In_xGa_1–xAs alloy is currently the most mature material system for photodetection at telecommunications wavelengths. In_xGa_1–xAs alloy is a variable band gap III–V semiconductor with excellent electronic transport and optical properties. The band gap of In_xGa_1–xAs can be tuned for wavelengths between 0.85 – 3.6 μm, making it ideal for near-infrared photodetection [24]. The development of In_xGa_1–xAs fabrication technology with various PD structures, including p-doped-intrinsic-n-doped (p – i – n), metal–semiconductor–metal (MSM) and avalanche photodetector (APD), will be briefly reviewed.

The In_xGa_1–xAs ternary system is attractive for its wide energy gap, ranging from 0.35 eV for pure InAs to 1.43 eV for pure GaAs. By simply changing the alloy composition, the photodetection responsivity can be maximized at the desired wavelength to enhance the signal-to-noise ratio [25, 26]. Especially for the In_0.53Ga_0.47As (E_g = 0.73 eV) alloy, which gives a complete lattice-match to the InP substrate, and outstanding photodetection performance for wavelengths between 1.0 – 1.7 μm [10]. Low background doping level and high carrier mobility have also been achieved for this system at room temperature to achieve good signal-to-noise-ratio (SNR) under high-speed operation [27]. In optical communication applications, PD based on In_xGa_1–xAs currently outperforms Si and Ge owing to its direct band gap properties and relatively large absorption coefficient, making it capable of covering both the C and L bands (see Figure 1) [28, 29], compared to Si and Ge that only cover the C band [1].

However, it is difficult to integrate In_xGa_1–xAs with CMOS circuits due to the large lattice mismatch between the III–V material and Si substrates, as InP/Si has an 8.1% mismatch. Over the past few decades, a lot of researches have been done to resolve the compatibility issue. The easiest way to integrate III–V with the CMOS technology uses epitaxial growth directly on an SOI substrate, but this leads to a large strain and high dislocation density, therein reducing the performance and reliability of the PD [30]. Lateral overgrowth [31] and an intermediate buffer layer [32] are also used to improve the crystal quality, but the performance is unsatisfactory. Later, flip-chip integration is used to mate the surface of the detector with the SOI substrate to improve performance [33]. The major drawback of this technique is that the integration process is carried out on a wafer level, which is time consuming and expensive. Today, a heterogeneous integration approach using wafer-bonding process is commonly adapted to obtain a defect-free III–V thin film that can be lithographically integrated with the Si substrate [34–37]. With modern metal-organic chemical vapor deposition (CVD) technology or molecular beam epitaxy, the In_xGa_1–xAs absorption layer can even be selectively grown on Si substrate [38].

Different types of PD structures based on the In_xGa_1–xAs ternary system have been developed to operate at telecommunications wavelength [4, 27]. The classic type is the p – i – n structure shown in Figure 2(a), which relies on both carrier drift and diffusion to generate photocurrents. Effenberger et al. proposed the use of a dual-depletion region to improve the device speed of the InGaAs/InP p–i–n detectors, with large device dimensions and high dark currents as a tradeoff [39]. Later, through evanescent-coupling of the p – i – n detector with the Si waveguide, a high responsivity of 1.1 A/W at 1550 nm was reported, with a low dark current of 10 pA and high QE of over 90% [34]. An alternative Si-waveguide butt-coupled PD with a 9 GHz 3 dB-bandwidth at –4 V bias and a high detection speed of 10 Gb/s has been demonstrated [38]. Furthermore, coupling the p – i – n PD with a high-Q cavity resonator could reduce the device dimension while maintaining high QE. Eighty-four percent QE was achieved by using a 200 Å InGaAs absorption layer [24, 40, 41].

MSM PDs consist of interdigitated Schottky metal contacts on top of an absorption layer, which are compatible with the current field-effect transistor technology as shown in Figure 2(b). The MSM PDs generally have higher speed and lower dark currents than p-i-n PDs due to its low capacitance, but suffer from smaller responsivity [42]. Initially, thin strained layers of GaAs and AlGaAs were used to create a higher Schottky barrier on InGaAs, which resulted in high leakage currents and low gains due to a significant trap density [43, 44]. Later, thin layers of undoped InAlAs were used for barrier enhancement, which significantly improved the device performance due to the latticematching between InAlAs and the InP substrate [45, 46]. Kim et al. reported a back-illuminated InGaAs MSM PD with a record responsivity of 0.96 A/W and 92% QE at 5 V bias, although the bandwidth is only 4 GHz [47]. Coupling with a cavity resonator can increase the bandwidth up to 100 GHz [48]. Recently, with flip-chip bonding of InGaAs MSM diode on the Si substrate, a responsivity of 0.74 A/W at 1550 nm and 1 μA dark current under low applied bias (∼1 V) has been achieved [49]. Cheng et al. demonstrated an InGaAs MSM PD that is monolithically integrated with an InP waveguide that fully covers the C and L bands with a nearly flat wavelength-dependent responsivity [50].

APDs are p-n junction photodiodes shown in Figure 2(c), operated at high electric fields in order to achieve photocurrent gains of > 1 [51], where one incident photon can generate more than one electron-hole pair [52]. In high speed long distance optical communication systems, APD is normally the best choice owing to its large internal gain, which provides sensitivity of 5 – 10 dB higher than the p – i – n photodiode [53]. Larger bandwidth and higher performance have been reported at 1300 – 1550 nm for various In_xGa_1–xAs APD structures, such as separated absorption grading charge multiplication [54], δ-doped separated absorption grading multiplication [55], InAlAs/InGaAs super-lattice structures [56], separated absorption charge multiplication structure with quantumdot resonant cavity [12], and floating guard ring structures [57, 58]. Using narrow InGaAs multiplication layer

Figure 2:

(a) The cross-section of an InGaAs p – i – n photodetector integrated with an SOI waveguide. (b) The cross-section of a waveguide integrated InGaAs MSM photodetector. (c) Schematic cross-section of InP/InGaAsP/InGaAs SACM APD with double-diffused floating guard ring configuration. Figures with reprint permission from: (a) Ref. [34] ©2010 OSA, (b) Ref. [35] ©2007 IEEE, and (c) Ref. [53] ©2007 IEEE.

Figure 3:

The effect of tensile strain on the band structure of Ge. (a) Schematic of how the band diagram changes as biaxial tensile strain is applied. (b) Plot of the band gap energies for the Γ and L bands as a function of tensile strain. Figure with reprint permission from Ref. [62] ©2010 Nature publishing group.

widths, a very low dark current of 10 – 300 nA with a 3 dB-bandwidth of 80 GHz was achieved [59]. Recent efforts focused on extending the capability of InGaAs APDs for error-free, high-speed modern communication (∼50 Gb/s) as well as single photon detection systems [60, 61].

Germanium (Ge) is another photodetection material system at the telecommunications wavelength. Today, fully CMOS-integrated Ge PDs are commercially available and shared the market with InGaAs PDs. In terms of material properties, Ge is more suitable for photodetection than InGaAs. Although Ge is an indirect band gap semiconductor, it has a small band gap of ∼0.8 eV, hence granting Ge the capability for photodetection at telecommunications wavelength without band gap engineering. On the other hand, the lattice mismatch between Ge and Si is 4.2% that is relatively smaller than that of InGaAs/Si, making Ge a better choice to be integrated with the CMOS circuit.

In order to develop high-performance Ge-on-Si PDs, the 4.2% lattice mismatch between Ge and Si has to be reduced. This lattice mismatch causes high surface roughness and threading dislocation densities, which degrades the device performance [62]. In 1984, Luryi et al. first obtained high quality epitaxial Ge-on-Si using a graded SiGe buffer layer to reduce the threading dislocation density [63]. This method is further improved by adjusting the Ge composition, growth temperature and using an additional ex situ chemical-mechanical polish step [64–66], and as a result, a low threading dislocation density of < 10⁶ cm^–2 has been achieved [66]. Alternatively, a twostep epitaxial growth technique has also been developed to directly grow Ge on Si without using the buffer layer, and a similar dislocation density ∼10⁶ – 10⁷ cm^–2 has been obtained [67, 68]. In addition, this two-step method can introduce tensile strain to the Ge layer during fabrication, which could transform Ge from an indirect to a direct band gap material (see Figure 3), hence greatly improving the optoelectronic properties of Ge [69–73]. Other techniques such as H₂ annealing are also adopted to improve the Geon-Si quality [74–76].

Ge PDs have structures that similar to InGaAs PDs, including the p – i – n, MSM and APD. The early Ge p – i – n PDs used a SiGe buffer layer which suffer from high dark currents (∼0.2 – 0.3 μA) due to large thermionic emission even from high quality Ge crystals [65]. Later, Liu et al. discovered that a thin Ge layer of the order of 1 – 2 μm could establish a strong built-in electric field, enough for carriers to easily overcome the recombination process at the lattice defects, hence reducing the overall dark currents and significantly improved the device sensitivity over a large bandwidth [77]. An internal QE of above 90% in the wavelength range of 1250 – 1340 nm was reported at 0 V bias as shown in Figure 4(a) [77]. Coupling the device with a resonant cavity could extend the photodetection wavelength to 1538 nm with a moderate QE of 59% [78]. Coupling the photodiode to a waveguide could further improve the responsivity and bandwidth while maintaining a small dark current. A responsivity larger than 1 A/W was achieved [79, 80], with a bandwidth > 30 GHz [80, 81], and a low dark current density 0.2 nA [82]. Recently, with the help of GeSn active layers,high-performance Ge p–i–n PDs are fabricated to extend the photodetection to the longer wavelengths up to 1800 nm [83, 84].

The performance of Ge MSM PDs is initially limited by the high dark currents due the low hole Schottky barrier height as a result of Fermi level pinning near the valence band edge. The dark currents can be larger than 150 μA even after coupling the photodiode with a waveguide [79, 85]. Efforts were made to modulate the Schottky barrier height in order to suppress the dark currents. Oh et al. used an amorphous Ge layer to increase the Schottky barrier height [86] and suppressed the dark currents by two orders of magnitude. Similarly, Liah et al. used an amorphous Si layer and reduced the dark currents by three orders of magnitude [87]. Recently, Ang et al. utilized a Si:C layer as shown in Figure 4(b) and reported an extremely low dark current of 11.5 nA with a high responsivity of 0.76 A/W and a moderate QE ∼60% [88, 89]. Besides manipulating the Schottky barrier, an alternative method of reducing the dark currents is through the adoption of the dopant-segregation technique. For example, implanting and segregating the As and boron/sulfur dopants can suppress the total dark currents by 3–4 orders of magnitude as shown in Figure 4(c) [90, 91]. With this modern fabrication technology, the responsivity of Ge MSM PDs had been boosted to 1.76 A/W at 5 V bias [92].

Ge-on-Si APDs combined the excellent optical absorption of Ge with the outstanding carrier-multiplication properties of Si. Since Si has a much smaller ionization ratio k-value of ∼0.1 than III–V materials such as InP with k-value of ∼0.5, the Ge-on-Si APD is believed to be better than III–V APDs in the high-speed optical communication systems in terms of higher gain bandwidth product and better sensitivity [62]. With a simple Ge_xSi_1–x/Si strainedlayer structure, the multiplication factor can be larger than 50 [93, 94]. Kang et al. achieved a gain-bandwidth product (Figure 4(d)) of 340 GHz with a small k-value of 0.09, and a sensitivity of –28 dBm at 10 Gb/s [19]. Assefa et al. manipulated the optical and electrical field within the Ge layer, and achieved 70% noise reduction with a high photodetection speed of over 30 GHz at only 1.5 V bias [95]. With state-of-the-art fabrication technology, the multiplication gain has been boosted to 680 at 8 V bias [96], the speed increased up to 43 Gb/s at 1550 nm with bit-error rate 10^–12[97]. Meanwhile, the input power can be reduced to as low as –35 dBm [98].

Si crystal has a relatively large indirect band gap ∼1.1 eV, corresponding to a cut-off wavelength below 1100 nm [99]. As such, Si is not the most suitable material for PDs operating at the telecommunications wavelength compared to the direct band gap III–V alloy such as InGaAs or Ge which has a smaller band gap. Nevertheless, all-Si PDs have the advantage of monolithic integration with the standard CMOS technology without changing the fabrication process, and hence could significantly reduce the fabrication costs [7]. In this section, a number of all-Si PDs operating at the telecommunications wavelength based on different physical mechanisms, including mid-band gap absorption (MBA), surface-state absorption (SSA), internal photon emission (IPE) and two-photon absorption (TPA), will be reviewed.

MBA PDs are developed based on the fact that high energy particles could introduce defect states located within the band gap of the intrinsic Si crystal, thus enabling detection of sub-band gap optical radiation. Fan et al. first discovered in 1959 that the damaged Si could interact with sub-band gap illumination to generate photocurrents [100]. Later, the MBA mechanism was shown to exist for both donor and acceptor defects [101, 102], where di-vacancies and interstitial clusters were the two

Figure 4:

(a) Responsivity of Ge p-i-n PD at 0 V and 2 V reverse bias in wavelength range of 650 – 1650 nm. The whole C band and a large part of L band is covered even at 0 V. (b) Schematics of the Ge MSM PD with a Si:C layer to enhance the Schottky barrier. (c) Schematics of the Ge MSM PD implanted with segregating sulfur dopant. (d) A 340 GHz gain-bandwidth Ge APD. Figure with reprint permission from: (a) Ref. [77] ©2005 American Institute of Physics, (b) Ref. [88] ©2008 IEEE, (c) Ref. [91] ©2008 IEEE, and (d) Ref. [19] ©2009 Macmillan Publishers Limited.

Figure 5:

(a) A monolithic p – i – n all-Si photodetector based on the MBA mechanism. The defect states are introduced by ion implantation. (b) The SSA photodetector fabricated by Baehr-Jones et al. [116]. (c) Schematic of the IPE mechanism. (d) Schematic of the TPA mechanism. Figures with reprint permission from: (a) Ref. [105] ©2006 American Vacuum Society; (b) Ref. [116] ©2008 Optical Society of American; (c) Ref. [121] ©2008 Institute of Physics.

primary defect types contributing to the sub-band gap photon absorption [103]. Knight et al. first proposed an all-Si MBA p – i – n photodiode with the defect states introduced by ion-implantation [104], and optimized its performance by post-annealing and exploiting helium ion as the implantation source [105, 106] (Figure 5(a)). Further, adding an external ring resonator cavity could increase the total optical absorption and achieve a complete C + L band coverage [107, 108]. With modern fabrication technology, Geis et al. were able to reduce the transversal cross-section of the waveguide to increase the absorption length and reduce carrier recombination, making the performance of the MBA PDs comparable with the III–V alloy and Ge PDs. They demonstrated a responsivity of 0.5–10 A/W at 1550 nm with a 60 V bias, and a bandwidth of > 35 GHz [109]. An extremely small bit error rate < 10^–12 is also reported [110]. Note that besides ion implantation, defect states can also be generated in Si by roughening the surface using ultrafast laser pulses [111, 112].

SSA PDs are based on a similar principle with the MBA PDs, where surface states are introduced into the band gap of the intrinsic Si, extending the optical absorption to longer wavelengths. It has been demonstrated that a large number of surface states are present for a clean Si or Ge surface and vanish as the surface becomes rough or oxidized [113–115]. By increasing the overlap between the optical mode and the waveguide surface using a small conduction arm, Baehr-Jones et al. first built the SSA PD (Figure 5(b)) in ultrahigh vacuum environment [116]. They reported a responsivity of 36 mA/W at 1575 nm under 11 V bias, and a corresponding dark current of 12 μA. Further improvements can be done by coupling the SSA PD with an external resonator cavity. A responsivity of 0.12 mA/W was achieved even at 0 V bias [117]. Ideally, surface states should have larger carrier mobilities and less recombination sites compared to the bulk defect states in the MBA PDs, resulting in higher detection speed and larger QE [118]. However, SSA is very sensitive to the Si surface conditions as well as the external temperature [114, 117], which requires a delicate fabrication process that hinders mass production. Presently, the SSA PD is still in its infancy.

IPE PDs rely on the principle that photo-excited electrons in metal can gain energy higher than the Schottky barrier and subsequently move into the conduction band of the semiconductor, as shown in Figure 5(c). The IPE PDs are widely implemented in infrared imaging systems due to the high switching speed and convenience in fabrication [119]. In optical communications, however, it is not favored due to the very low QE as the photo-excited electrons do not have enough energy or momentum to overcome the Schottky barrier [120, 121]. Elabd et al. first proposed to use ultra-thin metal films (∼2 nm) to increase the carrier escape-probability of the photo-excited electron [122]. A responsivity of 250 mA/W at 1500 nm was achieved with a trade-off in low total optical absorption. Later, metalporous silicon structures were exploited to improve the QE to ∼14.4% by increasing the surface-to-volume ratio [123]. Recently, it has been demonstrated that the QE can be significantly improved by coupling the IPE PD with a Si waveguide [124]. A responsivity of 4.6 mA/W at 1550 nm was reported with –1 V bias. The corresponding dark current is fairly low at 3 nA [125, 126]. Although current IPE PDs are limited by small bandwidths (∼3 GHz in Ref. [127]), it has the potential for high-speed communications due to the fast photoemission mechanism.

TPA PDs are based on the nonlinear TPA process as shown in Figure 5(d). Boggess et al. first discovered that an electron can absorb two photons (having individual energies below the semiconductor band gap) approximately at the same time, and reach the excited state in the conduction band [128]. Reintjes et al. first measured the TPA coefficient β = 1.5 cm/GW using 1060 nm picosecond laser pulse [129]. Later, β = 0.8 cm/GW was demonstrated at telecommunications wavelength (∼1500 nm) [130, 131]. Although the nonlinear behavior of the TPA offers the potential in ultra-short laser pulse detection, the small TPA coefficient hinders applications in optical communications. Coupling the TPA PDs with Si waveguide [132] or a high-Q resonant cavity can improve the responsivity to 16 mA/W at 1500 nm [133, 134]. However, it is limited by the low detection speed ∼0.1 GHz due to the high resistivity of the device [134]. With the continuous progress in the TPA APD development, the weak effect of the TPA is significantly amplified in the carrier multiplication process and the sensitivity is improved [135–137]. All-Si TPA PDs are also proposed as profilometers at both 1300 and 1550 nm in detecting ultra-short pulse [138, 139]. At the current stage, the research of the TPA PD is still focused on amplifying the TPA effect to improve the responsivity and speed.

In terms of material properties, HgCdTe is probably the most promising semiconductor for infrared photodetection in the spectrum of 0.7 ∼ 25 μm. HgCdTe was first synthesized in 1959 by Lawson et al. [140]. At present, we are already into the development of the third-generation HgCdTe PD products [10].

The HgCdTe system is believed to be suitable for infrared photodetection due to the following reasons [141]:

• –

  HgCdTe has an adjustable band gap, achieved through tuning the composition of each element, thereby granting the flexibility in spectral response over a wide infrared spectrum;

• –

  HgCdTe is a direct band gap semiconductor with high photon absorption coefficient even for a thin film [142], which leads to high QE and low noise generation (see Figure 6(a));

• –

  High lattice-matching (∼0.2% mismatch) between HgCdTe tertiary alloy and the CdZnTe substrate ensures high-quality crystal growth. In fact, it is the only material that has nearly the same lattice constant regardless of the composition of each element; and

• –

  Moderate dielectric permittivity ensures small device capacitance, and low thermal expansion coefficient ensures device stability.

The first and second generations of the HgCdTe PDs consist of a linear array of photoconductors and a twodimensional (2D) array of photovoltaic diodes, which are mainly used for thermal imaging at far infrared wavelengths (> 8 μm) [13, 141]. Only in the third generation of HgCdTe PDs, especially during the development of the APD (see Figure 6(a)), the potential of HgCdTe in nearinfrared photodetection is explored. For HgCdTe with a band gap of around 0.9 eV (corresponding to 1370 nm), the energy required to excite an electron from the valence band to the conduction band is identical to the energy to excite a carrier from the split-off band to the valence band (Figure 6(b)). This property provides the HgCdTe APDs a competitive advantage in the asymmetry between the avalanche gain of holes and electrons in the multiplication process. An ionization ratio k-value of 0.1 or less can be achieved, resulting in excellent SNR [143], and a figureof-merit better than that for the InGaAs and Ge APDs. At present, the HgCdTe APDs for photodetection at 1060, 1300, and 1550 nm have all been fabricated using liquid phase epitaxy or molecular beam epitaxy [144, 145]. The responsivity can reach 13.1 A/W with dark currents of 66 nA at 77.7 V bias, and the gain can be larger than 100 at room temperature [141].

Despite the outstanding performance, the HgCdTe APDs faced serious problems in mass production due to the weak Hg–Te bond as well as safety concerns for the highly toxic compounds. Also, large bulk areas and high operating voltages significantly increase power consumption of the APDs. In addition, high quality HgCdTe is usually grown on the CdZnTe substrate, which is difficult to integrate with the silicon readout circuit due to different thermal expansion coefficients and a 19% lattice mismatch [146, 147]. Finally, the fabrication cost associated with the CdZnTe substrate is also much higher than Si and Ge.

Since its first isolation in 2004, graphene is rapidly becoming an appealing material for photonics and optoelectronics due to its ultra-high bandwidth photodetection capability [14]. Graphene’s ultrafast carrier mobility, tunable optical properties via electrostatic doping, low dissipation and good stability make it an ideal platform for developing high speed PDs for optical communications [148]. In addition, graphene is compatible with CMOS technology, making it convenient for low-cost and large-scale integration into the CMOS circuits [149–151].

The band structure of a single layer graphene can be calculated with the tight-binding description, where the conduction and valence band meet at the six Dirac points of the Brillouin zone as shown in Figure 7(a) [15]. Near the Dirac point, the band diagram possesses linear energymomentum dispersion. Due to this peculiar dispersion property, the electron in graphene behaves like a massless Dirac fermion, with high carrier mobility at ∼1/300 speed of light. Besides the excellent carrier dynamics, the gapless band structure makes graphene capable of broadband photodetection from UV to even terahertz region [152]. High optical absorption coefficient of 7 × 10⁵ cm^–1 has been found from 300 to 2500 nm, which is much higher than conventional semiconductors [14].

On the other hand, due to the short interaction length of the atomic thin layer, the optical absorption of a single layer graphene is only 2.3% in visible and infrared region [153]. The opacity of graphene is solely determined by the fine structure constant α = e²/hc, which is a result of its two-dimensional nature and gapless band spectrum. Such a small absorption rate, although superior for an atomic thin-layered structure, is insufficient for high performance photodetection. Many researches have been performed to improve the optical absorption rate of the graphene [152]. It is found that the optical absorption of graphene layers is linearly proportional to the number of layers (see Figure 7(b)) [153, 154]. The total absorption of the multi-layer graphene can be treated equivalently as a superposition of each individual layer due to the weak van der Waals force between the adjacent layers. This property offers a convenient way to tune the absorption of the graphene PD without generating much noise. Other approaches such as integrating with other optical structures (e.g. plasmonic structure [155, 156], optical waveguide [149, 151] and optical cavity [157]), or utilizing the intrinsic plasmonic properties of graphene by patterning the graphene into nanostructure [158], or implementing hybrid material systems (e.g. polymer, nanoparticle, QD) to increase the light harvesting and facilitate the exciton separation [159], have also been reported in the literature [148, 152].

Generally, the physical mechanisms in the photodetection of graphene can be characterized into three major categories: photoelectric, photo-thermoelectric and photo-bolometric effects [152].

• –

  Photoelectric effect: graphene typically aligns lon-

Figure 8:

(a) A typical graphene p-n diode for photodetection and (b) its magnified version showing an interlayer between metallic p- and n-graphene (p-G and n-G) layers. For the electrodes, Ag was used. Figures with reprint permission from: (a) and (b) Ref. [176] ©2014 Macmillan Publishers Limited.

gitudinally between two electrodes, whereby the photo-excited electrons form exciton are separated and propelled under external bias to create photocurrent [160–162]. For the area far away from the metal–graphene junction, the photoconductive effect dominates the photoresponse, where the graphene behaves like a conventional semiconductor [163]. On the other hand, for the area beneath and close to metal–graphene junction, the photoresponse is dominated by the photovoltaic effect [164], where the Fermi energy of graphene is shifted according to the work function of the metal contact [165] and subsequently a built-in voltage can be established to separate excitons without an external bias [166]. Currently, most of the graphene PDs are fabricated based on this mechanism;

• –

  Photo-thermoelectric effect: Due to the strong electron–electron interaction, a photo-excited electron–hole pair can result in ultrafast heating of the carriers in graphene [167, 168]. During the long transport length for the hot carrier to interact, multiple excitations can further occur [167, 169]. By forming inhomogeneous thermoelectric power across the graphene channel, these hot carriers can produce a built-in bias based on Seebeck effect for a graphene interface junction [170] or a dual-gate graphene [171] to generate photocurrent;

• –

  Photo-bolometric effect: The bolometric effect is associated with the light-induced change in the conductance via optical heating [172], instead of direct photocurrent generation. The temperaturedependent conductance can be attributed to the change in both carrier mobility and carrier density. Yan et al. [173] explicitly demonstrated this effect on a uniform graphene without using graphene–metal junction to exclude the contribution from photoelectric and photo-thermoelectric effect. Freitag et al. [163] further distinguished the photoelectric, photo-thermal, and bolometric effect on a single layer graphene at room temperature.

Note that besides the three major mechanisms, other mechanisms may be responsible for graphene photodetection, including photogating, tunneling, and plasma-waveassisted mechanism. Readers can refer to Ref. [148, 152] for more detailed information.

The first graphene PD was reported by Xia et al., which is a transistor-based PD made out of single or double graphene layers, demonstrating an ultrafast photodetection of up to 40 GHz [174]. Mueller et al. have developed an interdigitated metal–graphene–metal structure to break the mirror symmetry of the internal electric field in the conventional field-effect transistor channel [175]. This device operated at 1550 nm with a responsivity of 6.1 mA/W. Coupling the PD with a microcavity resonator increased the total absorption to 40%, but with small bandwidth as a tradeoff [157]. Konstantatos et al. covered the graphene layer with a thin film of colloidal QDs to improve the device gain to 10⁸ with a responsivity of 10⁷ A/W, but the response speed was limited [159]. Later, Gan et al. developed an evanescently-coupled graphene-waveguide structure and achieved a responsivity of > 0.1 A/W, a uniform response between 1450 and 1590 nm, and high speed of 12 Gb/s [149]. Alternatively, Wang et al. reported a responsivity of 0.13 A/W at 1.5 V bias with a graphene/Siheterostructure waveguide PD [151]. Kim et al. further improved the responsivity to 0.4 ∼ 1.0 A/W in a broad spectral range from ultraviolet to near-infrared using an allgraphene p-n vertical junction structure as shown in Figure 8. The performance of such device was consistent under 6 months operations [176]. A similar structure was also demonstrated by Liu et al., showing a responsivity of > 1 A/W from visible to mid-infrared wavelengths at room temperature [177].

Note that our primary focus in this review has been on PDs in the telecommunications wavelength. For graphene, however, efforts are also spent on developing PDs in the terahertz (e.g. graphene-based bolometers, terahertz detectors), which is not achievable by other conventional semiconductors. Interested readers can refer to the recent reviews for more details [148, 152].

CNTs, first discovered by Iijima et al. in 1991 [178], are nearly ideal 1D systems with diameters of a few nanometers and lengths that be scaled up to centimeters. By controlling the arrangement of the carbon atom structure with respect to their axis (e.g. indices (n, m) [16]), CNTs could be direct band gap semiconductors, or even metals with nearly ballistic conduction [18]. The band gap of the CNT is inversely proportional to its diameter (the band gap ranges from 0.2 to 1.5 eV for CNT diameter from 0.8 to 3 nm), which allows CNT photodetection from ultraviolet to infrared [17, 179]. Besides the tunable band gap, CNT intrinsically has high absorption coefficient 10⁴ – 10⁵ cm^–1 in infrared region, which is one order of magnitude larger than the conventional semiconductors [180].

Photoexcitation creates an electron in the conduction band, leaving a hole in the valence band. For the bulk semiconductors, the electron–hole interaction is negligible and they behave like free carriers [181]. On the other hand, for low-dimensional materials, especially CNTs [182, 183] and QDs [184], substantial columbic interaction due to the strong quantum confinement between the electron and hole “quasi-particles” binds them into excitons [185], offering an alternative valley for charge transfer in the lowdimensional materials. Excitons can be efficiently formed in CNTs under photoexcitation due to the 1D confinement and subsequently decay into free electrons and holes. Photocurrent can be then produced via external bias [186] or photovoltage can be created by the build-in field near Schottky barrier or p-n junctions [187, 188]. The exciton binding energy is also inversely proportional to the diameter of the CNT, and can be further tuned by adjusting the electric field or the dielectric constant of the environment [189, 190]. In addition, carrier multiplication or ro-

Figure 9:

(a) Schematic of CNT/organic heterojunction and (b) Film morphology. Figures with reprint permission from: (a) and (b) Ref. [196] ©2009 American Chemical Society.

bust exciton is found to be effective in the CNT due to impact excitation [191, 192] with picosecond response being reported [193], thereby promising attractiveness in photodetection.

High-performance infrared PDs based on CNTs are usually fabricated with molecular beam epitaxial or CVD, and the photodetection mechanism is attributed to thermal or photo effects. For the thermal effect, the current is generated as a result of the change in conductance due to the temperature change under light illumination. St-Antoine et al. demonstrated a thermopile using single-walled CNT (SWCNT) thin film with a detectivity of 2 × 10⁶ cm·Hz^1/2/W in the visible and near-infrared region [194]. Lu et al. showed a bolometer with a higher detectivity 3.3 × 10⁶ cm·Hz^1/2/W at 1000 – 1300 nm using a multiwall-CNT (MWCNT) [195]. These bolometric devices are affected by the surrounding medium and are usually made in air or vacuum to improve the sensitivity. For the photo effect, the electrical signal is created based on the photoexcited exciton. Arnold et al. reported a detectivity > 10¹⁰ cm·Hz^1/2/W at 400 – 1500 nm with 3 dB bandwidth of 31 MHz using a CNTs/C₆₀ heterojunction (see Figure 9) [196]. Lu et al. achieved a detectivity of 2.3 × 10⁸ cm·Hz^1/2/W with a P3HT/CNT heterojunction from 1000 – 1300 nm [197]. Controlling the charity and uniformity of CNT may further improve the device performance. More details can be found in recent reviews [18, 179].

Besides the graphene and CNT, other low-dimensional materials have been reported with photodetection capabilities, including transition metal chalcogenide (TMDC), semiconducting nanowire, and QD.

Similar to graphene, the layered TMDCs (e.g. MoS₂, GaS, WS₂, GaTe, GaSe), also inherit the two-dimensional quantum limit. The exfoliated versions of TMDCs offer layer-dependent properties, which are complementary to yet distinct from those in graphene [198]. For example, TMDCs generally undergo an indirect- to directband gap transformation with decreasing layer numbers, which is essentially a result of quantum confinement and change in the atomic orbital hybridization [199]. Such band gap shift leads to the changes in the photoconductivity. Moreover, excitons are created in TMDCs due to the 2D quantum confinement and band splitting, which changes the absorption spectra [200]. TMDC-based PDs have been demonstrated in the literature using photovoltaic or photo-thermoelectric mechanism [201, 202]. However, the band gaps of most TMDC range from 1.1 – 1.9 eV (< 1130 nm) [203], which are not suitable for telecommunication applications. More details about TMDC PDs can be found in Ref. [198, 204].

Semiconducting nanowires such as Si, InGaN, InP, CdS etc, or “nanowhisker” were invented in 1990s [205], and have been actively investigated for photonic applications [206]. Similar to CNT, the optical and electronic properties of nanowires can be controlled by adjusting the sizes, dimensions, elementary compositions and morphology, for example, InGaN nanowire has a tunable band gap from UV to near infrared [207]. The large surface to volume ratio and high density of surface states enable photodetection with high sensitivity [208]. So far, nanowire PDs in various configurations (e.g. ohmic contact, Schottky contact, p-n dipole, APD, core-shell nanowire, etc) with the responsivity > 3000 A/W have been achieved [209]. Despite the excellent photodetection performance, the band gaps of semiconducting nanowires are usually larger than 1 eV [208, 210]. Therefore, the nanowire PDs are frequently operating in UV-visible region instead of infrared telecommunication region. Readers may refer to Ref. [209, 211] for more details.

QDs (e.g. InAs, GaAs, PbS, PbSe, CdSe etc) are the zero-dimensional nanostructures that can be synthesized from the solution phase, which can even outperform the epitaxial counterparts. Benefitting from the quantum confinement, the band gap and band offset of the QD depends highly on the size, shape and composition of the nanocrystal, offering a great tunability from visible to even long wavelength infrared regime [212, 213]. Meanwhile, the confinement makes the charge carriers of the QD occupy discrete energy levels and experience many quantum effects such as delocalization, variable range hopping and slow carrier relaxation [214]. Multiexciton generation with long exciton lifetime is also found to be significant in the QD that can be exploited to improve the photodetection performance [215–217]. QD PDs operating in telecommunication wavelength region have been established with either photodiode structures with fast responses [218–220], or photoconductor structures with high responsivity [221, 222]. The maximum responsivity can be larger than 2700 A/W with a dark current ∼0.7 mA/cm² at 100 V bias for PdS QDs [221]. However, the main trade-off of QD PDs is the relatively small bandwidth (typically megahertz) due to its size-confinement. More detailed review on QD PDs can be found in Ref. [213, 214, 223].

Waveguide PDs are common devices to convert optical to electronic signals by coupling light from waveguides to semiconductors. To increase speed and reduce footprint, its sensitivity needs to be enhanced. Plasmonic resonances have been studied to improve the performance of the waveguide PDs. The most direct approach is to exploit the electrodes of the PDs to excite surface plasmon resonance. Ren et al. have introduced plasmons by adding thin Al interdigitated electrodes on top of a Ge-based PD (Figure 10) [20]. The plasmons generated by the Al electrodes considerably enhanced the coupling of light from the silicon waveguide into the PD, enabling the use of smaller devices and hence higher speed. The resulting strong field intensities permeates the Ge active region, enabling high absorption under TM-mode injection. Measured results showed the TM-mode photocurrent is three times more than the TE mode that did not have plasmonic enhancement. The responsivity was measured to be 1.081 A/W at 1550 nm.

Plasmonics has also been used to enhance the performance of the Si waveguide IPE PDs (see Section 2.3), which normally consist of a metal-semiconductor interface where the Schottky barrier is formed. Plasmonic resonances can dramatically enhance the electromagnetic fields near the metal surface and increase its total absorption, resulting in an increased photocurrent. Levy et al. experimentally demonstrated an on-chip nanoscale silicon surface-plasmon IPE PD operating at the telecommunica-

Figure 11:

(a) SEM micrograph of the Schottky contact of an on-chip nanoscale silicon surface-plasmon Schottky photodetector. (b) Schematic illustration of an active optical photodetector consisting of an Au resonant antenna on a n-type silicon substrate. Figures with reprint permission from: (a) Ref. [21] ©2011 American Chemical Society, and (b) Ref. [224] ©2011 AAAS.

tions wavelength (Figure 11(a)) [21]. The fabricated PDs showed enhanced detection capability for shorter wavelengths that is attributed to the increased probability of the internal photoemission (IPE) process enhanced by the plasmon resonance. The responsivity of 13.3 mA/W has been measured at 1310 nm. In addition, the PD can be easily integrated with other Si photonic components. Halas et al. reported an active optical PD consisting of an array of independent rectangular gold nanorods (Figure 11(b)), which supported both longitudinal and transverse plasmon resonances [224], creating hot electron-hole pairs for photocurrents. The experiment measured photocurrent responsivity up to 9 μA/W at the telecommunications wavelength with the nanoantenna dimension of 50 nm (W) × 30 nm (T) × 158 nm (L).

Optical antennas are key devices in the focusing of light from free space or waveguides into ultra-small, nanometer-scale volumes. Plasmonic antennas have recently been used to enhance the performance of the PDs.

Figure 12:

(a) Top view of the open-sleeve dipole antenna consisting of a dipole antenna oriented in the y direction and two line electrodes in the x direction. (b) Cross-section of the germanium nanowire lying under the two line electrodes. (c) Cross-section showing germanium in the gap region between the two antenna arms. Figures with reprint permission from: (a), (b) and (c) Ref. [23] ©2010 Nature Publishing Group.

Figure 13:

Schematics of a plasmonic waveguide dipole antenna photodetector, consisting of two L-shaped metallic nanorods, which are used to form a nanoantenna, a nanocavity as well as a halfwavelength waveguide to efficiently couple the electromagnetic fields from the plasmonic waveguide to the detector.

Typically, the antennas are placed close to the active region of the PDs to confine optical near fields into a subwavelength volume [22]. Subwavelength MSM PDs using a plasmonic dipole antenna have been demonstrated as shown in Figure 12[23]. The gold dipole antenna was used to collect and concentrate light into a subwavelength germanium nanowire between the two dipole arms. The detector active region has an ultra-small volume of 150 nm × 60 nm × 80 nm, on the order of 10^–4 of the operating wavelength at 1310 – 1480 nm, much smaller than the traditional semiconductor PDs. An enhancement by a factor of 20 in the photocurrent has been achieved due to the plasmonic antenna resonance.

This concept was extended to plasmonic waveguide PDs, which could be a promising component for the nextgeneration chip-scale integration as it can both confine and guide optical signals in subwavelength structures. Placing a dipole antenna at the end of a metal–insulator–metal (MIM) waveguide allows effective coupling of the optical power from the waveguide to the antenna. The electromagnetic fields in the gap between two antenna

Figure 14:

(a) Schematics of a monopole antenna waveguide photodetector, whose nanocavity is formed by coupling a plasmonic hybrid waveguide with a monopole antenna. (b) Simulated power flow intensity inside the waveguide and the nanocavity. (c) The power flow inside a photodetector with dipole antenna is also shown, where “leaky power” in the gap lowers the coupling efficiency. Figures with reprint permission from Ref. [231] ©2009 OSA.

arms can be enhanced more than 100 times [226]. By filling semiconductor materials (e.g. Si, GaAs, Ge, etc) in the gap, a high performance PD can be easily built. Figure 13 shows a plasmonic MIM waveguide and a MSM PD consisting of a dipole antenna formed by two L-shaped metallic nanorods [227]. The L-shaped nanorods function as both a dipole nanoantenna and a half-wavelength nanocavity to efficiently couple and confine the electromagnetic fields from the plasmonic waveguide. This detector has a very small active volume (50 × 50 × 50 nm), and hence a small footprint and potentially THz detection speed [227]. However, optical power coupling is not entirely efficient due to power leakage arising from the gap in between the dipole antenna and the end of the MIM waveguide.

Among all plasmonic waveguides, hybrid plasmonic waveguides are the most suitable in intra-chip optic data transmission due to its excellent confinement and relatively low propagation loss [228–230]. The hybrid waveguide consists of a metal–insulator–semiconductor structure, which supports the design of a monopole antenna, which is more efficient than the dipole antenna configuration. The monopole is essentially a half-dipole mounted on a ground plane. If the ground plane is large enough, the electromagnetic waves reflected from it will seem to come from an image antenna that forms the missing half of the dipole. Figure 14(a) shows a hybrid plasmonic waveguide PD consisting of a monopole antenna [231]. The waveguide

Figure 15:

(a) Schematic of an MIM optical energy converter. Surface plasmon enhanced hot electrons in the top electrodes can largely increase the forward current and hence a net DC photocurrent is generated. (b) TEM image of quantum plasmonic tunnel junctions made of two silver nanoparticles bridged by a self-assembled monolayer in which CTP has been measured. Figures with reprint permission from: (a) Ref. [234] ©2011 American Chemical Society, and (b) Ref. [241] ©2014 AAAS.

is formed by sandwiching a thin layer of silica between an aluminum slab and a silicon nanowire. The metallic part of the waveguide serves as the conducting ground of the monopole antenna. The antenna feed can thus be placed at the terminal and subsequently the monopole antenna on top of it to form a nanocavity or the active volume of the PD. Simulation results show that 42% of optical power is absorbed in the active volume with dimensions 150 (L) × 60 (W) × 220 nm (H), which is fairly lager than the 27% power absorption with a dipole antenna [232]. Figure 14(b)– 14(c) shows the power flow intensity inside the waveguide and the nanocavity of the PDs with both the monopole and dipole antennas, respectively. The optical coupling was further improved using coupled cavities, and 78% absorption efficiency was achieved [233]. In addition, the monopole antenna is easier to fabricate compared to the dipole antenna, owing to the former’s gapless and monolithic built.

Unlike most PDs relying on semiconductors, plasmons could provide a direct conversion between photons and electrons, and thus plasmonic PDs could be built without using semiconductors. Melosh et al. have designed a purely plasmonic PD by using surface plasmon excitations in a simple MIM structure similar to a rectenna as shown in Figure 15(a) [234]. When light illuminates the top of the MIM structure, plasmons excited on the topmost metal layer can create a high concentration of hot electrons. If these hot electrons have enough energy, they can tunnel through the thin insulating barrier to produce photocurrents. Without plasmon excitation, the efficiency of this hot carrier tunneling is low and cannot be considered for photodetection.

When the MIM structure is replaced by a nanoparticle dimer separated by a subnanometer gap, charge transfer plasmon (CTP) could be excited [235–241]. The CTP has opened up new opportunities in nanoscale optoelectronics, single-molecule sensing and nonlinear optics [242]. Quantum mechanical effects become important when the two nanoparticles are placed so closely that electrons can tunnel across the gap. Theoretical studies showed that the CTP can be excited when the vacuum gap is in the scale of 0.3 – 0.5 nm [237, 238]. Recently, the CTP has also been experimentally observed at a large gap up to 1.3 nm, with a self-assembled molecular layer filling the gap as shown in Figure 15(b) [241]. The measured and simulated results showed that the frequency of the tunneling currents is approximately 200 THz and can be tuned by changing the molecules. This technology can be directly used for ultra-high speed PDs [243]. This is revolutionary considering that traditional semiconductor optoelectronic detectors can only operate in frequencies limited to tens of GHz. As these PDs also have nanometer footprints, they are promising PDs for next-generation optoelectronic circuits, with operation speeds over tens of thousands times higher than our familiar microprocessors.