductance effects are observed in hybrid 2D structures and operating points A and B, related to *gm* and Δ*Vg*, perform opposite directions.

Figure 15 compares the graphene-based detectors responsivity operating in visible and NIR with silicon and InGaAs photodiodes commercially available on the market [58,59]. The highest current responsivity, above 10<sup>7</sup> A/W, has been reached for hybrid Gr/quantum dot (QD) photodetectors with enhancement trapped charge lifetimes. As shown, the graphene high mobility along with the extension of the charges lifetime trapped in QDs caused a photodetector responsivity up to seven orders of magnitude higher in relation to the standard bulk photodiodes, where g ≈ 1. Higher responsivity of Si avalanche photodiode (APD), up to 100 A/W, is caused by avalanche process. However, due to the

long lifetime of trapped carriers, the response time of 2D material photodetectors is very slow (<10 Hz), what considerably limits real detector functions.

**Figure 15.** The graphene-based photodetectors spectral responsivity compared with commercial detectors. Black line presents spectral responsivity for ideal photodiode with 100% QE and g = 1. Red and green colors correspond to ≤1 ns, while the blue color ≥1 second response times. The graphene detectors are labelled with proper reference and brief description. The commercial photodiodes are marked in green (adapted after reference [58,59]).

It is interesting to underline the unusual electrical and optical properties of gold patched graphene nanostripe detectors presented by Cakmakyapan et al. [60]. The photodetector exhibits a spectral response in the ultrabroad range from visible to the IR with high responsivity ranging from 0.6 A/W (for wavelength 800 nm) to 11.65 A/W (for 20 μm) and frequency exceeding 50 GHz. As is shown in Figure 15, its current responsivity (black circles) coincides well with curve (black line) theoretically predicted for ideal photodiode in NIR spectral range.

2D materials show potential for operation in wide spectral range from UV to THz, although majority of them cover visible and SWIR (see Figure 16). Similarly, for graphene photodetectors, both high responsivity and short response time cannot be reached simultaneously in many 2D material-based photodetectors.

**Figure 16.** The layered 2D material photodetectors spectral responsivities at 300 K (after reference [43,61]). Black line shows spectral responsivity for ideal photodiode with 100% QE and g = 1. The responsivities of commercially available photodetectors (InGaAs and HgCdTe photodiodes) are presented for comparison reasons.

The two major factors determine the development of the 2D material high sensitivity photodetectors. It is a short carrier lifetime and low absorption in a thin active region (~100–200 nm). In consequence, the broadband operation sets a trade-off between high responsivity and response time. The 2D materials-based detectors display a large variation in their responsivity and response time [62–64] about 9 orders of magnitudes as is shown in Figure 17. In order to improve the IR absorption, the multiple layers instead of the single layer are chosen. In photogating effect photodetectors, 2D materials are used as the fast transfer channel for carriers. However, as is mentioned above, their overall disadvantage is the very slow response time attributed to traps and high capacitance. The response time is typically longer than ~1 × <sup>10</sup>−<sup>2</sup> ms, what indicates on considerably longer response time in comparison with commercial silicon, InGaAs, and HgCdTe photodiodes, while for HOT LWIR photodiodes is typically tens of nanoseconds. The up-left blank panel on Figure 17 shows the lack of photodetectors with both high responsivity and short response time. Figure 17 summarizes the responsivity and response time of different 2D material photodetectors. It is shown that black phosphorus is a good candidate for fast detection and falls into a region between graphene and TMDs.

**Figure 17.** Current responsivity versus response time for HOT 2D material photodetectors in relation to the commercial silicon, InGaAs and HgCdTe detectors (experimental data taken from reference [63]).

#### *5.2. Detectivity: HgCdTe Photodiode Versus 2D Material Photodetectors*

Figure 10 presents detectivity, D\* curves gathered from literature for HOT MWIR and LWIR photodetectors both for commercially available devices (PV Si and InGaAs, PC PbS and PbSe, HgCdTe photodiodes) and IB QCIP T2SLs, as well as for 2D material photodetectors. All experimental data gathered in Figure. 10 indicates on sub-BLIP photodetectors performance. As is shown, the literature data for 2D material photodetectors in LWIR above 3 μm is limited to several device structures. The Gr/FGr detector utmost detectivity in MWIR is comparable to HgCdTe. However, especially high detectivity (higher than for HgCdTe photodiodes) is demonstrated for black phosphorus arsenic (bPAs) detectors [46]. Their sensitivity enters the second atmospheric transmission window. Here it must be stressed, that the serious drawback of black phosphorus is surface instability in ambient conditions what can considerably limit its prospective applications [55,65]. More promising is stable TMD photodetectors like PdSe2/MoS2 heterojunction with record detectivity in LWIR range at room temperature. However, their practical application lies in perfect material synthesis and processing being still under development.

Figure 12 compares the experimental detectivity, D\* published in literature for different types of single element 2D material photodetectors operating at room temperature with theoretically predicted curves for P-i-N HOT HgCdTe detectors. As is presented, the detectivity values for selected 2D material photodetectors are close to data presented for commercial detectors (PV Si and Ge, PV InGaAs, PC PbS and PbSe, PV HgCdTe), and in the case of black phosphorus and TMD detectors are even higher. The enhanced sensitivity of 2D material photodetectors is introduced by bandgap engineering and photogating effect, what degrades the electronic material properties. In consequence, the layered-material photodetectors are characterized by limited linear dynamic range of operation and slow response time.

To summarize the discussion in this section about 2D material IR HOT photodetectors we can conclude that:


#### **6. Colloidal Quantum Dot Infrared Photodetectors**

Research on QD IR photodetectors based on self-assembled epitaxial QDs started in the mid-1990 and were initially very promising. Theoretical estimates carried out by Martyniuk et al. [67] in 2008 indicate that the self-assembled quantum dot infrared photodetectors (QDIPs) are suitable for noncryogenic operation especially in LWIR region. As it happens later, that epitaxial QDs suffer from the size control and low dots density. More recently, an attractive alternative to self-assembled epitaxial QDs has been colloidal

quantum dots (CQDs) with better size tunability of optical features and reduction of fabrication cost.

#### *6.1. Brief View*

In the last decade, a significant progress in fabrication of CQD photodetectors has been observed. In this approach, an active region is constructed based on 3D quantum confined nanoparticles synthesized by inorganic chemistry. CQDs offer a promising alternative to the single crystal IR materials (InGaAs, InSb, InAsSb, HgCDTe, as well as T2SLs see Figure 18). These nanoparticles could improve CQD photodetectors performance compared to epitaxial QDs due to many aspects gathered in Table 2 [50,68].

**Figure 18.** The wavelengths range that can be detected by materials commonly used in imaging applications.

**Table 2.** CQD photodetectors advantages and disadvantages in comparison with single crystal QD photodetectors.


CQD photodetectors are typically fabricated using conducting-polymer/nanocrystal blends, or nanocomposites [13,50,69,70]. Nanocomposites often feature narrow-bandgap, II-VI (HgTe, HgSe) [71,72], PbSe or PbS [73–75]. Usually, the reported IR photodetectors use CQDs embedded in conducting polymer matrices, such as poly [2-methoxy-5-(2 ethylhexyloxy)-1,4-phenylenevinylene] (MEH-PPV).

It is expected that the extension of application of CQD-based devices will be significant, especially in IR imaging which is currently dominated by epitaxial semiconductor and hybrid technologies [76]. Hybrid technology, due to the complexity of production

stages, reduces yield and increases overall cost. The IR CQD-based photodetectors are an alternative solution without these limitations.

The CQD layers are amorphous what permits fabrication of devices directly onto ROIC substrates, as shown in Figure 19 with no restrictions on pixel or array size and with a day cycle of production. In addition, the monolithic integration of CQD detectors into ROIC does not require any hybridization steps. Individual pixels are defined by the area of the metal pads arranged on the top of ROIC surface. To synthesize colloidal nanocrystals, wet chemistry techniques are used. Reagents are injected into a flask and, the desirable shape and size of nanocrystals are obtained by the control of reagent concentrations, ligand selection, and temperature. This so-called top-surface photodetector offers a 100% fill factor and is compatible with postprocessing at the top of complementary metal-oxide semiconductor (CMOS) electronics.

**Figure 19.** IR monolithic array structure based on CQDs.

The lead chalcogenides CQDs (primarily PbS) are the materials for SWIR photodetectors with detection to 2 μm. The peak can be adjusted using smaller dots by adding NIR bands to hyperspectral visible image sensors or using larger dots to include the InGaAs spectrum of image sensors [75]. From a performance standpoint, SWIR photodetectors based on PbS CQDs have reached detectivity, D\* comparable to commercial InGaAs photodiodes, with a values of >10<sup>12</sup> Jones at 300 K. HgTe CQDs have opened the MWIR spectral range. Detectivity, D\* between 10<sup>10</sup> to 10<sup>9</sup> Jones at 5-μm was demonstrated for HgTe CQD devices while maintaining a fast response time at thermoelectric cooling temperatures. It is unlikely that CQD IR detectors will ever reach the performance of currently popular InGaAs, HgCdTe, InSb and T2SL photodiodes.

Recent demonstrations of low-cost SWIR and MWIR CQD imaging arrays have heightened the interest in these devices. For both PbS and HgTe CQD photodetectors integration in camera imaging have been demonstrated [76]. It is expected that the successful implementation of this new class of IR technology may match the broad impact of cheap CMOS cameras that are widely used today. First SWIR cameras built on CQD thin film photodiodes fabricated monolithically on silicon ROICs have been launched [77,78]. The Acuros camera has resolution 1920 × 1080 (2.1 megapixels, 15-μm pixel pitch) and uses 0.4 to 1.7 μm broadband spectral response [77]. The IMEC's prototype imager has resolution of 758 × 512 and 5 μm pixel pitch. The CQD photodiodes on silicon substrate reach an external QE above 60% at 940 nm wavelength, and above 20% at 1450 nm, allowing uncooled operation with dark current comparable to commercial InGaAs photodetectors [78].

At present, CQD cameras are used in newer applications that require high-definition low cost imaging on smaller pixels without extreme sensitivity. It can be predicted that increasing the dot size while maintaining a good mono-dispersion, carrier transport and QE will improve maintaining low noise levels. Due to continuous development of deposition and synthesis techniques, much higher performance will be reached in the future.

#### *6.2. Present Status of CQD Photodiodes*

Figure 7b compares the detectivity, D\* temperature dependence versus cut-off wavelength ~5 μm for different material systems including commercially available HgCdTe and

HgTe CQD photodiodes. The gathered experimental data are also included. The estimated detectivity, D\* for CQD photodiodes are located below those for HgCdTe photodiodes. As is shown, at temperature above 200 K the theoretically predicted detectivity for HgCdTe photodiodes is limited by background. Rule 07 coincides well with theoretically predicted curve for Auger-suppressed p-on-n HgCdTe photodiode with doping concentration in active region 10<sup>15</sup> cm<sup>−</sup>3. As is marked in Section 3.3, at present stage of HgCdTe technology the doping concentration is almost two orders of magnitude lower (mid 1013 cm<sup>−</sup>3).

All experimental data gathered in Figures 12 and 20 indicates on sub-BLIP photodetectors performance. Both figures also clearly show that the detectivity values of CQD photodetectors are inferior in comparison with HgCdTe photodiodes and are generally worse also in comparison with 2D material photodetectors. Moreover, the theoretical predictions indicate on possible further HgCdTe devices performance improvement after decreasing of i-doping level in P-i-N photodiodes. For doping level of 5 × <sup>10</sup><sup>13</sup> cm−<sup>3</sup> the photodiode performance can be limited by background radiation in spectral band above 3 μm. It is shown that in this spectral region, the detectivity, D\* is not limited by detector itself, but by background photon noise at a level above 10<sup>10</sup> Jones in LWIR range (above one order of magnitude above Rule 07).

**Figure 20.** Room-temperature spectral detectivity curves of the commercially available photodetectors [PV Si and InGaAs, PC PbS and PbSe, HgCdTe photodiodes (solid lines reference [33])]. The experimental data for different types of CQD photodetectors are marked by dot points (reference [49,70,78–81]). Also, spectral detectivity of new emerging T2SL IB QCIPs are included [38]. PC—Photoconductor, PV—Photodiode.

#### **7. Conclusions**

In the last decade considerable progress in fabrication of SWIR and MWIR 2D material and CQD photodetectors has been demonstrated together with their integration into thermal imaging cameras. At current status of technology, the performance of both types of photodetectors is inferior in comparison with HgCdTe photodiodes. It seems that only PbS CQD photodetectors characterized by multicolor sensitivity and detectivity comparable to InGaAs detectors (which are currently the most common in commercial applications) have been located at the good position in IR material family at present time.

Discovery of graphene in 2004 gave a new impetus on technology development and investigations of 2D layered materials where their uncommon electronic and optical properties make them promising candidates for IR photodetectors. Despite spectacular demonstration of high detectivity like this achieved for black phosphorus layered photodetectors in MWIR spectral range [46] and noble TMD photodetectors like PdSe2/MoS2 heterojunction with record detectivity in LWIR range at room temperature [47], many challenges remain to be introduced to exploit the distinct advantages of these new materials. The prospect of commercialization of 2D material photodetectors depends on their large-scale integration with existing photonic and electronic platforms like CMOS technologies, high operability, spatial uniformity temporal stability, and affordability. Industry fabrication of devices is in the early stage of development and manufacturability.

In general, pristine narrow gap 2D materials are characterized by weak optical absorption and short carrier lifetime. Various ingenious approaches (electron trap layers, photogating effect with the graphene fast transfer channel) enhance sensitivity, however on the other side, degrade the electronic performance including carrier mobility. In this way high 2D material photodetector sensitivity collides with slow response time what seriously limits their practical applications.

In spite of sixty years development history of HgCdTe, it ultimate HOT performance limit has not been achieved. In order to achieve this goal, the doping concentration below <sup>5</sup> × 1013 cm−<sup>3</sup> is required. This level of doping concentration has been recently achieved in fully-depleted HgCdTe FPAs by Teledyne Technologies.

At present stage of HgCdTe technology, the semiempirical rule Rule 07 (specified in 2007), widely popular in IR community as a reference for other technologies, was found not to fulfil primary expectations. In this paper, it was shown that the potential properties of HOT HgCdTe photodiodes operating above 3 μm guarantees achieving more than order of magnitude higher detectivity (above 1010 Jones) in comparison with value predicted by Rule 07, and this detectivity is limited by background. In this context it is rather difficult to compete 2D material and CQD photodetectors with HgCdTe photodiodes.

**Author Contributions:** Conceptualization, A.R., P.M.; writing—Original draft preparation, A.R., P.M., M.K. and W.H.; writing—Review and editing, A.R., P.M., M.K. and W.H.; visualization, A.R., P.M., M.K. and W.H.; supervision, A.R. and P.M.; project administration, A.R., P.M. and M.K.; funding acquisition, A.R., P.M. and M.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Polish National Science Centre:, grant number OPUS-UMO-018/31/B/ST7/01541, HARMONIA-UMO-2018/30/M/ST7/00174, OPUS-UMO-2017/27/B/ ST7/01507 and UMO-2019/33/B/ST7/00614.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Abbreviations**



#### **References**


### *Review* **Low-Light Photodetectors for Fluorescence Microscopy**

**Hiroaki Yokota 1,\*, Atsuhito Fukasawa 2, Minako Hirano <sup>1</sup> and Toru Ide <sup>3</sup>**


**Abstract:** Over the years, fluorescence microscopy has evolved and has become a necessary element of life science studies. Microscopy has elucidated biological processes in live cells and organisms, and also enabled tracking of biomolecules in real time. Development of highly sensitive photodetectors and light sources, in addition to the evolution of various illumination methods and fluorophores, has helped microscopy acquire single-molecule fluorescence sensitivity, enabling single-molecule fluorescence imaging and detection. Low-light photodetectors used in microscopy are classified into two categories: point photodetectors and wide-field photodetectors. Although point photodetectors, notably photomultiplier tubes (PMTs), have been commonly used in laser scanning microscopy (LSM) with a confocal illumination setup, wide-field photodetectors, such as electron-multiplying charge-coupled devices (EMCCDs) and scientific complementary metal-oxide-semiconductor (sC-MOS) cameras have been used in fluorescence imaging. This review focuses on the former low-light point photodetectors and presents their fluorescence microscopy applications and recent progress. These photodetectors include conventional PMTs, single photon avalanche diodes (SPADs), hybrid photodetectors (HPDs), in addition to newly emerging photodetectors, such as silicon photomultipliers (SiPMs) (also known as multi-pixel photon counters (MPPCs)) and superconducting nanowire single photon detectors (SSPDs). In particular, this review shows distinctive features of HPD and application of HPD to wide-field single-molecule fluorescence detection.

**Keywords:** low-light photodetectors; fluorescence microscopy; time-resolved fluorescence microscopy; hybrid photodetector (HPD); single-molecule fluorescence detection

#### **1. Introduction**

The development of the light microscope has enabled investigation of the fine structures of biological specimens under magnification. In recent decades, fluorescence microscopy—a form of highly sensitive optical microscopy—has evolved and is a necessary element of life science studies [1–3]. Microscopy has elucidated biological processes in vitro and in vivo, and also enabled tracking of biomolecules in real time. Development of highly sensitive photodetectors and light sources, in addition to evolution of various illumination methods and fluorophores, has helped microscopy acquire single-molecule fluorescence sensitivity, enabling single-molecule fluorescence imaging and detection [4–7]. Single-molecule fluorescence microscopy led to the emergence of super-resolution microscopy [8–10].

Low-light photodetectors used in fluorescence microscopy are classified into two categories: point photodetectors and wide-field photodetectors [5,11]. Point photodetectors, notably photomultiplier tubes (PMTs), are most commonly used in laser scanning microscopy (LSM). The detectors are also used in point-like excitation and detection to study freely diffusing biomolecules, such as protein molecules and nucleic acids (DNA and RNA) in solution [11]. Wide-field photodetectors, such as electron-multiplying charge-coupled

**Citation:** Yokota, H.; Fukasawa, A.; Hirano, M.; Ide, T. Low-Light Photodetectors for Fluorescence Microscopy. *Appl. Sci.* **2021**, *11*, 2773. https://doi.org/10.3390/ app11062773

Academic Editor: Jacek Wojtas

Received: 19 February 2021 Accepted: 17 March 2021 Published: 19 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

devices (EMCCDs) and scientific complementary metal-oxide-semiconductor (sCMOS) cameras, are used in wide-field illumination and detection to study surface-immobilized or slowly-diffusing biomolecules and organelles. These biomolecules and organelles include protein molecules, nucleic acids, and lipids, and nucleus and mitochondria, respectively. The two photodetectors are distinct in many aspects. Point photodetectors have high temporal resolution (high sampling frequency) but have no spatial resolution without scanning. Wide-field photodetectors, in contrast, have spatial resolution (typically submicrometer precision) but are limited to relatively low frame rates. This review focuses on low-light point photodetectors and presents their fluorescence microscopy applications and recent progress.

In fluorescence microscopy, the fluorescence emission can be characterized not only by intensity and position but also by lifetime [12]. Fluorescence microscopy uses two fluorescence detection methods: steady-state and time-resolved fluorescence detection. Time-resolved measurements contain more information than is available from steady-state measurements. Fluorescence lifetime measurement by time-resolved detection provides data that is independent of fluorophore concentration and allows us to obtain information on the local ambient environment around the fluorophore, such as pH, ion concentrations, temperature, and fluorescence resonance energy transfer (FRET) efficiency [13]. The laser scanning fluorescence microscope, an indispensable imaging device in the biological sciences, is one of the most widely used fluorescence microscopy. LSM with a confocal illumination setup provides a base for various fluorescence microscopes using steady-state and time-resolved fluorescence detection, such as two-photon microscopy and fluorescence lifetime imaging microscopy (FLIM). This review introduces these point detectors, describes their operating principles, and compares their specifications. These photodetectors include conventional PMTs, single photon avalanche diodes (SPADs), hybrid photodetectors (HPDs), in addition to newly emerging multi-pixel photon counters (MPPCs) (also known as silicon photomultipliers (SiPMs)) and superconducting nanowire single photon detectors (SSPDs). In particular, this review shows distinctive features of HPD, and notes the applications of HPD to wide-field single-molecule fluorescence detection and the development of multi-pixel photodetectors.

#### **2. Fluorescence Microscopy**

Fluorescence measurements are characterized by their high sensitivity, up to singlemolecule detection. Because biological samples commonly exhibit low contrast, fluorescence microscopy makes good use of fluorescence phenomenon to enhance the contrast. Fluorescence microscopy acquires data of target biological samples through fluorescence emissions characterized not only by intensity and position, but also by lifetime. Fluorescence microscopy uses the two fluorescence detection methods: steady-state and timeresolved fluorescence detection. LSM with a confocal illumination setup provides a base for various fluorescence microscopes using steady-state and time-resolved fluorescence detection, such as two-photon microscopy and FLIM.

#### *2.1. Fluorescence*

Fluorescence is a photophysical phenomenon of the emission of light through the excitation of a fluorophore from the ground state to an excited electronic state upon the absorption of light, with the light energy equivalent to the energy gap between the two states [10,12]. Figure 1 shows a Jablonski diagram that illustrates the electronic energy levels of a fluorophore and the transitions between them are represented by arrows. S0, S1, and S2 represent the singlet ground, and first and second excited electronic states, respectively. The vibrational ground states and higher vibrational states of each electronic state are illustrated with black and gray lines, respectively. The transition from the ground state to the excited state by the light absorption occurs in less than 10−<sup>15</sup> s. A fluorophore is usually excited to some higher vibrational level of the excited state. The electron usually rapidly relaxes to the lowest vibrational level of S1. This process is called internal conversion and

generally occurs within 10–12 s. Then, the excited state is relaxed to the ground state in a few nanoseconds (10–9 s), which is accompanied by the radiation of the fluorescence emission. Because internal conversion is generally complete prior to emission, the last relaxation step to the ground state accounts for most of the overall process. Thus, the relaxation time is called the fluorescence lifetime. The wavelength of the fluorescence is longer than the excitation wavelength because the energy from the absorbed photon is partially lost via non-radiative decay. This shift in wavelength is called the Stokes shift.

An electron in the S1 state can also flip its spin thus creating the first triplet state T1, which is termed intersystem crossing. Transition from T1 to S0 is forbidden, thus the rate constants for triplet emission (phosphorescence) are several orders of magnitude smaller than those for fluorescence, which results in a long-lived dark state called blinking. While in the T1 state, the fluorophore may experience photobleaching, which is an irreversible fluorescence switching-off process.

Table 1 shows characteristics of several fluorophores commonly used in fluorescence microscopy, including dyes, a quantum dot (Qdot) [14–16], and a fluorescent protein [17,18]. Photobleaching can be a representative photostability indicator of a fluorophore and the higher resistance to photobleaching allows longer or brighter fluorescence observation.


**Table 1.** Characteristics of fluorophores commonly used in fluorescence microscopy.

The markings +, ++, and +++ indicate "poor," "moderate," and "excellent," respectively. Adapted with permission from [7].

#### *2.2. Laser Scanning Fluorescence Microscopy (LSM)*

Laser scanning fluorescence microscopy (LSM) is one of the most widely used forms of biological fluorescence microscopy. LSM creates fluorescence images by sequentially recording the fluorescence intensity of each pixel by scanning a focused laser beam across a specimen using confocal optics to obtain only in-focus fluorescence. LSM with a confocal illumination setup provides a base for various fluorescence microscopes using steady-

state and time-resolved fluorescence microscopy. Figure 2 shows schematics of confocal microscopy and fluorescence microscopy with confocal optics.

**Figure 2.** Schematics of confocal microscopy and fluorescence microscopy with confocal optics: (**a**) confocal microscopy; (**b**) two-photon microscopy; (**c**) fluorescence correlation microscopy (FCS); (**d**) fluorescence lifetime imaging microscopy (FLIM).

#### *2.3. Confocal Microscopy*

Confocal microscopy limits the observed volume to reduce out-of-focus signals. Minsky introduced the concept of confocal microscopy [19] and issued an original patent for a microscope in 1961. In confocal microscopy, two pinholes that are conjugated in the identical image plane are placed at a focal point in the light path. Light from outside of the focal plane is not focused on the pinhole(s) and only fluorescence very near to the sample's focal point reaches the detector (Figure 2a). Thus, the confocal laser scanning microscopy enables three dimensional reconstruction of specimens.

#### *2.4. Two-Photon Micriscopy*

Two-photon microscopy, which was first reported by the Watt W. Webb group in 1990 [20], makes use of the phenomenon that two photons are absorbed by a fluorophore simultaneously. The fluorophore can be excited by light with one-half the energy of each photon or twice the wavelength. The two-photon excitation light is generated by increasing the photon density using a focused high-power femtosecond pulse laser (Figure 2b). Fluorescence emitted from the focus point is detected by a point detector (commonly a PMT) and a fluorescence image is acquired by LSM. Two-photon microscopy enables deep imaging because the microscopy uses near-infrared laser excitation light that exhibits better tissue penetration and collects the localized fluorescence signal.

#### *2.5. Fluorescence Correlation Spectroscopy (FCS)*

Fluorescence correlation spectroscopy (FCS) is also based on confocal optics with a continuous wave laser(s) and monitors the mobility of molecules, typically translation diffusion into and out of a small volume (Figure 2c) [12]. FCS analyzes time-dependent fluorescence intensity fluctuations in a tiny observed volume on the order of femtoliter.

When a fluorophore diffuses (or fluorophores diffuse) into the illuminated volume, a fluorescence burst is detected due to steady-state fluorescence emission from the fluorophore(s). If the fluorophore diffuses (or fluorophores diffuse) quickly out of the volume, the burst duration is short, whereas if the fluorophore diffuses (or fluorophores diffuse) more slowly the photon burst duration persists for longer. Correlation analysis of the time series enables the diffusion coefficient of the fluorophore to be determined.

The fluorescence intensity fluctuation depends on the size and number of the molecules passing through the illuminated volume, which provides information on biomolecular interaction in vitro and in vivo.

#### *2.6. Fluorescence Lifetime Imaging Microscopy (FLIM)*

Fluorescence lifetime imaging microscopy (FLIM) is an advanced tool that maps the fluorescence lifetime distribution through time-resolved fluorescence detection (Figure 2d) [21,22]. The fluorescence lifetime can respond to changes in pH, temperature, and ion concentrations such as calcium concentration. Its capability to offer both localization of target fluorophores and the fluorophores' local microenvironment exhibits its superiority to fluorescence intensity based steady-state imaging because the lifetime of a fluorophore is mostly independent of its concentration. FLIM can be performed using the time-domain method in which the sample is excited with a pulse laser, or the frequencydomain method in which the sample is excited with intensity-modulated light, commonly sine-wave modulation [12].

#### **3. Single Point Detectors**

#### *3.1. Performance Indices*

#### 3.1.1. Dark Count

Dark count refers to the tiny flow of electricity in a photodetector operated under a totally dark condition. This dark current should be minimized. Dark count is caused by several phenomena that vary with photodetectors. The dark count of PMTs results from thermionic emission from the photocathode and dynode, and ionization of residual gases (ion feedback) [23]. The dark count of SPADs and SiPMs results from thermionic emission from the depletion layer. The dark count of HPDs is negligible due to its high electron bombardment gain. The dark count of SSPDs, which is low and dependent on the cooling temperature and bias current [24], results from energy dissipation in the nanowire and the blackbody radiation at room temperature through the optical fiber [25].

#### 3.1.2. Instrument Response Function

The transit time and its fluctuation are the major determinants of the time response of a photodetector [23]. The transit time refers to the travel time of the photoelectron. The full-width half-maximum (FWHM) of the instrument response function is a standard for the time response. Many detectors exhibit a non-Gaussian instrument response function. The rise and fall times of a detector are evaluated from the waveform. Transit time spread (TTS) is the fluctuation of the transit time of the single photoelectron pulse.

#### 3.1.3. Afterpulse

Afterpulses are spurious pulses that may appear subsequent to the input signal [23]. An afterpulse is best characterized by the autocorrelation function of the detected photons. For vacuum tube-based photodetectors, positive ions generated by the ionization of residual gases in a detector create afterpulses that appear several hundreds of nanoseconds to several microseconds later than the input signal. This phenomenon is called ion feedback [23]. Among solid-state photodetectors, SPADs and SiPMs experience high afterpulse noise with high count rate measurements. The afterpulse noise is generated by thermally released trapped carriers [26]. HPDs and SSPDs are free of afterpulses.

#### 3.1.4. Sensitivity

The sensitivity of a photodetector is largely determined by the quantum efficiency of the photocathode material. Most photocathodes used in vacuum tube-based photodetectors are made of compound semiconductors. GaAsP exhibits the highest quantum efficiency in the visible region (about 45%). A photocathode (extended red GaAsP) that is more sensitive to a longer wavelength than GaAsP is available for Hamamatsu Photonic products. Silicon and group III-V compound semiconductors, such as InGaAs and GaAs, are also used in solid-state photodetectors. Figure 3 shows the spectral response of these photocathodes.

**Figure 3.** Spectral response of photocathodes.

These photocathodes are usually used in a vacuum and under the ambient temperature ranging from room temperature down to 273 K. As the ambient temperature is lowered, the spectra slightly shift to shorter wavelength due to the bandgap broadening of the semiconductors. Thus, the spectral response of these photocathodes does not change significantly in the temperature range.

#### 3.1.5. Active Area

The diameter or surface of the active area of a photodetector determines the observed area. Confocal optics uses a pinhole in a plane conjugate with the image plane in the sample—the light from the pinhole is easy to focus on a small point detector, such as a SPAD [13]. Point photodetectors with a larger active area are often useful because the light does not need to be focused and they can used to observe a large area without scanning.

#### *3.2. Photomultiplier Tube (PMT)*

Photomultiplier tubes (PMTs) are the most widely used point detectors in fluorescence microscopy. A PMT is a vacuum tube that contains a photocathode, focusing electrodes, an electron multiplier (dynodes), and an anode (Figure 4) [23]. Incident photons are absorbed by the photocathode, which ejects primary electrons (~3 eV). The electrons are accelerated by a high voltage to hit a series of dynodes. Then, additional electrons (5–10 electrons) are ejected and exponentially amplified. The electron current is then detected by an external electrical circuit. Typical PMTs have 8–10 dynodes with a cathode-to-anode voltage gap of ~1 kV and current gain of 106 to 107.

**Figure 4.** Schematic of a photomultiplier tube (PMT).

The main photocathodes used in PMTs are bi-alkali, with spectral response peaks around 400 nm and range of up to 700 nm. The advent of the GaAsP photocathode with its spectral response peak around 500 nm invigorated the LSM market because its spectral response ranges from visible light (500–700 nm) to near infrared light (900 nm). Laser scanning microscopes with a GaAsP photocathode were released to the market in the late 2000s. Current laser scanning microscopes incorporate PMTs with these multialkali photocathodes.

PMTs have been the leading low-light point photodetector for some time and, thus, a large number of power supplies and signal processing circuits for them are available. However, troubles can arise from unwanted noise generated by residual gas molecules (ion feedback), and large variation in responsivity and gain caused by difficulty in controlling metal evaporation to produce photocathodes and secondary electron surfaces. In addition, single photons can be detected with PMTs, but discrimination of single versus multiple photons is difficult.

#### *3.3. Single Photon Avalanche Diode (SPAD)*

The single photon avalanche diode (SPAD) is a solid-state photodetector composed of three semiconductor layers, called p-layer, i-layer, and n-layer (Figure 5) [27]. The n-layer has extra electrons, whereas the p-layer has holes. The average gain for an avalanche photodiode (APD) is around 100, which is insufficient for single-photon detection. Therefore, SPADs are usually operated in "Geiger-mode," where an applied bias voltage is greater than the diode's breakdown voltage. Then, when a charge is generated by an incident photon, the charge multiplication (or avalanche) occurs until it saturates corresponding to a current typically specified by the components. These APDs with a single pixel are referred to as SPADs, and those with multiple pixels are referred to as silicon photomultipliers (SiPMs) or multi-pixel photon counters (MPPCs). Although Geiger-mode driven SPADs are suitable for single photon counting, SPADs suffer several drawbacks. The active area cannot be increased because fabrication of a large semiconductor surface increases the number of defects. Geiger-mode driven SPADs have high dark counts due to the Geiger discharge and high afterpulse noise. The dead time is relatively long (about 50 ns), during which photon counting is inoperative because the mode needs to be reset for every single photon detection. To reduce dark counts, SPADs are typically cooled to 210–250 K.

**Figure 5.** Schematic of a single photon avalanche diode (SPAD).

#### *3.4. Hybrid Photodetector (HPD)*

The hybrid photodetector (HPD) is a hybrid of an avalanche diode (AD) and a photocathode, both of which are in a vacuum tube (Figure 6) [23,26,28–31]. When light is incident onto the photocathode, photoelectrons are emitted from the photocathode. The photoelectrons are then accelerated by a high negative voltage to directly bombard the AD where electron-hole pairs are generated and the signal is amplified. The amplification is termed "electron bombardment gain".

**Figure 6.** Schematic of a hybrid photodetector (HPD).

In the case of an HPD manufactured by Hamamatsu Photonics, the electron bombardment gain is approximately 1500 with the photocathode supply voltage of −8 kV. The signals of electron-hole pairs are further amplified to 80-fold (avalanche gain) by applying a reverse voltage of about 400 V to the avalanche diode. Then, the total gain will therefore be as much as about 120,000.

#### 3.4.1. Features of HPD

HPDs have significant advantages over PMTs and other low-light photodetectors in the detection of fluorescence, as discussed below. However, a few disadvantages of HPDs also exist. The extremely high cathode supply voltage (−8 kV) is difficult to deal with, which can be problematic when incorporated into systems.

Low Afterpulse

HPDs have a notable feature of lower afterpulse due to their uncomplicated internal structure. Therefore, the major cause of afterpulses, i.e., ion feedback, is highly unlikely to occur in HPDs. Afterpulses evaluated for an HPD and a PMT are shown in Figure 7. This graph shows the probability at which afterpulses may be generated by a single photoelectron input. In contrast to the PMT's multiple afterpulses in a time range from 100 ns to 1 μs, this HPD exhibited only a small number of afterpulses in the time range.

**Figure 7.** Afterpulse noise of PMT and HPD [32]. The data were obtained through detection of single photons from a continuous wave (CW) light source (wavelength = 470 nm). The cathode supply voltage and the reverse voltage of HPD were −8 kV and 398 V, respectively.

Comparable low afterpulses have been reportedly achieved solely for superconducting nanowire single photon detectors (SSPDs). These SSPDs have micrometer order active areas and must be operated at a liquid helium temperature [25], as discussed in Section 4.2.

#### High Resolution of Photon Counting

HPDs exhibit better pulse height resolution than PMTs. Gain fluctuation of HPDs is significantly lower owing to much higher electron bombardment gain (about 1500 at a photocathode supply voltage of −8 kV) than the first dynode gain of an ordinary photomultiplier tube (typically as low as 5–10). The first gain mostly determines the signal-to-noise ratio of the electron multiplication, which in turn represents the detector's capability to distinguish between one and multiple photons. As a result, HPDs offer high resolution of photon counting. As shown in Figure 8, signal peaks that correspond to 1, 2, 3, 4, and 5 photoelectrons can be identified in the output pulse height distribution.

#### High Timing Resolution

Time response characteristics of HPDs are largely determined by the junction capacitance of the internal avalanche diode, provided the diameter of the internal avalanche diode is around or larger than 1 mm. The internal avalanche diode with a diameter of 1 mm and a very low capacitance (4 pF), which is incorporated in the current HPD product, realizes a fast response. Figure 9a shows the time response waveforms of an HPD and a PMT. The FWHM for the HPD is 0.6 ns, which is smaller than that for PMT (1.6 ns).

**Figure 8.** Output height distribution of an HPD [32]. The incident pulsed lights (wavelength = 470 nm) were adjusted to make the photocathode emit three photoelectrons on average. The cathode supply voltage and the reverse voltage were −8 kV and 380 V, respectively.

**Figure 9.** Time response characteristics of an HPD and a PMT [32]: (**a**) time response waveforms of an HPD and a PMT; (**b**) transit time spread (TTS) of an HPD and a PMT. The data were obtained using a pulse laser (wavelength = 405 nm) with a pulse width of 77 ps. The cathode supply voltage and the reverse voltage of HPD were −8 kV and 390 V, respectively.

The TTS determines the instrument response function of HPD. The following three factors mainly affect the TTS for HPD: (i) the transit time within the photocathode; (ii) the variation in the time taken for the photoelectrons to move in the vacuum from the photocathode to the avalanche diode; and (iii) the electron transit time within the avalanche diode. The TTS for an alkali photocathode is about 50 ps [31]. The measured raw TTS for a GaAsP photocathode in an HPD is about 113 ps, which is larger than that for alkali because the GaAsP layer is thicker than the alkali layer. This raw TTS value included the laser pulse width (77 ps) and temporal resolution of the measurement system (30 ps), so the net TTS should be much smaller. Figure 9b shows the time response waveforms of an HPD and a PMT. The raw TTS for the GaAsP photocathode in a PMT is 300 ps, which is larger than that of an HPD. These characteristics are very important for time-resolved fluorescence detection because it contributes to accurate fluorescence lifetime measurement.

#### Large Active Area

HPDs have a large effective area of more than several millimeters in diameter, enabling high photon collection efficiency. The active area of HPDs is comparable with that of PMTs, which is a marked contrast to SPADs with an effective diameter of only 10 micrometers. Table 2 lists the characteristics of PMTs, SPADs, and HPDs.

#### **Table 2.** Characteristics of point photodetectors [23,27,32].


#### 3.4.2. HPD and Fluorescence Microscopy

The features of HPDs mentioned in Section 3.4.1 show the preeminence of HPDs in fluorescence microscopy applications, such as LSM, FLIM, and FCS. In practice, HPDs are incorporated into commercial microscopes and products of Becker & Hickl GmbH and PicoQuant.

#### FCS

In FCS, the presence of afterpulses deforms the correlation spectra. To avoid this issue, the fluorescence signal is commonly divided into two and detected by two detectors, and the cross-correlation between the two signals is calculated. This procedure is complicated and decreases the signal-to-noise ratio of each acquired datapoint. In contrast, the afterpulse-free feature of the HPD makes FCS measurement simpler, and a single HPD provides better data than PMT [13,26]. As shown in Figure 10, the autocorrelation spectrum acquired by an HPD is of good quality, whereas the spectrum acquired by PMT contains overlapped afterpulse noise below the 2 μs region.

**Figure 10.** Autocorrelation data obtained by an HPD and a PMT [32]. The data were acquired using 100 nM Alexa Fluor 532 dye solution.

#### FLIM

FLIM utilizes time-correlated single photon counting (TCSPC) and acquires higher temporal resolution data (<ns) than FCS. Figure 11 shows a comparison of fluorescence lifetime measurement data obtained by an HPD and a PMT. Afterpulses are also problematic for FLIM. The afterpulse raises the baseline of the PMT data. Consequently, the dynamic range of PMT is lower by an order of magnitude than that of HPD and seriously interrupts determination of the decay time constant.

**Figure 11.** Fluorescence lifetime measurement data obtained by an HPD (**a**) and a PMT (**b**). The data were acquired using fluorescein dye solution. Adapted with permission from [13].

Compared with SPADs, HPDs can obtain brighter FLIM images due to the larger active area of the HPD. Figure 12 shows FLIM images acquired by an HPD and a SPAD. Although the quantum yields of the photodetectors were similar, the image acquired by the HPD collected twice as many photons as acquired by the PMT.

**Figure 12.** FLIM images acquired by HPD (**a**) and SPAD (**b**). Adapted with permission from [13].

#### 3.4.3. Other Types of HPDs

This subsubsection briefly refers to other types of HPDs that were developed by Hamamatsu Photonics (Table 3) and applications of the HPD to single-molecule fluorescence microscopy. These other types of HPDs are cooled HPD and MPPC (SiPM) incorporated HPDs.

**Table 3.** Other types of HPDs [32–35].


Table 3 lists the characteristics of these HPDs.

The cooled HPD, in which the photocathode is cooled by the Peltier element, reduces thermal electronic noise from the photocathode to one-tenth that of the non-cooled HPDs (Figure 13). This HPD is supposed to be useful for the detection of extremely low light, including single molecule detection.

**Figure 13.** Dependence of dark count rate of the developed cooled HPD on the applied Peltier current [32]. Application of 1.2 A Peltier current cooled the HPD to about 283 K at an ambient temperature of 298 K.

The MPPC-incorporated HPD was developed to solve the difficulties of operating the HPD under high voltages such as 8 kV. The MPPC-incorporated HPD can operate with a lower voltage that operates PMTs due to the MPPC's high gain, and is able to detect single photons. The first prototype of an MPPC (SiPM)-incorporated HPD with a GaAsP photocathode with a diameter of 3 mm and a 25.4-mm (1-inch) bi-alkali photocathode type of MPPC (SiPM)-incorporated HPD were developed by Hamamatsu Photonics, and Barbato et al. evaluated its characteristics [33,34]. As shown in Figure 14, the HPD operates with lower voltages (photocathode voltage ~−3 kV and MPPC bias voltage ~ +70 V) than those of the current HPD product that incorporates an AD (photocathode voltage ~−8 kV and avalanche photodiode bias voltage ~+450 V). The low voltage operation capability facilitates installation of the HPD into various apparatus, including fluorescence microscopes. Recently, Hamamatsu Photonics developed an HPD with a 50.8-mm (2-inch) diameter [35].

**Figure 14.** Schematic of multi-pixel photon counters (MPPC)-incorporated HPD.

3.4.4. Application of HPDs to Single-Molecule Fluorescence Microscopy

We applied the cooled-photocathode HPD to single-molecule fluorescence microscopy. We applied the HPD to low-background wide-field single-molecule fluorescence detection with high temporal resolution as a proof-of-principle demonstration. The fluorescence collected by an objective was divided into two, each simultaneously detected by HPD or imaged by EMCCD. The HPD allowed the fluorescence intensity of a mobile single molecule fluorophore to be determined at higher temporal resolution than conventional high-sensitivity CCD cameras. Specifically, the cooled-photocathode HPD detected fluorescence of a single Qdot while performing two-dimensional diffusion with 0.1 ms temporal resolution (Figure 15).

**Figure 15.** Wide-field sub-millisecond single-molecule fluorescence detection by a cooled HPD [32]: (**a**) Schematic of the observed mobile Qdot. The Qdot (Qdot655 streptavidin conjugate) was attached to biotinylated phosphoethanolamine (PE), 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(cap biotinyl), via biotin-streptavidin interaction in a 1,2-dioleoyl-snglycero-3-phosphocholine (DOPC) lipid bilayer formed on a coverslip. The Qdot was excited by a CW laser (wavelength = 532 nm); (**b**) Time course of a mobile Qdot fluorescence intensity was obtained by an HPD with a time resolution of 0.1 ms (red). The repeated fluorescence-on and –off are caused by blinking. The intensity profile matches the intensity simultaneously obtained by EMCCD (blue) with a time resolution of 31.319 ms; (**c**) Trajectory of the Qdot obtained by EMCCD. The inset is a fluorescence image of the observed mobile Qdot.

The HPD also enabled wide-field single-molecule fluorescence lifetime measurement with nanosecond temporal resolution. We succeeded in obtaining time courses of fluorescence lifetime of a single Qdot whose fluorescence images were simultaneously monitored by an EMCCD (Figure 16).

**Figure 16.** Wide-field single-molecule fluorescence lifetime measurement by a cooled HPD: (**a**) Schematic of the observed mobile Qdot. Sample preparation for the observation was identical to that described in the caption of Figure 15. The Qdot was excited by a pulse laser (wavelength = 520 nm); (**b**) Time courses of the Qdot fluorescence (upper, blue) obtained by an EMCCD and the lifetime (lower, red) simultaneously obtained by an HPD. The lifetime was obtained by fitting each decay curve drawn using data accumulated in three seconds with a single-exponential (red); (**c**) Trajectory of the Qdot obtained by EMCCD. The inset is a fluorescence image of the observed mobile Qdot.

A group at Tohoku University incorporated two HPDs into a specially equipped line confocal optical system in which a slit was used instead of a pinhole to improve the time resolution of single-molecule FRET study. They achieved FRET observation of single protein molecules that flowed unidirectionally in a flow cell at a time resolution of 10 μs and analyzed the high-speed folding process of protein molecules [36].

#### **4. Emerging Point Detectors**

This section introduces newly emerging photodetectors that can be used for fluorescence microscopy. These photodetectors are silicon photomultipliers (SiPMs) also known as multi-pixel photon counters (MPPCs) and superconducting nanowire single photon detectors (SSPDs).

#### *4.1. Silicon Photomultiplier (SiPM)*

Silicon photomultipliers (SiPMs), which offer single photon detection capability, are an emerging photodetector in a variety of industries and biological fluorescence microscopy. Caccia summarized applications of SiPMs to biophotonics including fluorescence microscopy [37]. SiPM consist of a SPAD array. The sum of pulses from all SPADs is the SiPM output. Note that the number of SiPM output is single irrespective of the number of SPADs in the SiPM. Many photons can be simultaneously detected by the SPADs. Figure 17 shows a schematic of the operating principle of the SiPMs.

**Figure 17.** Schematic of the operating principle of silicon photomultipliers (SiPMs): SiPMs consist a SPAD array and every SPAD acts as an element that generates an all-or-nothing current pulse. When one or more photons are absorbed, a current pulse is generated. A current pulse is also produced by the dark count, whereas a current pulse is not generated when photon absorption fails. The output is the sum of the SPADs. Adapted with permission from [38] © The Optical Society.

SiPMs have several advantages over PMTs, including low fabrication cost, low operating voltage, and extremely high damage thresholds (high durability). In addition, silicon diodes have high quantum efficiency in the near-infrared region used for deep tissue imaging. Due to these factors, SiPMs can be better suited for high-speed imaging. Although SiPMs can detect intense light, their dark count rate is larger than that of PMTs, which is the major tradeoff. The advantage of high damage thresholds is reportedly distinct for confocal fluorescence microscopy and two-photon microscopy in clinical sites where surgical marking inks emit intense fluorescence [39]. SiPMs have not yet been widely used for biological imaging. Giacomelli et al. evaluated the performance of commercial SiPMs by comparing the SiPMs with a GaAsP PMT for LSM. They reported that the SiPM sensitivity exceeds the PMT sensitivity for moderate- to -highspeed LSM, whereas the PMT exhibited better sensitivity due to its lower dark counts for low speed LSM [39]. Modi et al. also compared SiPMs products with a GaAsP PMT for two-photon imaging of neural activity [38]. They showed that SiPM exhibited a signal-to-noise ratio that was comparable to or better than PMTs in usual calcium imaging, though dark counts of the SiPMs were higher than that of the PMT. They concluded that the low pulse height variability of the SiPMs surpassed the weak point and resulted in high performance.

#### *4.2. SSPD*

In recent decades, the superconducting nanowire single photon detector (SSPD) has become an increasingly popular device, with applications ranging from sensing to quantum communications for single photon detection with high efficiency, precise timing, and low noise [24,40]. The crucial high-speed feature of the SSPD is represented by the TTS (<50 ps) the fluctuation between the true arrival time of a photon and the electrically registered arrival time recorded by the system. Korzh et al. showed that the use of low-latency materials lowered the TTS of the SSPD, and demonstrated that the temporal resolution can be 2.6 ± 0.2 ps for visible wavelengths and 4.3 ± 0.2 ps at 1550 nm using a specialized niobium nitride SSPD [41].

Figure 18 shows a schematic of the SSPD. The SSPD consists of superconducting nanowire with a thickness of a few nanometers that senses photons. Single photon absorption by the SSPD suppresses superconductivity, which in turn generates a voltage spike that can be used to detect the photon.

**Figure 18.** Schematic of the operating principle of the superconducting nanowire single photon detector (SSPD): The SSPD consists of a superconducting nanowire with a thickness of a few nanometers that senses photons. A bias electrical current flows through the nanowire during the operation: (**a**) the entire area is in the superconducting state prior to photon absorption; (**b**) single photon absorption takes place in the superconducting nanowire; (**c**) superconductivity is locally suppressed by energy excitation via the photon absorption; (**d**) the bias current makes the suppressed area resistive and the resistive area expands across the nanowire, which in turn generates a voltage spike; (**e**) as current is diverted, the resistive area relaxes to the superconducting state. Adapted with permission from [42]. Copyright 2020, American Chemical Society.

The SSPD is free of afterpulses because it returns to the superconducting state without generating afterpulses after photon detection. The afterpulsing-free characteristic of the SSPD is a clear advantage for time-resolved fluorescence microscopy. The SSPD has been used for fluorescence microscopy and applied to FCS [43,44]. Although the SSPD has these excellent features, its operation is inconvenient due to its small active area and low operating temperature. The very small active area (~10 μm) makes the optical alignment difficult and the extraordinary low operating temperature (≤4 K) requires the liquid helium cooling system.

Detailed comparisons have been made between SSPDs, SPADs, and other photoncounting technologies in [27,40,44].

The characteristics of SiPMs and SSPDs are summarized in Table 4.


**Table 4.** Characteristics of emerging photodetectors [24,27].

#### **5. Summary and Outlook**

This paper provides an overview of low-light point photodetectors used in fluorescence microscopy and introduces several point detectors and their operating principles, focusing on HPDs that exhibit high timing resolution and low afterpulse. In addition, we demonstrate application of HPD to wide-field single-molecule fluorescence detection.

Fluorescence imaging with a point photodetector needs scanning to acquire an image and thus temporal resolution of the imaging is often limited. To overcome this limitation, excitation methods other than point excitation, such as multifocal excitation line excitation, have been proposed [45,46]. Regarding photodetectors, imagers with a large number of pixels have been reported by many groups [47–50]. For example, Zickus et al. reported scan-less wide-field FLIM using a camera consisting of a 500 × 1024 SPAD array at a rate of 1 Hz [51]. Michalet et al. firstly reported the evaluation of a multi-pixel (8 × 8) HPD developed by Hamamatsu Photonics [11]. Fukasawa, an author of this paper, and his colleagues, reported another multichannel HPD that was composed of 32 channels (two lines of 16 pixels) on a chip, in which the size of each pixel was 0.8 × 0.8 mm (Figure 19) [52]. It was confirmed that the timing resolution and afterpulse characteristics of the multichannel HPD are identical to the conventional single channel HPD. Wollman et al. reported an 1024-element SSPD array (a 32 × 32 row-column multiplexing architecture) [53]. Fast acquisition methods for FLIM and multi-pixel photodetectors have been reviewed by Liu et al. [22]. These point detector arrays can be widely applied to simultaneous multiparameter observation, including simultaneous multi-wavelength fluorescence observation.

**Figure 19.** A photograph of the developed multichannel HPD. Adapted with permission from [52]. Copyright 2016, Elsevier.

Low-light photodetectors find more applications in various fields not limited to biological fluorescence microscopy [23]. These applications include flow cytometry and polymerase chain reaction (PCR) in life science, positron emission tomography (PET) for medical diagnosis, elementary particle (neutroino etc.) detection and collision experiments in high energy physics, and semiconductor wafer inspection in industry. In the near future, combination of fluorescence microscopy and other modalities may make lowlight photodetectors evolve further and may provide more detailed information on target biological specimens.

**Author Contributions:** Conceptualization, H.Y.; writing—original draft preparation, H.Y.; writing review and editing, H.Y., A.F., M.H., and T.I.; investigation, H.Y. and A.F.; formal analysis, H.Y. and A.F.; supervision, H.Y.; visualization, H.Y. and A.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the Science Research Promotion Fund of the Promotion and Mutual Aid Corporation for Private Schools of Japan (to H.Y., M.H., and T.I.) and the research grant of Tokai Foundation for Technology (to H.Y. and A.F.).

**Data Availability Statement:** The data that support the findings of this study are available from the corresponding author upon reasonable request.

**Acknowledgments:** We would like to thank Yoshihiro Takiguchi for his advice on pulse lasers.

**Conflicts of Interest:** The author declares that there are no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Abbreviations**


#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Applied Sciences* Editorial Office E-mail: applsci@mdpi.com www.mdpi.com/journal/applsci

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel: +41 61 683 77 34

www.mdpi.com

ISBN 978-3-0365-4494-6