Integration of low-dimensional materials for energy-harvesting applications: current progress, scope, challenges, and opportunities

1.1 III-V semiconductors on silicon for optoelectronics applications

Silicon technology is one of the remarkable achievements the last century has witnessed. Due to the abundance of silicon material, the extraordinary silicon properties, and the significant advancement in silicon lithography processes, various large-scale electronics can be easily realized. Extremely fast and reliable computer and mobile phone processors are becoming a reality, where, >20 years ago, personal computers and smart phones were a dream. Thanks to the scientific advancement in silicon technology, silicon electronics have enabled us to virtually communicate with anyone around the globe with just click of a button.

Silicon-based solar cells, however, cannot achieve >25% conversion efficiency. Mainly because silicon can only cover a small portion of the solar spectrum, rendering the rest of the spectrum untouched. Therefore, a new integrable approach to efficiently use the entire spectrum is required.

III-V materials such as gallium arsenide (GaAs) and gallium nitride (GaN) are widely integrated for optoelectronic applications. Ideally, integrating these semiconductor solar junctions on silicon solar junction is an intriguing method. However, due to the noticeably large lattice mismatch between III-V materials and silicon, integration of photovoltaics based on III-V semiconductors on silicon wafers has been a staggering challenge. It is desirable to have a junction that spans the entire solar spectrum, which spans from ultraviolet to near infrared. However, use of multiple junctions, or tandem cell, is one practical approach to cover this solar spectrum, where each junction is designed for a specific spectrum range, thus contributing to the total current. To date, the highest reported efficiencies using four solar junctions are 38% and 46% for non-concentrated and concentrated illuminations, respectively [3], [4].

The downside of this multijunction approach is the expensive cost of the growth technique and the substrate along with the limited choice of materials. Also, the design of multijunction devices requires current matching in order to achieve the highest efficiency acquired, which imposes another challenge in the fabrication of solar junction.

One approach to tackle the silicon lattice mismatch is to grow vertical nanowires on top of silicon pn junctions. Due to their low dimensionality, nanowires can overcome the lattice mismatch by strain relaxation at the edges. For instance, GaAs nanowires can be grown using the metalorganic chemical vapor deposition (MOCVD) technique. Doping of MOCVD-grown GaAs nanowires can be controlled by modifying the gas flow rate. In Figure 1A and B, the GaAs nanowire tandem cell and the scanning electron microscopy (SEM) image of as-grown GaAs nanowires are shown, respectively. In order to obtain the maximum possible current, a low-resistance connecting junction is required between the nanowire and the silicon junctions.

Figure 1:

(A) Schematic diagram of the GaAs nanowire on silicon tandem cell. (B) SEM image of the grown GaAs nanowires showing high growth yield. The scale bar is 1 μm. (C) Measured current-voltage characteristic of the GaAs nanowire junction (blue circles), silicon junction (red triangle), and tandem cell (black square). The measured open-circuit voltage is approximately the summation of the open circuit current of both junctions [5].

A recent attempt to fabricate high-efficiency tandem solar cells using GaAs nanowires on top of silicon pn junctions have shown satisfactory results. Yao et al. measured efficiencies close to 11.4% for these two junctions [5]. In Figure 1C, the measured current-voltage characteristics of the tandem cell are shown along with each stand-alone junction under the AM 1.5 G solar spectrum. The obtained current density for the tandem cell is 20.64 mA/cm², with open-circuit voltage of 0.956 V and fill factor of 0.578.

The integration of GaAs nanowire junction with silicon junction is aimed to produce efficiencies close to or higher than the Shockley-Queisser limit. In fact, the simulated result should exceed the silicon solar cell efficiency. The advantage of this technique is in the fabrication processes, where a SiN mask is used to pattern the GaAs nanowires, making it easy to grow controlled diameter nanowires. However, the low 11.4% efficiency measured for the tandem cell indicates major constrains. First, the design should optimize each junction parameter separately in order to obtain conversion efficiencies close to the theoretically predicted efficiency of each cell. In light of this, the GaAs nanowire junction lacks a concrete doping characterization method, although photoluminescence measurement is a viable approach. Also, defect states and other non-radiative recombination states should be handled with care. A careful choice of the passivation layer can reduce such non-radiative recombination states, which is absent in Figure 1A. Secondly, a low-resistance connecting junction between the nanowire junction and the silicon junction is required in order to obtain the maximum conversion current. Although this measured efficiency is low, this proof-of-concept tandem cell is a promising approach to obtain efficiencies that can surpass the Shockley-Queisser limit with two junctions.

Another promising approach to solve the lattice mismatch between III-V semiconductors and silicon is the inclusion of a buffer layer. That is, a 2D layer on silicon connected to the III-V material by van der Waals epitaxy can alleviate the lattice mismatch problem [6]. van der Waals epitaxy is advantageous in the sense that the stress caused by the lattice mismatch can be relaxed due to weak bonds at the interface between 3D/2D. The choice of this layer is crucial in order to prevent instability of the grown III-V semiconductor and ensure the sticking of the material on top of the 2D layer.

Figure 2A and B demonstrate this approach where graphene is used as the buffer layer between III-V semiconductors, in this case GaAs, and the silicon substrate. Here, graphene was used due to its thermal stability, which can accommodate high growth temperatures. GaAs was grown using molecular beam epitaxy with different growth parameters. The top graphene layer is under strain in order to lattice match the bulk GaAs layer. The final structure is expected to produce high-quality GaAs grown on silicon substrate, as illustrated in Figure 2B.

Figure 2:

(A) Atomic diagram of the GaAs/multilayer graphene/silicon layers. (B) Visual representation showing the stacking of the GaAs/graphene/silicon layers [6].

Although this is a viable approach to III-V semiconductors on silicon, there are still major challenges that hinder the fabrication of optoelectronics using this approach. One major issue is the limitations of the buffer layer between the silicon and the III-V semiconductor. That is, the surface energy of the 2D material layer can be low, which tends to produce island growth of GaAs on graphene instead of a uniform GaAs film [6]. Thus, the ability to control the growth sites is still open for research. Moreover, the stability of the grown GaAs on top of the buffer layer is of a noticeable concern. This is because the adsorption energy of Ga and As on graphene is pretty low, although an ultra-smooth GaAs surface has been achieved on graphene/silicon substrate using optimized growth parameters, as shown in Figure 3A and B. Producing high-efficiency GaAs solar cells on silicon substrate using a 2D buffer layer is still a challenge.

Figure 3:

(A) SEM image of grown GaAs with Ga pregrown on multilayer graphene buffer layer. The surface shows smooth morphology, which is confirmed using (B) atomic force microscopy image. The measured root mean squared (RMS) roughness is around 0.6 nm with peak-to-peak value of 3 nm [6].

1.2 Organic photovoltaic

OPV is a type of solar cell that exploits organic semiconductors. Unlike conventional solar cells, OPVs can be produced with inexpensive, environmentally friendly, carbon-based materials. OPV’s flexibility, transparency, lightweight nature, high absorption property, and the decent indoor performance are some of the key features that allow them to be integrated with portable electronic and biotronic devices [7].

Carbon-based molecular chains constitute the absorption layer or the active layer of the OPVs. These molecular polymers are deficient in delocalized charges when compared to inorganic counterparts. Hence, much less conductivity is notable in them [8]. However, specially prepared polymers can conversely provide delocalized electrons through the π bond system [8], [9]. As a famous example for polymers with delocalized electron encouraging bond arrangement, aromatic benzene molecule offers multiple π bonds that contribute to the conductivity enhancement in the material.

Specially prepared polymers can be put together to form a semi-pn junction with an internal electric field. This pn junction can be called acceptor/donor junction. The acceptor type of conductive polymers (analogues to p-type doped material) can take the form of a long molecular chain such as P3HT, while the donor type of “0D” carbon allotrope derivatives called fullerenes (analogues to n-type material) can take the form of PCBM. The donor/acceptor interface can be engineered to build various types of OPV devices.

The donor-acceptor interface can be realized either in a planar (PHJ) or bulk heterojunction (BHJ) configuration [10], [11]. The PHJ is considered the simplest structure of OPVs, which consists of sequentially deposited donor/acceptor layers through solution spin coating. The active layer is sandwiched between selective transport layers, in addition to the contacts at both ends. Low efficiencies were observed with PHJs due to the extremely low nominal diffusion length for organic polymers, which is in the order of 10 nm, nearly five orders of magnitude less than silicon.

BHJ polymers, on the other hand, can be realized to supply more dispersed junction contour length, hence a higher probability for exciton to be dissociated. The method used to make a BHJ layer is to blend both the donor material and the acceptor material in an optimized ratio. Figure 4A shows the structure of both FPJ and BHJ OPV device, in addition to the typical band diagram of OPV.

Figure 4:

(A) BHJ structure (left), FPJ structure (middle), and typical band interfaces (right) of an OPV device. (B) Device schematic and (C) J-V curves at different ZnO layer thickness of the structure in (B) [12]. (D) Device structure and (E) measured J-V curves of the highest recorded efficiency of tandem organic solar cell [13].

OPVs have a total of four steps in order to collect a charge at the contacts. These four steps are (i) light absorption, (ii) exciton generation and diffusion, (iii) charge transport, and (iv) charge collection. The majority of organic semiconductors have high band gap values, which overwhelm the energy and the amount of absorbed photons. A great deal of research is taking place in terms of band-gap engineering to produce narrower band gaps.

In the pursuance of high-efficiency OPVs, research efforts have mainly focused on four critical directions. These directions include engineering new types of narrow-band-gap polymers; experimenting with different device structures in terms of tandem and buffer layer incorporation; light management techniques; and the incorporation of nanoparticles, nanotubes, or nanowires in different layers of the OPV structure.

The first direction was extensively studied in the last decade, and had significant impact in terms of breaking efficiency records [12], [13], [14], [15]. Band-gap engineering in polymer materials has allowed the tuning of the interface levels of the polymers. Low-band-gap polymers are usually used as the donor materials and fullerene derivatives, such as PC71BM, have been widely adopted as the acceptor component given their relatively high electron affinity and charge carrier mobility. By definition, lower-band-gap materials will have higher catchment area of the solar spectrum, hence an increase in the internal quantum efficiency and consequently a boost to the short circuit current and power conversion efficiency. As an example, Figure 4B shows a recently studied OPV structure that employs a narrow-band-gap material [12]. The structure is created using an inverted OPV approach. The carrier selection layer was deposited using a low-temperature mist CVD method. In contrast to the high-temperature and low-vacuum traditional methods, this approach offers low-temperature and open-air pressure processing for depositing the ZnO selective layer. Avoiding high-temperature and high-vacuum processes are of a great importance to increase the economic feasibility, increase the process expediency, and protect the vulnerable organic layers from any damage and degradation, which could arise from high-temperature deposition. One of the advantages of using ZnO as a buffer layer is its selectivity in conducting electrons and blocking the holes in inverted structures. Additionally, ZnO offers higher electron mobility and light transparency.

Figure 4C shows the measured J-V characteristics at various buffer layer (ZnO in this case) thicknesses. According to the measured J-V curves, the best performance for such a cell occurs with ZnO thickness of 14 nm. The cell parameters at the optimal ZnO thickness are J_SC=16.6 mA/cm², V_oc=0.71 V, and FF=73%, with conversion efficiency of 8.6%.

It is worth mentioning that the current world efficiency record for single-layer OPV has been obtained very recently in 2016 by chemically modifying the solvent of the active polymers [16]. The work has made a simple approach, which is to replace the non-halogenated solvent with a hydrocarbon-based solvent, which also reduced the degradation of the active polymer itself. This high efficiency was reported by the same group that held the previous 11.5% world record OPV efficiency according to the latest National Renewable Energy Laboratory chart.

In terms of device structure innovation, tandem OPVs have offered a great utilization of covering a large range of the solar spectrum. You et al. have obtained the current world record that is held by this kind of structure [13]. As Figure 4D shows, the device structure incorporated two active layers with different effective band gaps, with an interlayer between them to aid common charge collection. The incorporation of two layers with different band gaps is to cover a wide range of the solar spectrum. At both ends (top and bottom), buffer layers were deposited to provide charge transport, finalized by an Ag contact from one side and a transparent ITO from the other. The J-V characteristics of the final structure are illustrated in Figure 4E. The obtained cell characteristics are J_SC=10.4 mA/cm², V_oc=1.4 V, FF=71%, and cell efficiency around 10.7%. The observed high V_oc value indicates that the two tandem layers are connected in series.

OPVs require careful examination of the contacts and the buffer layer work functions. The energy band alignment between the electrode of choice and the active layer polymer will significantly affect the charge collection (hence, the conversion efficiency). Failing to optimize the electrode/active layer interface will cause an accumulation of uncollected charges [17]. Aside from aligning the work functions and band edges, other techniques such as facilitating a buffer layer that blocks charges from going to the undesired electrode have been reported [18]. Moreover, degradation of the active layer could take place if the buffer layers as well as the contacts are not carefully chosen [19].

In terms of light management and optical properties, OPVs have witnessed many techniques to improve the efficiency. One example was the incorporation of metallic nanoparticles to use localized surface plasmon resonance (LSPR) [20]. LSPR is an optical phenomenon that takes place upon the interaction of metal nanoparticles with light. When the light wavelength is larger than the nanoparticle size, the LSPR commence. High oscillation of surface electrons forms a localized electric field around the nanoparticles, which, in turn, will increase the mobility. In addition to the mobility increase, the oscillation of electrons will emit light around the band-gap wavelength, which will increase the absorption, and ultimately the efficiency. LSPR can be tuned by many factors such as the material type, the geometrical size of the nanoparticles, the shape of the nanoparticles, and the spacing between these particles. Xue et al. have shown that the efficiency and fill factor have increased by 60% upon the incorporation of metallic nanoparticles [21]. Patterning of the buffer layer along with the antireflection coating layer is a technique that has been used to boost photon harvesting [22], [23]. It did not, however, provide a landslide increase in the OPV metrics.

One of the key prospects for improving the open circuit voltage in OPVs is to reduce the density of trap states, which can be done by accurate control of the phase separated morphology, and by altering the interfacial arrangement of the donor/acceptor materials [24]. This approach of minimizing the interfacial trap states would help the fabrication of a thick photoactive layer, which, in turn, would increase the short circuit current without compromising the fill factor. Enhancing the contact work function alignment by accurate choice of materials, in conjunction with reducing the width of the Gaussian density of state, is one of the encouraging routes for enhancing the V_oc.

2.1 Graphene-based devices

The atomically thin graphene layer exhibits extraordinary electrical and optical properties. Some of these properties include high electron mobility [25], high mechanical strength with Young’s modulus of 1 TPa [26], and high thermal conductivity [27]. Yet, graphene is not applicable for most of the semiconductor applications due to the absence of the band gap. It is only possible to open a band gap in bilayer graphene by applying a vertical electric field [28], which poses a major constraint for fabricating large-scale electronics based on graphene devices.

Graphene optoelectronics, however, have been extensively studied. Fundamentally, monolayer graphene is relatively transparent between visible to infrared [29], [30]. It can be grown using CVD [31]. Recently, Kim et al. have shown bright light emission from a suspended graphene flake [32], proving that graphene can be used as a light emitter. Nevertheless, due to the semimetal nature of graphene, photodetectors based on graphene do not show a photovoltaic effect, which is the essential mechanism in current conduction in solar cells. Instead, they show a photo-thermoelectric effect where a temperature gradient is created optically, forcing the generated carriers to move from the hotter side to the colder side [33]. There have been extensive studies on graphene photodetectors. Here, we will focus on graphene/silicon junction, which can be easily scaled for commercialization.

Graphene/silicon junction is a promising device structure for large-scale optoelectronic applications. Graphene/silicon junction is fundamentally a Schottky junction, which can be useful for various energy-harvesting applications, including low-energy optical switches when used as photodetectors. Several groups have investigated the optical and the electrical properties of this structure [34], [35], [36], [37], [38], [39]. Figure 5A shows a typical graphene/silicon device where the graphene is in contact with the silicon substrate (in this case, n-type silicon substrate). The gold electrodes can be modified according to the preference of the group, as in the case of Figure 5B, where the measurements were taken with silver paste and gallium-indium eutectic electrodes as top and bottom contacts, respectively. In this figure, the current density-voltage characteristics are illustrated for the graphene/silicon device when operated as a solar cell and at different stages of treatment [40]. The black curve illustrates the solar cell J-V curve for a bare graphene/silicon junction. The observed efficiency is 3.78% with J_SC=22.9 mA/cm² and V_oc=0.387 V. These parameters can be enhanced by various methods. The first method is doping the graphene, which is illustrated as the red curve in Figure 5B. The purpose of doping in this structure is to enhance the open-circuit voltage as well as the fill factor. Consequently, the observed efficiency after HNO₃ doping is 8.91% with J_SC=23.89 mA/cm² and V_oc=0.548 V. The yellow curve in the figure corresponds to the doped graphene/silicon junction after coating with TiO₂ layer. Such a layer acts as an antireflection layer, minimizing the reflected incident light. For this particular cell, the performance parameters measured are V_oc=0.60 V, J_SC=32.5 mA/cm², FF=73%, and efficiency of 14.1%.

Figure 5:

(A) Schematic diagram of the graphene/silicon device. (B) Measured I-V curves of the graphene/silicon device when used as a solar cell. The black curve is the solar I-V of bare graphene on silicon. The red curve is after doping the graphene with HNO₃, and the yellow curve is the solar I-V after spin coating with TiO₂ [40]. (C) Solar I-V curves of the device in (B) showing the degradation of the cell performance after a long period of time. The recovered curve is measured after doping the cell with HNO₃. The inset shows the short-circuit current as a function of time. (D) Measured I-V curves of the graphene/silicon device at several incident laser powers when used as a photodetector [36].

The stability of the graphene/silicon junction solar cell after treatment is a major issue. In Figure 5C, the cell characteristics are measured on different days. The cell characteristics degraded after 20 days. This degradation occurred due to the instability of the doping technique used to dope the graphene flake. In fact, the cell efficiency is slightly enhanced after re-doping the graphene flake with HNO₃, giving 14.5% efficiency. To our knowledge, this efficiency obtained represents the highest efficiency reported to date on graphene/silicon structure. Yet, the stability of this structure to maintain this high efficiency is a subject of research.

The graphene/silicon junction can also be used as a photodetector [36]. In Figure 5D, the current-voltage characteristics of the device are illustrated at various incident optical powers. As shown, the maximum observed photocurrent occurs when the device is reverse biased, where the excited carriers are injected from the silicon to the graphene flake. This is because in the reverse biased condition, the difference in the Fermi level opens large numbers of available states, allowing large numbers of photoexcited carriers [36]. For this structure, the reported incident photon conversion efficiencies can be as high as 57% and the maximum responsivity is 435 mA/W. This high efficiency along with the high responsivity illustrates the potential use of graphene/silicon devices for photodetection applications.

We believe the obtained efficiency for this type of graphene devices can be optimized by various techniques. First, the existence of the native oxide between the graphene flake and the silicon substrate can degrade the device performance [35]. This layer could either be grown during the transfer of graphene or even after the graphene deposition. Such an issue can be prevented by careful fabrication techniques and considering different passivation layers. Another method to increase the efficiency is doping the graphene p-type (for n-type Si substrate). This will shift the Fermi energy in the graphene region to the valence band, making it easy to bias the device, hence low-power operations with high conversion efficiency. However, the doping method, doping concentration characterization, device stability with time, and doping-to-defect ratio are some of the crucial key points that need to be carefully addressed.

2.2 Carbon nanotube-based devices

Carbon nanotube research in the last few years have witnessed major improvements in terms of solving the grand challenges of carbon nanotube synthesis, which include purity and chirality control. Solving the synthesis issue will substantially put carbon nanotube electronic and optoelectronic devices on the industrial roadmap. A recent study to fabricate thin film transistors with high-purity semiconducting nanotubes has shown a large I_on/I_off ratio with large turn-on current using ultra-high-purity semiconducting carbon nanotubes. The high-purity semiconducting nanotube solution has enabled the realization of large-scale integration of nanotubes for electronics applications [41]. The potential of using carbon nanotube devices for electronic applications can be realized in the near future.

In an attempt to control the nanotube’s chirality, Liu et al. have used a carbon nanotube seed to “clone” the same chirality by optimizing the CVD growth parameters [42]. The resulting elongated nanotube exhibited the exact chirality as the starting seed. The method highlights the first attempt to produce a specific chirality nanotube, which depends on the starting seed. Yet, the challenge of scaling this method to industrial applications has been hinged on the full control of the placement of the cloned nanotubes in the desired positions. Some proposed solutions include nanotube transfer from one substrate to another. However, this approach can induce defects and possible degradation of the nanotube’s performance.

Although all these attempts to solve the synthesis issues for large-scale integration are still under investigation, the optoelectronic properties of carbon nanotubes have been extensively studied in the past few years [43], [44], [45], [46], [47], [48].

Nanotubes have a direct band gap, which do not require any phonons to emit or absorb a photon. They exhibit strong exciton binding energy due to the 1D nature of these materials. In terms of applications, these excitonic transitions can be used for various biomedical sensing and communication applications [49], [50]. When biased in a pn junction configuration, semiconducting nanotubes exhibit ideal diode behavior [51], efficient photocurrent detection [52], and photocurrent enhancement at the excitonic transitions.

Recent studies on carbon nanotube/silicon junction, similar to the graphene structure in Figure 6A, have shown high conversion efficiencies [53], [55]. Unlike the graphene/silicon junction, semiconducting carbon nanotube/silicon junction is ideally a pn junction. However, due to uncontrollable CVD growth process, a mixture of metallic and semiconducting nanotubes are produced, rendering a Schottky junction with the underlying silicon substrate. Figure 6A illustrates the device structure where the nanotube thin film is lying on top of the silicon substrate. In Figure 6B, the solar cell J-V characteristics of CVD-grown nanotubes on n-type silicon substrate are demonstrated. The measurements were taken under non-concentrated AM 1.5 analogues to Figure 5B; the device is coated with TiO₂ antireflection coating layer and chemically doped using HNO₃. The resulting solar cell characteristics, after applying the treatment processes, are V_oc=0.61 V, J_SC=32 mA/cm², FF=77%, and efficiency of 15%, which is quite high for such a device. Nevertheless, the carbon nanotube/silicon junction suffers from time stability, similar to the graphene/silicon junction, due to doping degradation, which can be resolved by chemically doping the cell with HNO₃.

Figure 6:

(A) Schematic diagram of the carbon nanotube/silicon junction with TiO₂ layer as an antireflection coating layer. (B) The measured J-V of the structure in (A) at different treatment processes. The highest efficiency obtained after HNO₃/H₂O₂ doping and TiO₂ coating [53]. (C) Schematic diagram of carbon nanotube photodetector using a split gate structure. (D) Photocurrent enhancement with increasing gate voltage (Vg1=-Vg2) for the structure in (C) [54].

It should be noted that semiconducting nanotubes suffer from a large contact resistance. In fact, research is still undergoing to optimize the contact resistance of carbon nanotubes [56], [57]. The current findings suggest that nanotubes with diameters between 1.5 and 1.6 nm start to collapse under the deposited metal and affect the device performance.

Quasi-metallic nanotubes, on the other hand, are surfacing to become useful for optoelectronic applications. Ideally, metallic nanotubes do not exhibit a band gap [58], mainly due to the absence of this band gap whenever quasi-metallic nanotubes are lying on a substrate [59]. One of the advantages quasi-metallic nanotubes exhibit is low contact resistance due to the absence of Schottky contacts, as opposed to their semiconducting counterparts. They also exhibit higher photocurrent values compared to semiconducting nanotubes [60]. Figure 6C shows a suspended quasi-metallic carbon nanotube pn device. The split gate in the bottom of the trench can be biased separately to induce electrostatic doping. In Figure 6D, we showed photocurrent enhancement of this carbon nanotube photodetector with increasing electrostatic doping [54]. The obtained efficiency was <1%.

The peak in Figure 6D corresponds to the excitonic transition E¹¹ of the nanotube. In fact, a recent study showed that quasi-metallic nanotubes can exhibit photo-thermoelectric effect, similar to graphene photodetectors, if the mini-gap is less than the exciton binding energy (50 meV in this case) [61]. For band gaps >50 meV, the photovoltaic effect dominates the photocurrent transport. Although the efficiency of this photodetector is <1%, this work was the first demonstration of quasi-metallic nanotubes as photodetectors. Thanks to the mini-gap that metallic carbon nanotubes exhibit, optoelectronics based on quasi-metallic nanotube devices can be realized.

3.1 Photovoltaic devices

The possibility of implementing nanodevices for energy-generation applications is rapidly increasing. Many of the unknown fundamental physical and chemical properties are being discovered on a daily basis. Advancements in material fabrications have led to new emerging devices based on 2D materials. The most prominent 2D materials to date are molybdenum disulfide (MoS₂), tungsten diselenide (WSe₂), tungsten sulfide (WS₂), and phosphorene. Experimental investigations using these devices show that photovoltaic devices can be easily scaled down to the nanosize without complex fabrication processes. For example, Buscema et al. have shown that black phosphorus multilayers exhibit a photovoltaic effect when an electrostatic homojunction is created [62], analogous to the nanotube in Figure 6C. Other similar studies on MoS₂ and WSe₂ show similar effects [63], [64]. We should mention that devices based on the split gate structure are ideal for understanding the physical phenomena associated with the material under study. Nonetheless, electrostatic doping requires two separate electrodes, hence two voltage sources, which render the device inefficient and impractical for industrial applications. A more scalable design is required to achieve similar or even better performance.

Tsai et al. measured the solar cell performance of MoS₂/silicon junction. The results yielded V_ov=0.41 V, J_sc=22.36 mA/cm², FF=57.26%, with efficiency of 5.23% [65]. Although these cell characteristics are lower than those of graphene/silicon junction, it was the first demonstration of MoS₂/silicon junction solar cell, which can be further improved by several treatments, similar to the graphene/silicon junction discussed above.

Another interesting photovoltaic approach is van Der Waals stacking of 2D materials to create pn junctions. Many groups have investigated this approach with different 2D layers [64], [66], [67], [68]. The maximum reported external quantum efficiency, to our knowledge, is 34% for multilayer MoS₂/WSe₂pn junction [67]. In fact, Furchi et al. measured the solar cell response of monolayer MoS₂/WSe₂pn junction and found a conversion efficiency <1% [69]. Similar conversion efficiency has been obtained using GaTe-MoS2 pn junctions. Although 2D stacked structures exhibit high quantum efficiency, the solar cell behavior of such structures requires adequate improvements in order to achieve low-cost 2D solar cells.

As discussed in Section 1, the tandem cell approach is very promising given less fabrication processes (and hence cost effectiveness). In Figure 7A, the solar spectrum is illustrated. Most of the low-dimensional photovoltaic devices we mentioned in this article fall in the visible to mid-infrared regime (approximately between 500 and 1200 nm). Although integrating some III-V semiconductors can cover the entire spectrum, there are still wide solar spectrum regions that need to be investigated using 2D materials, especially in the infrared. In fact, carbon nanotubes and graphene have been used as infrared detectors [70], [71], [72]. The detection mechanism of these carbonaceous devices is mainly bolometric. They also suffer from low conversion efficiency. Another candidate for infrared detection studies is black phosphorus, where the band gap is around 0.3 eV [73]. To this date, no reported studies have been published due to the recent discovery of this 2D layer.

Figure 7:

(A) Solar spectrum vs. wavelength. The region between the dashed lines is where most low-dimensional devices discussed in this article can cover. (B) Schematic illustration of the thermoelectric power generator. The contrast in the temperature between the top and the bottom of the thermoelectric material results in current generation, which is detected by the external resistor. (C) Electrical conductivity, (D) Seebeck coefficient, and (E) power factor at different thermal conductivities for several 2D materials obtained from the literature. The arrows indicate the desired area where a good thermoelectric materials should fall in. (F) Estimated conversion efficiency of some of the 2D materials discussed in Table 1. For an ideal thermoelectric device, the maximum attained efficiency should reach the Carnot efficiency (black curve).

3.2 Thermoelectric devices

So far, we have discussed photovoltaic devices that operate by converting the incident photons to electricity. Another source of energy that solar power provides is heat. Some geographic areas exhibit high temperatures, especially during the summer. This wasted heat can be a good source of energy if invested wisely. Thermoelectric devices are one way to use this wasted heat. Figure 7B shows a schematic diagram of a thermoelectric power generator. The Seebeck phenomenon, which is the sole mechanism of thermoelectric devices, arises when the surface of a material is thermally excited. This leads to a temperature gradient across the material. During the process, carriers tend to move from the hot side to the cold side. For an n-type (p-type) material, electrons (holes) will move from the top to the bottom, creating an electric current. For a thermoelectric generator to work, three main features are required: a heat source, a heat sink, and a thermoelectric material.

Unfortunately, current thermoelectric devices are poorly used for power-generation applications. Reasons such as low heat-to-electricity conversion, toxicity, and expensive materials are some of the major obstacles that face thermoelectric devices [74], [75], [76]. The most common material used for thermoelectric devices is bulk bismuth telluride Bi₂Te₃ and its alloys. Materials based on Bi₂Te₃ have thermoelectric figure of merit (ZT) values ranging between 0.5 and 1 [76], [77]. Thermoelectric generators based on Bi₂Te₃ suffer from low conversion efficiency and complex fabrication techniques. Nevertheless, researchers have been trying to fabricate materials with ZT >1 [78]. In principle, this is a challenging task since ZT=S²σT/κ, where S is the Seebeck coefficient, σ is the electrical conductivity, T is the absolute temperature, and κ is the thermal conductivity. It is essential to have high electrical conductivity with low thermal conductivity, in order to have high ZT. However, it is quite an obstacle to obtain materials with these specifications.

At the nanoscale, due to low dimensionality, which causes the confinement of the lattice temperature (heat), it is predicted that material thermal properties should be substantially enhanced [76], [79]. For instance, black phosphorus and phosphorene (monolayer black phosphorus) exhibit low thermal conductivity with high electrical conductivity in the armchair direction, which are two important thermoelectric properties that are hard to find. Black phosphorus has thermopower (S) as high as 2000 μV/k for monolayer, with thermal conductivity as low as 18 (W/m-K), and electrical conductivity in the order of 10³ S/m [80], [81]. Using these parameters, several studies have found that black phosphorus exhibit a thermoelectric figure of merit (ZT) ranging between 1 and 2.5 for monolayers [81], [82]. In fact, due to its recent discovery, and according to a recent study on the electron transport of phosphorene, it was found that the thermoelectric properties can be dramatically degraded if the electron-phonon coupling effect was not taken into account. The modified ZT drops to 0.15 at room temperature [83]. Thus, we believe concrete and reliable experimental studies on the fundamental thermoelectric properties of phosphorene and black phosphorus are crucial.

Another recently investigated 2D material is MoS₂. Buscema et al. measured the Seebeck coefficient of monolayer MoS₂, and found it to fall in the range between 4×10² and 1×10⁵ [84], which is significantly larger than that of bulk MoS₂. This large Seebeck coefficient can be tuned by applying a vertical electric field across the flake, as in the case of gate voltage. Yan et al. also measured the thermal conductivity of monolayer MoS₂ using the Raman spectroscopy technique [85]. The obtained thermal conductivity value was 34 (W/m-K). This value is much less than the simulated values by approximately twofold. Accordingly, the thermoelectric figure of merit of monolayers MoS₂ ranges between 0.24 and 1.3. We should mention that calculated ZT values assume ballistic transport while neglecting the various scattering mechanisms in each material.

In Table 1, we summarize the properties of several 2D materials. We should note that there is significant discrepancy between the values reported for the thermoelectric properties of each material. Most of the values in the table are based on calculations instead of experimental measurements, where certain assumptions are made (i.e. ballistic electron transport). This can be a major source of error in obtaining the actual thermoelectric properties. Moreover, thermoelectric calculations ignore the doping technique, which renders the calculated ZT values unrealistic if the doping cannot be realized. Lastly, the contact resistance at the thermoelectric material-electrode interface is ignored, which imposes another barrier to overcome in order to obtain high heat-to-electricity conversion efficiency.

In Figure 7C and D, we plot the electrical conductivity vs. thermal conductivity and the thermopower vs. thermal conductivity using values obtained from literature, respectively. We also plot the power factor (PF=S²σ) in Figure 7E. The arrows in these figures indicate the desired direction for an ideal thermoelectric material. Accordingly, phosphorene (armchair) and MoS₂ exhibit high thermoelectric properties, compared to Bi₂Te₃, although the thermal conductivity of Bi₂Te₃ is one of the lowest compared to other materials. While graphene exhibit one of the highest electrical conductivities, it suffers from very high thermal conductivity, which renders graphene thermoelectric devices impractical. Other 2D materials plotted in the figures also exhibit noticeably good thermoelectric properties. Thus, we believe 2D materials, such as phosphorene and MoS₂, can be potential candidates for thermoelectric device applications.

To this end, we can estimate the thermoelectric efficiency based on some of the reported ZT values in Table 1. The simplified efficiency is given by [75], [101]

η=ΔTThot1+ZT-11+ZT+TcoldThot,

where η is the device efficiency;T_hot and T_cold are the hot and cold temperatures at both ends of the material, respectively; and ΔT is the difference between T_hot and T_cold. In Figure 7F, we plot the estimated efficiencies for some of phosphorene and MoS₂ reported ZT values. Of course, the plotted curves are estimates only and assume constant ZT across the calculated temperature range, which can be a source of error in finding the actual efficiencies [102]. It is worth mentioning that to obtain high ZT values for monolayer flakes, robust studies on the doping concentration should be investigated. One common doping method is electrostatic gating, as in the case of Figure 6C; however, this method requires a voltage source, which is not efficient for energy-generation applications. Chemical doping to obtain high ZT, however, is still another room of research.