Advances in High–Speed, High–Power Photodiodes: From Fundamentals to Applications

Abstract

High–speed, high–power photodiodes play a key role in wireless communication systems for the generation of millimeter wave (MMW) and terahertz (THz) waves based on photonics–based techniques. Uni–traveling–photodiode (UTC–PD) is an excellent candidate, not only meeting the above–mentioned requirements of broadband (3 GHz~1 THz) and high–frequency operation, but also exhibiting the high output power over mW–level at the 300 GHz band. This paper reviews the fundamentals of high–speed, high–power photodiodes, mirror–reflected photodiodes, microstructure photodiodes, photodiode–integrated devices, the related equivalent circuits, and design considerations. Those characteristics of photodiodes and the related photonic–based devices are analyzed and reviewed with comparisons in detail, which provides a new path for these devices with applications in short–range wireless communications in 6G and beyond.

1. Introduction

High data rate and high–quality transmission, large–capacity data storage, wireless short–range links, sensing, and imaging bring new challenges for current communication systems. Parts of the of microwave (MW), millimeter wave (MMW), and terahertz (THz) bands, which cover the frequency from 20 GHz to 3 THz, as shown in Figure 1, provide an appropriate choice for the above–mentioned challenges [1,2,3,4]. The exciting MMW and THz bands (20 GHz~1 THz), which are located between the radio waves and light waves, combine both their qualities and advantages, including large available bandwidth, high information capacity, transparent for some kinds of materials, high–spatial resolution, smaller beam scattering than infrared and visible light, and greater safety for the human body than x–rays and gamma rays [5,6,7]. Based on the above–mentioned advantages of the frequency band, it would bring about many potential applications [7], such as high–capacity wireless links, short–range wireless communication systems, and inspection system, high–resolution imaging transmission, spectroscopy, remote gas sensing and detection, security system and astronomical radio telescope system. More–over, this frequency band would support 6G technology for a high data rate (100~100 Gbit/s), helping towards a fully connected world in the future.

However, the large atmospheric attenuation (Figure 1) [8,9] in wireless communication systems needs to be considered (i.e., the attenuation caused by the rain and fog, free–space propagation loss (FSPL), gas loss, and signal loss from fiber to chip [8,9,10,11]) with the exception of focusing on the two key figures, i.e., the data rates or information capacity and transmission distance. To compensate for the FSPL limitation, high–gain and high–output power antennas or phased array antennas for beam forming/beam steering [12,13] should be considered for both transmitter and receiver systems. For the transmitter in communication systems, using large–scale photonic integrated circuit (PIC) techniques has an unparalleled impact on the continually reducing footprint, coupling loss, low power consumption and cost, while gradually increasing stability, reliability, and functionality [14]. Photonics–based technology is not only used for the development of MMW to THz communication techniques in systems with lower loss in optical fibers, large bandwidth, and miniaturized device/system [6,7,8,9,15,16,17,18,19,20,21,22,23,24,25] but it can also reduce some drawbacks which are brought forth by electronics–based technology, including high propagation loss in coaxial cables, narrow bandwidth, incompatibility with system operations at high–frequency bands and bulky feeding methods [26,27,28,29].

In this scheme, photonic–based techniques are derived from optical to RF signal conversions utilizing the photomixing method [30]. Since there is an output power limitation using a single photomixer, the InP–based semiconductor devices have been employed and led by a uni–traveling–carrier photodetector (UTC–PD) [31,32], operating at a long fiber–optics communication band (1.31~1.55 μm), which can be used to further enhance the output power up to a mW level around 300 GHz [9]. However, the new requirements of high data rate transmission in future 6G poses another challenge for THz systems. In addition, due to the lack of appropriate power amplifiers in the THz system transmitting terminally along with the difficulty of miniaturizing the monolithically integrated system employing Si–lens, the combination of high–speed, high–power UTC–PD (or UTC–PD array) and high–gain, high–output power antenna and/or arrays would be a better choice to tackle the power limitation problem. Furthermore, the optical–to–electrical (O/E) conversion efficiency and heat dissipation at a high bias voltage also needs to be considered. We could use resonant cavity enhanced (RCE) structure [33] employing distributed Bragg reflector (DBR) [34], subwavelength grating (SWG) mirrors [35] as bottom reflectors and dielectric layers as top reflectors to increase O/E conversion efficiency, while using high conductivity AlN [36,37], diamond [38] or SiC [39] substrates to reduce heat dissipation. Moreover, the slow light effect resulting from Bragg grating structures [40,41] with a remarkably low group velocity might offer a possible and promising solution to successfully compress optical signals and enhance light–matter interactions, and the enhanced O/E conversion efficiency in PDs could be possible while the joule heat problem at a higher bias voltage, device footprint reduction, and low power consumption could also be solved in the future.

This review work is structured as follows. Section 2 presents the overall design considerations for photodiodes (PDs), including the principles of PIN–PD and UTC–PD, 3–dB bandwidth analysis and discussion, equivalent circuit models, and saturation current and RF output power. Section 3 shows the solutions of bandwidth–responsivity trade–off, which are integrated or quasi–integrated with high–reflected mirrors for III–V and group–IV photodiodes, and microstructure photodiodes with microholes for enhancing light–matter interactions. Section 4 narrates the photodiode photonic–integrated applications, which relate to the photodiodes integrated with short stubs for output power improvement as well as the integrations with planar antennas for transmitters/emitters from MMW to THz bands. The discussion and conclusions are given in Section 5 and Section 6.

2. Overall Design Considerations for Photodiodes

2.1. Principles of PIN–PD and UTC–PD

The band diagrams for PIN–PD [42,43] and UTC–PD [5,31,32,44,45] are shown in Figure 2. PIN–PD has a simple three–layer structure, i.e., wide–bandgap P– and N–layers, and a depleted absorber. Both holes and electrons in the depleted absorber of PIN–PD contribute to the response, but the output response mainly depends on the low–velocity carrier holes, whose transport velocity is an order of magnitude lower than that of electrons, which limits the photoresponse to some extent.

The UTC–PD band structure, by dividing the PIN–PD absorber into two layers, is composed of a p–type neutral absorber and a wide–bandgap (depleted) carrier collector [31,32]. By employing a bandgap grading or/and doping grading in the absorber, the quasi–field is formed to reduce the electron traveling time effectively. In the meantime, the photogenerated majority holes in the InGaAs absorber respond quickly within several picoseconds due to the high concentration of electrons, leading to a negligible effect on the photoresponse; therefore, the high–velocity electrons dominate the photoresponse of UTC–PDs. Furthermore, the velocity of electrons ($3~5\times {10}^7 \mathrm{cm}/\mathrm{s}$) is usually 6~10 bigger than that of holes ($5\times {10}^6 \mathrm{cm}/\mathrm{s}$) [5], together with the quasi–ballistic transport (overshoot velocity) [46] in the InP collector, which leads to a quite short total delay time in the whole UTC–PD. For PIN–PD and UTC–PD with the same absorber (${\mathrm{W}}_{\mathrm{a}}$), the UTC–PD (${\mathrm{W}}_{\mathrm{a}}\cong {\mathrm{W}}_{\mathrm{c}}$) represents a superior frequency response with a shorter traveling time in the collector due to the large difference in carrier velocities, though the carriers in UTC–PD travels ~three times the average distance than that in PIN–PD [44].

During high optical input, the charge intensity increases together with the increasing photocurrent in the depletion, producing a space charge effect due to the charge accumulation, while the field and potential profiles are modulated resulting in output saturation and a nonlinear response. For PIN–PD, as shown in Figure 2a, the charge storage happens at high injection conditions, resulting in the band bending and field screening, which finally causes output current saturation. However, in UTC–PD, shown in Figure 2b, only electrons participate in the space charge effect, whose velocity overshoot, even facing the decreased electrical field. Finally, the saturation occurs because of the current blocking produced from the heterojunction interface between the absorber and collector; while this state will be slower than that in PIN–PD, it is because UTC–PD can withstand a charge density an order of magnitude higher than PIN–PD [5,44]. In addition, the space charge in UTC–PD can not only be compensated by doping the spacer layer at the absorber/collector (InGaAs/InP) heterojunction interface but also can be reduced by employing a heavily n–doped InP charge (cliff) layer to increase the electrical field of the spacer layer [47]. Therefore, UTC–PD is a typical representative of high–speed response and high–output current.

In a typical UTC–PD structure, the bandgap discontinuity (the abrupt conduction band barrier) at the InGaAs absorber and InP collector interface causes the current blocking effect, thus leading to the device output current saturation. To smooth this discontinuity and suppress the blocking effect, the compositional graded quaternary system material ${\mathrm{In}}_{1-\mathrm{x}}{\mathrm{Ga}}_{\mathrm{x}}{\mathrm{As}}_{\mathrm{y}}{\mathrm{P}}_{1-\mathrm{y}}$ (for spacer layers) is usually employed to improve device performance [48]. Therefore, the design of ${\mathrm{In}}_{1-\mathrm{x}}{\mathrm{Ga}}_{\mathrm{x}}{\mathrm{As}}_{\mathrm{y}}{\mathrm{P}}_{1-\mathrm{y}}$ lattice–matched to InP [49] must be considered, while the cutoff wavelength–related refractive index and permittivity is shown in Figure 3a. However, the introduction of quaternary system material (Figure 3b) might bring certain difficulties for epitaxial growth and the device fabrication process. Therefore, a dipole–doped regime, as shown in Figure 3b is incorporated to reduce the conduction band barrier, which uses a lower dipole–doping concentration ($\le 5\times {10}^{18}{ \mathrm{cm}}^{-3}$ to avoid out–diffusion) to fully suppress the conduction band offset only left with a spike–like barrier for electrons to tunnel through [50,51,52].

2.2. 3–dB Bandwidth Analysis and Discussion

Response time is a physical quantity that characterizes how quickly a PD responds to an incident light signal, that is, it represents the total time it takes for a PD to convert an optical signal into an electrical signal.

The total response time of PD generally consists of three parts: (a) diffusion time of electrons in the absorption layer, ${\mathrm{\tau }}_{\mathrm{a}}$, (b) circuit time constant, ${\mathrm{\tau }}_{\mathrm{RC}}$, determined by junction capacitance and load resistance, and (c) drift time of electrons in the collection layer, ${\mathrm{\tau }}_{\mathrm{c}}$ [44]. PIN–PD is usually made up of parts (a) and (b), while (c) belongs to UTC–PD [44,53,54]. Therefore, the total response time for general UTC–PD can be given as

$$\mathrm{\tau }=\sqrt{{\mathrm{\tau }}_{\mathrm{a}}^2+{\mathrm{\tau }}_{\mathrm{c}}^2+{\mathrm{\tau }}_{\mathrm{RC}}^2},$$

(1)

while the PIN–PD can be dealt with as a special case, just having a single intrinsic/depletion layer, ${\mathrm{W}}_{\mathrm{a}}$. The RC–delay time–limited bandwidth, ${\mathrm{f}}_{\mathrm{RC}}$, and transit–time–limited bandwidth, ${\mathrm{f}}_{\mathrm{RC}}$, usually consist of the main 3–dB bandwidth of a PD, which here the diffusion is neglected at the high–frequency response. We assume that a PD has one single intrinsic layer (Figure 4) with the i–layer thickness of $\mathrm{d}$ and the area of $\mathrm{A}$, so the ${\mathrm{f}}_{\mathrm{RC}}$ can be expressed as

$${\mathrm{f}}_{\mathrm{RC}}=\frac{1}{2\pi \mathrm{RC}}=\frac{1}{2\pi \mathrm{R}}\frac{\mathrm{d}}{\mathrm{\epsilon }\mathrm{A}},$$

(2)

where the junction capacitance $\mathrm{C}=\mathrm{\epsilon }\mathrm{A}/\mathrm{d}$ approximately equals PD capacitance (in ideal cases), the ε is the relative permittivity of i–layer, and R is the resistance and equals the sum of diode series resistance, ${\mathrm{R}}_{\mathrm{S}}$, and the load resistance, ${\mathrm{R}}_{\mathrm{L}}$, has a typical value of 50 Ω.

It needs to be noted that Equation (2) will be changed if a PD has multiple dielectric/depletion layers and the corresponding values will be replaced by the relative permittivity, ${\mathrm{\epsilon }}_{\mathrm{req}}$, and the total depletion thickness, ${\mathrm{d}}_{\mathrm{T}}$, respectively, which are shown as follows [55]:

$${\mathrm{\epsilon }}_{\mathrm{req}}={\left[{{\displaystyle \sum }}_{\mathrm{m}=1}^{\mathrm{n}}\frac{{\mathrm{d}}_{\mathrm{m}}}{{\mathrm{d}}_{\mathrm{T}}{\mathrm{\epsilon }}_{\mathrm{rm}}}\right]}^{-1},$$

(3)

and

$${\mathrm{d}}_{\mathrm{T}}={\mathrm{d}}_1+{\mathrm{d}}_2+\cdots +{\mathrm{d}}_{\mathrm{n}}.$$

(4)

Therefore, the total capacitance, ${\mathrm{C}}_{\mathrm{T}},$ and the RC–limited 3–dB bandwidth, ${\mathrm{f}}_{3\mathrm{dB},\mathrm{RC}},$ can be rewritten as

$${\mathrm{C}}_{\mathrm{T}}=\frac{{\mathrm{\epsilon }}_0{\mathrm{\epsilon }}_{\mathrm{req}}\mathrm{A}}{{\mathrm{d}}_{\mathrm{T}}}=8.854\times {10}^{-6}\frac{{\mathrm{\epsilon }}_{\mathrm{req}}\mathrm{A}}{{\mathrm{d}}_{\mathrm{T}}}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}\left(\mathrm{pF}\right),$$

(5)

and

$${\mathrm{f}}_{3\mathrm{dB},\mathrm{RC}}=\frac{1}{2{\pi \mathrm{RC}}_{\mathrm{T}}}=\frac{1}{2\pi \mathrm{R}}\frac{{\mathrm{d}}_{\mathrm{T}}}{{\mathrm{\epsilon }}_0{\mathrm{\epsilon }}_{\mathrm{req}}\mathrm{A}},$$

(6)

where ${\mathrm{\epsilon }}_0$ is the vacuum permittivity, with a value of $8.854\times {10}^{-12}\mathrm{F}/\mathrm{m}$.

For a UTC–PD with the same thickness of the absorber and collector, assuming a pure diffusion case without quasi–field, the 3–dB bandwidth can be given as [44,54]

$${\mathrm{f}}_{3\mathrm{dB},\mathrm{tr}}\cong \frac{1}{2{\pi \mathrm{\tau }}_{\mathrm{a}}}=\frac{1}{2\pi \left({\mathrm{W}}_{\mathrm{a}}^2/3{\mathrm{D}}_{\mathrm{e}}+{\mathrm{W}}_{\mathrm{a}}/{\mathrm{v}}_{\mathrm{th}}\right)},$$

(7)

where ${\mathrm{D}}_{\mathrm{e}}$ is the electron diffusion constant, ${\mathrm{v}}_{\mathrm{th}}$ is the thermionic velocity and the related equations and values are shown in

Table 1. The explanations for parameters [44] used for the 3–dB bandwidth calculation.

Symbol	Value and Units	Explanations
${\mathrm{k}}_{\mathrm{B}}$	$1.38\times {10}^{-23} \mathrm{J}/\mathrm{K}$	Boltzmann constant
$\mathrm{q}$	$1.6\times {10}^{-19} \mathrm{C}$	Electron charge
$\mathrm{T}$	$300 \mathrm{K}$	Temperature
${\mathrm{m}}_0$	$9.11\times {10}^{-31} \mathrm{kg}$	Electron mass in vacuum
${\mathsf{\mu }}_{\mathrm{e}}$	$5000{ \mathrm{cm}}^2/\mathrm{Vs}$	Electron mobility in p–InGaAs
${\mathsf{\mu }}_{\mathrm{h}}$	150 ${\mathrm{cm}}^2/\mathrm{Vs}$	Hole mobility in p–InGaAs
${\mathrm{v}}_{\mathrm{e},\mathrm{os}},{\mathrm{v}}_{\mathrm{ns}}$	$4\times {10}^7 \mathrm{cm}/\mathrm{s}$	Electron overshoot velocity in InGaAs and InP
${\mathrm{v}}_{\mathrm{hs}}$	$4.5\times {10}^6 \mathrm{cm}/\mathrm{s}$	Hole saturation velocity
${\mathrm{D}}_{\mathrm{e}}$	$2{\mathrm{k}}_{\mathrm{B}}{\mathrm{T}\mathsf{\mu }}_{\mathrm{e}}/\mathrm{q}$	Electron diffusion constant
${\mathrm{m}}^{*}$	$0.043{\mathrm{m}}_0$	Electron effective mass at the band edge
${\mathrm{v}}_{\mathrm{th}}$	$\sqrt{2{\mathrm{k}}_{\mathrm{B}}\mathrm{T}/{\pi \mathrm{m}}^{*}}$	Thermionic–emission velocity
${\mathrm{v}}_{\mathrm{e},\mathrm{dri}}$	$~{\mathsf{\mu }}_{\mathrm{e}}\left(\varphi /{\mathrm{W}}_{\mathrm{a}}\right)$	Electron drift velocity in absorber
${\mathrm{v}}_{\mathrm{e},\mathrm{diff}}$	$~3{\mathrm{D}}_{\mathrm{e}}/{\mathrm{W}}_{\mathrm{a}}$	Electron diffusion velocity in absorber
${\mathrm{\tau }}_{\mathrm{a}}$	${\mathrm{W}}_{\mathrm{a}}^2/3{\mathrm{D}}_{\mathrm{e}}+{\mathrm{W}}_{\mathrm{a}}/{\mathrm{v}}_{\mathrm{th}}$	Carrier traveling time by electron diffusion
${\mathrm{\tau }}_{\mathrm{c}}$	${\mathrm{W}}_{\mathrm{c}}/{\mathrm{v}}_{\mathrm{e},\mathrm{os}}$	Electron drift time in collector
${\mathrm{k}}_{\mathrm{B}}$	$1.38\times {10}^{-23} \mathrm{J}/\mathrm{K}$	Boltzmann constant

.

In addition, for the electron drift/delay time–limited bandwidth in the collector of UTC–PD, ${\mathrm{f}}_{3\mathrm{dB},\mathrm{co}}$, can be considered as

$${\mathrm{f}}_{3\mathrm{dB},\mathrm{co}}\cong \frac{1}{2{\pi \mathrm{\tau }}_{\mathrm{c}}},$$

(8)

where the electron overshoot velocity in the InP collector is ${\mathrm{v}}_{\mathrm{e},\mathrm{os}}=4\times {10}^7 \mathrm{cm}/\mathrm{s}$, while the corresponding electron drift time ${\mathrm{\tau }}_{\mathrm{c}}={\mathrm{W}}_{\mathrm{c}}/{\mathrm{v}}_{\mathrm{e},\mathrm{os}}$ which is around 0.5~1 ps for a collector with a thickness of 200~400 nm. However, the electron drift time, ${\mathrm{\tau }}_{\mathrm{c}}$, usually can be ignored for a designed UTC–PD which has a similar thickness for the absorber and collector due to the relatively greater carrier traveling time (${\mathrm{\tau }}_{\mathrm{a}}$) in the absorber. The ${\mathrm{\tau }}_{\mathrm{a}}$ can be separately calculated to be 3.74 ps and 11.2 ps for the 200 nm and 600 nm absorber by [44,54]

$${\mathrm{\tau }}_{\mathrm{a}}=\frac{{\mathrm{W}}_{\mathrm{a}}^2}{3{\mathrm{D}}_{\mathrm{e}}}+\frac{{\mathrm{W}}_{\mathrm{a}}}{{\mathrm{v}}_{\mathrm{th}}}.$$

(9)

It is quite clear that we can ignore the electron drift time when the thickness of the absorber approximately equals the collector.

Therefore, the total bandwidth, ${\mathrm{f}}_{3\mathrm{dB},\mathrm{UTC}}$, as a general case, can be written as

$${\mathrm{f}}_{3\mathrm{dB},\mathrm{UTC}}=\frac{1}{2\pi \mathrm{\tau }}=\frac{1}{2\pi \sqrt{{\mathrm{\tau }}_{\mathrm{a}}^2+{\mathrm{\tau }}_{\mathrm{c}}^2+{\mathrm{\tau }}_{\mathrm{RC}}^2}},$$

(10)

while the similar values for the absorber and collector, i.e., ${\mathrm{W}}_{\mathrm{a}}\cong {\mathrm{W}}_{\mathrm{c}}$, so Equation (10) can be changed to

$${\mathrm{f}}_{3\mathrm{dB},\mathrm{UTC}}=\frac{1}{2\pi \mathrm{\tau }}\cong \frac{1}{2\pi \sqrt{{\mathrm{\tau }}_{\mathrm{a}}^2+{\mathrm{\tau }}_{\mathrm{RC}}^2}},$$

(11)

and the ${\mathrm{f}}_{3\mathrm{dB},\mathrm{UTC}}$ also has another form, as shown in Equation (12), for treating PD with a small junction area less than 10 μm².

$${\mathrm{f}}_{3\mathrm{dB},\mathrm{UTC}}=\frac{1}{2\pi \mathrm{\tau }}\cong \frac{1}{2\pi \sqrt{{\mathrm{\tau }}_{\mathrm{a}}^2}}=\frac{1}{2{\pi \mathrm{\tau }}_{\mathrm{a}}}.$$

(12)

For PIN–PD, both of the holes and electrons are in the single intrinsic/depletion layer, ${\mathrm{W}}_{\mathrm{a}}$, while the hole and electron saturation velocity are ${\mathrm{v}}_{\mathrm{hs}}$ and ${\mathrm{v}}_{\mathrm{ns}}$, so the 3–dB bandwidth for a PIN–PD with a thin absorber can be given as [43,54]

$${\mathrm{f}}_{3\mathrm{dB},\mathrm{PIN}}\approx \frac{3.5}{2{\pi \mathrm{\tau }}_{\mathrm{a}}},$$

(13)

where the ${\mathrm{\tau }}_{\mathrm{a}}$ equals ${\mathrm{W}}_{\mathrm{a}}/\overline{\mathrm{v}}$, and the average velocity formula is $1/{\overline{\mathrm{v}}}^4=0.5\left(1/{\overline{\mathrm{v}}}_{\mathrm{ns}}^4+1/{\overline{\mathrm{v}}}_{\mathrm{hs}}^4\right)$. Meanwhile, a thinner absorber can make PIN–PD obtain a larger transit–time–limited bandwidth, but it also causes a decrease in the RC–limited bandwidth by Equation (2), while the drawback can be overcome in UTC–PD, just by independently designing the absorber and collector. Therefore, a thinner absorber can be used in UTC–PD for a larger transit–time–limited bandwidth without sacrificing the RC–limited bandwidth by well–designing a relatively thick collector. Additionally, some references show the detailed theoretical and experimental analysis of the characteristics for PIN–PD and UTC–PD, including energy band, electrical field, doping mechanism, optimization of absorber and collector, bandwidth, responsivity, saturation, and output power [56,57,58,59,60,61,62] as well as transient response [63], dark current [64,65], and P–contact shapes [66]. According to the above–mentioned bandwidth equations and the values from Table 1, the 3–dB bandwidths for PIN–PD and UTC–PD can be calculated as shown in Figure 5a, including the assumed values during the calculation.

It can be seen that the bandwidth of UTC–PD is proportional to $1/{\mathrm{A}}^2$ [44,54] when the absorber thickness is larger than 100 nm, while it is proportional to $1/\mathrm{A}$ for PIN–PD [44,54]. Figure 5c is from [44] as a reference. The well–designed UTC–PD has improved 3–dB bandwidths over 150 GHz, 220 GHz [70], while the maximum value is up to 310 GHz (Figure 5b), and the 10–dB and 15–dB bandwidths are separately 750 GHz and >1 THz, but the responsivity is only 0.07 A/W [67]. To improve the low responsivity of UTC–PD, the refracting–facet UTC–PD (RF–PD) [68] and total–reflection UTC–PD (TR–PD) [69] are proposed to increase the light propagation length in the absorber, where RF–PD uses an angled edge to reflect the light while TR–PD employs a V–grooved mirror integrated with two PD to totally reflect the back–illuminated light. A high responsivity of 1 A/W with 50 GHz 3–dB bandwidth (0.32 A/W and 170 GHz) is achieved for RF–PD, while the TR–PD shows 0.83 A/W responsivity and a 58 GHz 3–dB bandwidth, as shown in Figure 5d.

Up to now, many PIN–PD and UTC–PD structures are proposed to improve the device performance, including frequency response, output power, saturation current, and responsivity, while the common band diagrams, including simple p–i–n type (Figure 2a), partially depleted absorber (PDA) type, dual depletion type, general UTC type, hybrid absorber type, charge–compensated (CC) hybrid absorber type and near–ballistic (NB)/NB–CC type [31,32,38,44,47,71,72,73,74,75,76,77,78,79,80], are summarized and shown in Figure 6.

In practice for UTC–PD, we need to choose one of them for adequate design (such as device size, bandwidth, absorber and collector layers, doping, electrical field, electron mobility, diffusivity, and diffusion time, etc.) according to the requirements and applications. We take the UTC–PD design as an example and assume three types of p–doping in the absorber with the values of $2.5\times {10}^{17}{ \mathrm{cm}}^{-3}$, $1\times {10}^{18}{ \mathrm{cm}}^{-3}$, and $2.5\times {10}^{18}{ \mathrm{cm}}^{-3}$, corresponding the electron mobility (μ_e) of 6000, 5000 and 4000 ${\mathrm{cm}}^2/\mathrm{Vs}$ [44]. Using equations shown in Part 2.2, the electron traveling time in the absorber (${\mathrm{\tau }}_{\mathrm{a}}$), the electron drift time in the collector (${\mathrm{\tau }}_{\mathrm{c}}$), and the electron diffusion constant (i.e., electron diffusivity, ${\mathrm{D}}_{\mathrm{e}}$) can be calculated as given in Figure 7a–c and summarized in Figure 7d. From Figure 7d, we find that the ${\mathrm{\tau }}_{\mathrm{a}}$ is much bigger than ${\mathrm{\tau }}_{\mathrm{c}}$ and indicates the electron diffusion time in the absorber which determines the total carrier transit time [82], so it only needs to consider ${\mathrm{\tau }}_{\mathrm{a}}$ and use Equation (7), i.e., ${\mathrm{f}}_{3\mathrm{dB},\mathrm{tr}}$, to evaluate the device’s 3–dB bandwidth at MMW and THz frequency bands when the absorber thickness is smaller than 100 μm.

2.3. Equivalent Circuit Models

The time–delay equivalent circuit model for air–bridge type waveguide PIN–PD was thoroughly analyzed by Wang et al. [84,85,86], as shown in Figure 8. Fundamentally, a basic air–bridge typed–PD is a current generator, and the corresponding equivalent circuit is composed of four parts, as shown in Figure 9, including the PD itself, the air bridge, the coplanar waveguide (CPW), and the 50 Ω load [86,87]. The photocurrent can be obtained and simulated using a current source, ${\mathrm{I}}_{\mathrm{ph}}$, which is in parallel with the junction capacitance, ${\mathrm{C}}_{\mathrm{j}}$, and the large junction shunt resistance (megohms, usually $\ge 100 \mathrm{k}\mathsf{\Omega }$), ${\mathrm{R}}_{\mathrm{j}}$, and in series with a small series resistance, ${\mathrm{R}}_{\mathrm{s}}$ (or ${\mathrm{R}}_{\mathrm{c}}$) (related to ${\mathrm{P}}^{+}$, ${\mathrm{N}}^{+}$ layers and ohmic contact resistance of metallic connection), as well as the air–bridge inductance from p–contact to p–electrode pad, ${\mathrm{L}}_{\mathrm{b}}$ (or ${\mathrm{L}}_{\mathrm{s}}$), determined by the bridge geometry. The ${\mathrm{C}}_{\mathrm{dx}}$ is the capacitance between the air bridge and the n–electrode pad. The characteristic impedance, ${\mathrm{Z}}_{\mathrm{c}}$, of the CPW, which can be measured with open and short circuit yielding a p–electrode pad capacitance, ${\mathrm{C}}_{\mathrm{p}}$ and an n–electrode pad–induced pad inductance, ${\mathrm{L}}_{\mathrm{p}}$ (or ${\mathrm{L}}_{\mathrm{G}}$), is calculated by ${\mathrm{Z}}_{\mathrm{c}}=\sqrt{{\mathrm{L}}_{\mathrm{p}}/{\mathrm{C}}_{\mathrm{p}}}$ with a common value of $50 \mathsf{\Omega }$ (e.g., ${\mathrm{L}}_{\mathrm{p}}=55 \mathrm{pH}$, ${\mathrm{C}}_{\mathrm{p}}=22 \mathrm{fF}$, so ${\mathrm{Z}}_{\mathrm{c}}=50 \mathsf{\Omega }$ in [87]). One thing to note is that some PDs do not use air bridge during the device fabrication process, so the parasitic capacitance between the p–electrode and n–electrode, ${\mathrm{C}}_{\mathrm{dx}}$ and air–bridge inductance, ${\mathrm{L}}_{\mathrm{b}}$ (or ${\mathrm{L}}_{\mathrm{s}}$) could not be used in the equivalent circuit model, as shown in Figure 10.

To evaluate the opto–microwave conversion properties of high–speed PDs by using S–parameters (such as reflection coefficients S22 and optoelectronic conversion properties S21), a small–signal radio frequency (RF) equivalent circuit for PD is reported considering synthetically both the analysis of the carrier transit effect [88] and the external parasitic resistance–capacitance (RC) delay–time constant [85]. The small–signal RF equivalent circuit is shown in Figure 11, which uses the voltage (RF voltage, ${\mathrm{V}}_{\mathrm{ac}}$)–controlled current source (VCCS, ${\mathrm{I}}_{\mathrm{ac}}={\mathrm{g}}_{\mathrm{m}}{\mathrm{V}}_{\mathrm{ac}}$, ${\mathrm{g}}_{\mathrm{m}}$ is denoted as transconductance representing optical–to–RF conversion quantum efficiency, i.e., responsivity) to replace of the current source, ${\mathrm{I}}_{\mathrm{ph}}$ in Figure 9 and Figure 10, while the overall frequency response (O–E 3–dB bandwidth) of PD is determined by both of the transit time and the RC–delay time constant, i.e., $1/{\mathrm{f}}_{3\mathrm{dB}}^2=1/{\mathrm{f}}_{\mathrm{tr}}^2+1/{\mathrm{f}}_{\mathrm{RC}}^2$. The transit–time bandwidth, ${\mathrm{f}}_{\mathrm{tr}}=1/\left(2{\pi \mathrm{\tau }}_{\mathrm{tr}}\right)$ of ac–photocurrent, ${\mathrm{I}}_{\mathrm{ac}}$ is determined by the carrier transit–time effect (${\mathrm{\tau }}_{\mathrm{tr}}={\mathrm{R}}_{\mathrm{t}}{\mathrm{C}}_{\mathrm{t}}$), which is indirectly affected by selecting the values of input circuit elements of ${\mathrm{C}}_{\mathrm{t}}$, ${\mathrm{R}}_{\mathrm{t}}$, and ${\mathrm{g}}_{\mathrm{m}}$. The signal source impedance is ${\mathrm{R}}_0$. The RC–delay time constant bandwidth can be calculated using Equation (6). In addition, the elements in an external circuit, such as in the air bridge, CPW pad (and P–contact shapes [66]), and load, also have an effect on the overall frequency response of PDs.

The similar and/or improved equivalent circuit models and RF small–signal equivalent circuit models, which are given from some specific examples for both PIN–PD and UTC–PD, with and without air bridge, can be found in previous works [25,85,86,87,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119]. Some typical improved RC–delay time models are the modified UTC–PD (MUTC−PD) (Figure 12a) [89] and the dipole–doped UTC–PD (Figure 12b) [50], which are exploited as follows. For the MUTC–PD [89] (Figure 12a) and dual–drifting layer UTC–PD (DDL–UTC–PD) [100], Li et al. focus on the analysis of the absorbing depleted region (resistance, ${\mathrm{R}}_{\mathrm{u}}$, and capacitance, ${\mathrm{C}}_{\mathrm{u}}$) and the non–absorbing depleted region (resistance, ${\mathrm{R}}_{\mathrm{j}}$, and capacitance, ${\mathrm{C}}_{\mathrm{j}}$). Based on the RF small–signal equivalent circuit model without considering the air–bridge circuit, as shown in Figure 13a, the improved equivalent circuit model of the RC–delay time is shown in Figure 13b, while the further simplified one without considering ${\mathrm{R}}_{\mathrm{u}}$ and ${\mathrm{C}}_{\mathrm{u}}$ is reported by Han et al. [90], as shown in Figure 13c. In the above–mentioned RC–delay time models, the bulk material resistance is ${\mathrm{R}}_{11}$, while ${\mathrm{R}}_{22}$ represents P– and N–contact resistance, ${\mathrm{C}}_{\mathrm{p}–\mathrm{elect}}$ is the P–electrode parasitic capacitance. Based on this model, the experimental and circuit analysis exhibit the bandwidth of an MUTC–PD up to 40 GHz, while the affected factors of the device performance, such as space charge screening, self–induced electrical field and over–shoot effects, are also discussed in detail [89].

In particular, the dipole–doped structure [50], as shown in Figure 12b, has been proposed by Wang et al. to tackle the bandgap discontinuity between the InGaAs absorber and the InP collector interface, and also to simplify the difficulty of epitaxy growth. The corresponding equivalent circuit for the RC–delay time [96] proposed by Meng et al. is shown in Figure 13d, where the resistance and capacitance for the dipole–doped part are denoted as ${\mathrm{R}}_{\mathrm{dd}}$ and ${\mathrm{C}}_{\mathrm{dd}}$. ${\mathrm{R}}_{\mathrm{aborb}}$ and ${\mathrm{R}}_{\mathrm{collect}}$ represent the resistance of the absorber and collector, while the junction capacitance is ${\mathrm{C}}_{\mathrm{j}}$. Through carefully choosing the values for circuit components, the simulated bandwidth matches the measured value very well from 10 MHz to 20 GHz and can be estimated to be ~62.5 GHz at a −5 V bias voltage.

To carefully analyze and depict UTC–PD impedance and photoresponse behavior, an accurate circuit model is proposed [99]. This model not only provides detailed explanations for inconsistent values at low frequencies (here, from DC to 30 GHz, the CPW parasitic effects are negligible) between the classical circuit models of PDs and measurements but even presents a comprehensive study for both the measured device impedance and S11 up to 110 GHz, while showing the values of frequency photoresponse up to 67 GHz. The RC–delay time of this new model is shown in Figure 13e, where two $\mathrm{RC}$ parallel circuits (${\mathrm{R}}_2{\mathrm{C}}_2$ and ${\mathrm{R}}_3{\mathrm{C}}_3$) are used to model the two spacers (or transition layers) in UTC–PD, while the ${\mathrm{R}}_4{\mathrm{C}}_4$ parallel circuit denotes the carrier collector, and the ${\mathrm{R}}_1$ represents the series resistance with a small value (related to ${\mathrm{P}}^{+}$, ${\mathrm{N}}^{+}$ layers and ohmic contact resistance of metallic connection/doped materials). In addition, the heavily p–doped neutral thick absorber usually can be ignored due to a negligible resistive effect [99]. It is noted that the parasitic capacitance, ${\mathrm{C}}_{\mathrm{p}}$, and inductance, ${\mathrm{L}}_4$ of the CPW pad cannot be ignored if the designed device photoresponse is above 75 GHz. This model provides useful and accurate information to optimize the UTC–PD complex impedance over the frequency range, thus providing a method for power maximization from UTC–PD to an antenna and then radiating to the air, by employing complex conjugate impedance matching between them [99,120]. Based on this new accurate model, some UTC–PD structures are designed and integrated with antennas [22,25,121]. In Figure 14, the matching network is employed to optimize the impedance matching to increase the RF output power, while stray capacitance is introduced in model–2 (without in model–1) to represent the parasitic effect resulting from the flip–chip bonding [25]. Compared with Figure 14a, the accurate equivalent circuit model is used for UTC–PD in Figure 14b, and the extra capacitance is also employed [22].

Moreover, by optimizing the impedance transmission line of the signal pad in CPW, a new transmission line with a characteristic impedance of 96 Ω is introduced forming a new low–inductive peaking CPW [36,37,101]. Subsequently, the charge–compensated MUTC–PD (CC–MUTC–PD) [78] is flip–bonded with this new CPW on aluminum nitride (AlN) (or diamond submounts for thermal dissipation [38,122]) die for further increasing the bandwidth and decreasing the output power roll–off at >120 GHz frequency [121]. Similarly, an inductive 115 Ω characteristic impedance CPW, as shown in Figure 15, is proposed and fabricated to improve the PD bandwidth from 28 GHz to 37.5 GHz [90], while further optimization is applied for MUTC–PD to obtain higher 3–dB bandwidth (and RF output power) from 156 GHz (−0.16 dBm @ 170 GHz [111]) up to 220 GHz (−4.94 dBm @ 230 GHz [113]). Note that some special examples with detailed parameters have been reported in the above–mentioned literatures cited in this paper, so here they are not given.

2.4. Saturation Current and Output Power

High–power PDs with a broadband frequency response are key components serving as microwave photonic antenna systems [123,124,125], phased array antennas, MMW imaging and sensing systems [5], and analog fiber–optic links wireless communications [8,126,127,128,129]. The high–speed and high–power PD is a promising candidate, such as the UTC–PD which can be monolithically integrated with a matching circuit using a coplanar waveguide short stub to achieve saturation output power, obtaining a value of −13.2 dBm (20.8 mW) at 100 GHz with a −3 V bias voltage [90].

Furthermore, narrow bandwidth high–power also serves as an analog photonic emitter and oscillator systems. The charge–compensated MUTC–PD (CC–MUTC–PD) with a cliff layer [78] is an excellent candidate to achieve high power and high frequency which are realized by the flip–chip techniques on AlN [36,37] or diamond [38] submounts, as shown in Figure 16a, for thermal dissipation. Figure 16b shows that the dissipated power of the device bonded on the diamond submount is faster than that of the devices bonded both on AlN and no bonding. Similarly, the simulated results indicate that the lowest junction temperature is produced from the device bonded on the diamond submount when using the same input power, as shown in Figure 16c.

Considering to use the well–designed high impedance CPW signal transmission line, bonding on AlN submount, a 9 μm CC–MUTC–PD with 70 GHz bandwidth achieves an RF output power of −2.6 dBm at 160 GHz [121]. To further improve the RF output power, high thermal conductivity substrates, including AlN and diamond, are employed bonding with CC–MUTC–PDs using flip–chip techniques [114], which shows 3–dB bandwidth of 105 GHz, 117 GHz, 129 GHz, and 150 GHz for PDs with diameters of 10 μm, 8 μm, 6 μm and 4 μm at a −3 V bias voltage, achieving an RF output power of −3 dBm at 150 GHz and −5.7 dBm at 165 GHz. Using the same methods for heat dissipation, the record RF output powers for gated modulation at 1 GHz and 10 GHz are separately 41.5 dBm (14.2 V) and 40 dBm (10 W) for 40 μm CC–MUTC–PD at a −36 V bias voltage [122], as shown in Figure 17.

In brief, the PD saturation current and RF output power is affected by several factors, including device area, PD types, bias voltage, fabrication quality, CPW design, junction heat with high–bias voltages, short microwave stub [131], submounts, measurements and so on, which need to be considered comprehensively. Furthermore, the superior conductivity of SiC substrate also is an excellent candidate for heat dissipation. Using the wafer–bonding technique, InP–based UTC–PD bonded on the SiC substrate [39] can increase the RF output power to a mW–level, i.e., 0.22 dBm (1.05 mW) at 300 GHz with 15.5 mA photocurrent and 2.01 dBm (1.59 mW) with the optimized bias at 18 mA photocurrent [132,133], while the output power is up to 4.04 dBm (2.53 mW) at 273 GHz with the optimized bias voltage at 18 mA photocurrent when this PD integrates with a taper slot antenna [134,135], as shown in Figure 18. The comparisons of 3–dB bandwidth, responsivity, saturation current or 1–dB compression saturation current and RF output power for different PDs [67,94,95,97,101,111,113,114,115,119,121,122,123,124,130,131,134,136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152] are shown in

Table 2. The 3–dB bandwidth (BW), saturation current ($I_{Sat}$ or $I_{ph}$), responsivity ($R$), bias voltage ($V_{bias}$) and output power ($P_{out}$) comparisons for PDs.

PD Type *	BW (GHz) **	R (A/W)	${\mathit{I}}_{\mathit{S}\mathit{a}\mathit{t}}$ (mA)	${\mathit{P}}_{\mathit{o}\mathit{u}\mathit{t}}$$\left(\mathbf{dBm}\right), {\mathit{V}}_{\mathit{b}\mathit{i}\mathit{a}\mathit{s}}$	Ref.
UTC–PD	310	0.07	NA	11 (12.59 mW) @ 100 GHz, −0.5 V	[67]
NBUTC–PD	120 (64 μm2, −3.0 V)	0.15	24.6	6.28 (4.25 mW) @ 110 GHz, −5 V	[94]
NBUTC–PD	270 (3.5 μm, −3.0 V)	0.12	17	6.12 (4.093 mW) @ 170 GHz, −2 V	[95,146]
NBUTC–PD	225 (3 μm, −1.5 V)	0.08	13	−1.74 (0.67 mW) @ 260 GHz, −2 V	[97]
CC–MUTC–PD	>110 (4 μm)	0.17	27	7.80 (6.03 mW) @ 110 GHz (6 μm), −4 V	[101]
MUTC–PD	156 (4.5 μm, −3 V)	0.165	14.6	−0.53 (0.885 mW) @ 170 GHz, −3.5 V	[111]
MUTC–PD	150 (4.5 μm, −3 V)	0.165	14.6	−0.16 (0.964 mW) @ 170 GHz, −3.5 V	[111]
MUTC–PD	200 (3.6 μm)	0.07	8.4	−2.14 (0.611 mW) @ 200 GHz, −2 V	[113]
MUTC–PD	230 (3 μm)	0.07	5.85	−4.94 (0.321 mW) @ 220 GHz, −2 V	[113]
CC–MUTC–PD	105 (10 μm, −5.0 V)	0.15	>30	14 (25.12 mW) @ 80 GHz (10 μm), −2 V	[114]
CC–MUTC–PD	117 (8 μm, −3.0 V)	0.15	>20	11.3 (13.49 mW) @ 100 GHz (8 μm), −2 V	[114]
CC–MUTC–PD	117 (8 μm, −3.0 V)	0.15	>20	8.50 (7.08 mW) @ 110 GHz (8 μm), −2.5 V	[114]
CC–MUTC–PD	150 (4 μm, −3.0 V)	0.15	8.0	$-3.0 (0.5 \mathrm{mW}) @ 150 \mathrm{GHz} (4 \mathsf{\mu }\mathrm{m}$), −2.5 V	[114]
CC–MUTC–PD	$100 (7 \mathsf{\mu }\mathrm{m}$, −1.1 V)	0.11	4.9	$-19.3 (11.75 \mathsf{\mu }\mathrm{W}$$) @ 300 \mathrm{GHz} (7 \mathsf{\mu }\mathrm{m}$), 0 V	[115]
CC–MUTC–PD	$125 (5 \mathsf{\mu }\mathrm{m}$, −1.1 V)	0.11	25.3	$-10.5 (89.13 \mathsf{\mu }\mathrm{W}$$) @ 300 \mathrm{GHz} (10 \mathsf{\mu }\mathrm{m}$), −1.8 V	[115]
WG–MUTC–PD	$165 (2\times 15 {\mathsf{\mu }\mathrm{m}}^2,$ −1.2 V)	0.312	10.6	–1.69 (0.678 mW) @ 215 GHz, −2 V	[119]
CC–MUTC–PD	$70 (9 \mathsf{\mu }\mathrm{m}$)	0.2	40	–2.60 (0.55 mW) @ 160 GHz, −5 V	[121]
CC–MUTC–PD	$13 (40 \mathsf{\mu }\mathrm{m}$, −6 V)	0.7	234 (−10 V)	41.5 (14.2 W) @ 1 GHz, −36 V	[122]
CC–MUTC–PD	$13 (40 \mathsf{\mu }\mathrm{m}$, −6 V)	0.7	234 (−10 V)	40 (10 W) @ 10 GHz, −36 V	[122]
CC–MUTC–PD	$13 (40 \mathsf{\mu }\mathrm{m}$, −6 V)	0.7	234 (−10 V)	29.8 (0.955 W) @ 15 GHz, −10 V	[122]
UTC–PD	$120 (8 \mathsf{\mu }\mathrm{m}$)	NA	21	0 (1 mW) @ 120 GHz, −3V	[123,124]
CC–MUTC–PD	$10 (50 \mathsf{\mu }\mathrm{m}$)	0.75	300	32.7 (1.86 W) @ 10 GHz, −14 V	[130]
CC–MUTC–PD	$11 (50 \mathsf{\mu }\mathrm{m}$)	0.75	228	30.1 (1.02 W) @ 10 GHz, −11 V	[131]
UTC–PD on SiC	NA	NA	15.5	0.22 (1.05 mW) 300 GHz, −1 V	[134]
UTC–PD on SiC	NA	NA	18	2.01 (1.59 mW) 300 GHz, bias optimized	[134]
PIN–PD	$1 (90 \mathsf{\mu }\mathrm{m}$)	NA	620	26.5 (0.447 W) @ 1 GHz, −5 V	[136]
PIN–PD	$>6 (34 \mathsf{\mu }\mathrm{m}$)	NA	260	26 (0.398 W) @ 6 GHz, −5 V	[136]
PIN–PD	$>5 (50 \mathsf{\mu }\mathrm{m}$)	0.55	120	25.8 (0.38 W) @ 5 GHz	[137]
PIN–PD	$7 (70 \mathsf{\mu }\mathrm{m}$)	0.6	200	29 (0.794 W) @ 5 GHz, −5.8 V	[138]
PIN–PD	$40~50 (10 \mathsf{\mu }\mathrm{m}$)	~0.7	30	14 (25.12 mW) @ 50 GHz, −2 V	[139]
MUTC–PD array	$23.8 (40 \mathsf{\mu }\mathrm{m}$)	0.438	86.8	16 (39.81 mW) @ 12 GHz, −3 V	[140]
MUTC–PD	$34 (20 \mathsf{\mu }\mathrm{m},$ −4 V)	0.41	~70	17.1 (51.29 mW) @ 30 GHz, −4 V	[141]
MUTC–PD	$~45 (15 \mathsf{\mu }\mathrm{m},$ −4 V)	0.41	38	14.6 (28.84 mW) @ 40 GHz, −4 V	[141]
WG–UTC–PD	$>67 (3\times 10 {\mathsf{\mu }\mathrm{m}}^2,$ −3 V)	0.19	13	1.1 (1.29 mW) @ 100 GHz, −3 V	[142]
TTR–UTC–PD	$>110 (3\times 20 {\mathsf{\mu }\mathrm{m}}^2$)	0.5	16	0 (1 mW) @ 110 GHz, −8 V	[143]
NBUTC–PD	$290 (64 {\mathsf{\mu }\mathrm{m}}^2$, −3.0 V)	0.15	13.6	1.85 (1.53 mW) @ 110 GHz, −3 V	[144]
NBUTC–PD	$280 (64 {\mathsf{\mu }\mathrm{m}}^2$, −3.0 V)	0.15	18	4.12 (2.58 mW) @ 110 GHz, −3 V	[144]
NBUTC–PD	$200 (64 {\mathsf{\mu }\mathrm{m}}^2$, −3.0 V)	0.15	29	8.10 (6.46 mW) @ 110 GHz, −3 V	[144]
NBUTC–PD	$180 (64 {\mathsf{\mu }\mathrm{m}}^2$, −3.0 V)	0.15	37	10.7 (11.75 mW) @ 110 GHz, −3 V	[144]
Type–II UTC–PD	$140 (3.2 \mathsf{\mu }\mathrm{m}$, 0 V)	0.09	8	$-13.9 (40.74 \mathsf{\mu }\mathrm{W}$) @ 160 GHz, 0 V	[145]
NBUTC–PD	$315 (3.5 \mathsf{\mu }\mathrm{m}$, −0.5 V)	0.1	13	0.174 (1.04 mW) @ 280 GHz, −1.2 V	[147,148]
RCE–UTC–PD	$300 (3\times 3 {\mathsf{\mu }\mathrm{m}}^2$)	0.12	9.8	–1.25 (0.75 mW) @ 300 GHz, −1 V	[149]
Two UTC–PD	70 per PD	0.17	20	0.79 (1.2 mW) @ 300 GHz, −3.9 V	[150]
Type–II UTC–PD	$330 (3 \mathsf{\mu }\mathrm{m}$, −0.5 V)	0.11	13	–3 (0.5 mW) @ 320 GHz, −1 V	[151]
MUTC–PD	~100	0.22	20	–2.7 (0.54 mW) @ 350 GHz, −2.5 V	[152]

* Waveguide–type PDs are denoted as ‘WG’, while others are vertical double–mesa PDs. ** PD diameter, area, and bias voltage are shown in brackets.

.

3. Solutions of Bandwidth–Responsivity Trade–Off

3.1. High–Reflected Mirrors for Photodiodes

High–performance PD based on III–V compound semiconductors serves as a key component and has already been demonstrated in various applications in wireless communications and radio–over–fiber and antenna systems [5]. The key parameters of PDs, which are responsivity, bandwidth, and saturation power, have been well designed for increasing one of them. For tackling the trade–off between bandwidth and responsivity in PD, several kinds of structures have been reported. The first structure is the resonant cavity enhanced (RCE) structure [33]. By selecting distributed Bragg reflectors (DBRs) as a bottom mirror in the PD, the incident light is reflected to the active layer, to achieve secondary absorption, as shown in Figure 19a. The detailed requirements for DBR design, which is also amenable to the RCE scheme, i.e., many number of different semiconductor material combinations, have been already reported in a review article [153].

Based on the reflection theory of RCE, many kinds of reflected mirror models, as shown in Figure 19b–f, which are used in vertical–mesa and waveguide types of PDs to improve the responsivity without sacrificing the bandwidth, have been proposed in the last 30 years. Most of the fabrication of reflected mirrors are the material combinations of DBRs (Figure 19b) [154,155,156,157,158,159,160,161,162,163,164,165], while periodic (or non–periodic) strip or concentric circular subwavelength gratings (SWGs), two–dimensional square bulk SWGs (Figure 19c) [140,161,166,167,168,169,170,171,172,173,174], Au mirrors [149], and even the scattering structure formed by a periodic cylindrical particle array [175], also can be employed as reflected ones. Those types of mirrors are usually either on the PD bottom (i.e., bottom mirrors) realized by direct epitaxial growth by metal–organic chemical vapor Deposition (MOCVD) or quasi–monolithically integrated using micron–level–thick benzocyclobutene (BCB) bonding (or integrated [69]) techniques, or on the top of the PD (i.e., top mirrors) implemented by employing the metal contacts (or strip gratings) by magnetic sputtering (or electron beam evaporation) or the flip–chip method with chemical etching [158,176,177,178].

Figure 20 shows PIN–PD and UTC–PDs integrated or quasi–integrated (BCB bonding) with two kinds of bottom mirrors [34,35,140,157,179] It can be seen from Figure 20a that the GaAs–based PIN–PD is integrated with 4 DBRs (three GaAs/AlGaAs DBRs and one Si/SiO₂ DBRs) forming a four–mirror and three–cavity(M4C3) structure operating at 1550 nm [157]. The fabricated M4C3 structure achieves 70% peak quantum efficiency, 36 GHz 3–dB bandwidth, and quite a narrow spectrum linewidth (full–width at half–maximum (FWHM)) of 0.75 nm which is well–suited to high–density wavelength division multiplexing (WDM) communication systems [155]. The HR–UTC–PD [34], as shown in Figure 20b, is quasi–integrated with high–reflectivity (HR) DBR mirrors, which makes the responsivity increase by 23.2% without sacrificing the 3–dB bandwidth. As depicted in Figure 20c, the FR–UTC–PD [35] is quasi–integrated with a focusing reflection (FR) mirror, which is designed with a non–periodic concentric circular high–contrast grating (NP–CC–HCG) on the silicon substrate. Based on the 84.59% FR–efficiency, the integrated device achieves 36.5% improvement for the responsivity, without altering the 3–dB bandwidth. Figure 20d shows a symmetric–connected UTC–PD array (SC–PDA) integrated with a two–dimensional (2D) SWG beam–splitter [140], where the integrated device obtains an RF output power of 16 dBm at 12 GHz with a 87.9 mA saturation current and 0.438 A/W responsivity, showing stronger high–power handling capability than that of a single photodetector (~11 dBm@12 GHz and 46.7 mA, 0.179 A/W), without complex coupling, phase matching as well as any cooling techniques. Therefore, these kinds of mirrors perfectly realize the decoupling between the responsivity and bandwidth for photodiodes, while the improvement of responsivity and/or bandwidth can be further enhanced if using the well–optimized bonding process in the future.

3.2. Microhole Design for Photodiodes

Moreover, the application of micro–holes [180,181,182,183,184,185] etched from the top until through the active regions is also an effective way to assist the light–matter interactions. Figure 21 shows PDs with different kinds of micro– or/and nano–holes in active layers for enhancing light–matter interactions to balance the frequency response and responsivity [180,184]. Figure 21(a1,a2) reports a novel design that uses the etched periodic or photon–trapping micro– and nanoholes (600~1700 nm diameters) with square (Figure 21(a3)) or hexagonal (Figure 21(a4)) shapes that penetrate from the top mesa to the bottom contact layer in the Si photodiode [180]. Such a structure reveals that the lateral propagating slow and stationary optical modes can be generated only using normal incident light, which promotes a well–rounded interaction of light with Si that is an order of magnitude greater compared with the same thick Si–film absorber while ensuring an ultrafast carrier transport. This design enables the Si–based PDs to monolithically integrate with CMOS and extends its applications for optical data links in data communications and computer networks [180].

Given the 1550 nm wavelength application, microstructure PIN–PD (MPIN–PD) is proposed and analyzed numerically by employing an InP substrate, as shown in Figure 21b [184]. The well–designed central obconical shape surrounded by a V–groove trench not only greatly facilitates the interaction of light and InGaAs absorber, but also increases the device responsivity and bandwidth simultaneously. With the optimized dimensions of the V–groove and center cone, together with the optimized circular electrodes, the efficiency and bandwidth of MPIN–PD are separately increased by 61% and 21%. In addition, the back–to–back type of PD [186], waveguide–coupled PD [187], dual–absorber PD [58,161,169] and near–ballistic PD [79] also can be used to achieve high responsivity.

With the evolution of the RCE scheme and tapered–holes structures, a novel model for vertical–mesa PD, as shown in Figure 19d, which is realized based on tapered holes and bottom mirrors, could be used to solve the trade–off between bandwidth and responsivity. For waveguide PDs, the responsivity could be improved by exploiting an SWG high–focused mirror which is made by periodic (or non–periodic) strip or concentric circular SWG bonded at the back of the device (Figure 19f), while the usage of DBR mirrors (Figure 19e) have already been simulated [188] and achieved experimentally [189] showing similar performance for the trade–off.

3.3. High–Reflected Mirrors for Group–IV Photodiodes

However, the PDs based on III–V compound semiconductors are not easy for CMOS compatibility, so the group–IV PDs are gradually paid more attention by researchers, which more easily enables electronic photonic integrated circuits (EPICs) with low–cost, compact, and large–scale integrations [190]. It is to be noticed that the Si–based PD cannot be directly used for the C–band and L–band fiber–optic telecommunication regimes due with the bandgap of 1.12 eV only resulting in a ~1107 nm cutoff wavelength.

Therefore, germanium (Ge) will play an important role for the above–mentioned telecommunication bands owing to a relatively small bandgap (less than 0.7 eV at room temperature), though its optical response rolls off rapidly beyond 1500 nm [191] while this defect can be solved by depositing the Ge–layer on the Ge–on–insulator (GOI) platform [190]. By introducing 0.16% tensile strain to the Ge active layer, the operation wavelength is extended to cover the entire range of telecommunication C– and L–bands (1530–1620 nm). The high refractive index Ge absorber with the low refractive indices of the bottom insulator and the deposited SiO₂ top layer separately serve as two mirrors which finally constitute a RCE–Ge PD on the GOI platform [190] to increase the responsivity, similar to the model shown in Figure 19b. Such RCE–Ge PD on GOI (i.e., GOI–PD) indicates that the responsivity can be greatly increased up to 0.15 A/W at 1550 nm while covering the entire C– and L–bands due to the resonant cavity effect.

Meanwhile, germanium–tin (GeSn) PD on Si is another typical group–IV candidate for SWIR applications which only needs to incorporate Sn into Ge shrinking the direct bandgap and causing redshift of the absorption edge from C–band to the mid–infrared range and beyond [192,193]. GeSn–based (2.5% Sn) PIN–RCE–PD on SOI use the buried oxide (BOX) and the deposited SiO₂ as the bottom and top mirrors forming multiple reflections to enhance light–matter interaction, which shows ~0.4 A/W responsivity at 0 V bias voltage in the C–band, which is much higher than that of the conventional GeSn PIN–PDs with the values of 0.1~0.3 A/W [194]. The introduction of 10% Sn extends the operation wavelength to a 2 μm band, where the metal–semiconductor–metal (MSM) RCE–GeSn–PD can work at 1550 nm and 2 μm wavelengths by optimizing the cavity length. Based on the RCE resonance effects, a record responsivity of 0.43 A/W is obtained with a −3 V bias voltage at 2 μm wavelength at room temperature, revealing that it is promising for CMOS–compatible photonic–integrated circuit applications covering the entire telecommunication bands in a 2 μm range [195].

4. Photodiode Photonic–Integrated Applications

4.1. Photodiodes Integrated with Short–Stubs

With the development of MMW to THz communication techniques, the photonics–based technology provides lower loss in the optical fiber and a large bandwidth while needing the PDs with a high output power and broadband high–frequency characteristics.

Furthermore, we could not ignore the connected compact rectangular waveguide (WR) output port for practical use in a package module [90,152,196,197,198,199,200]. Figure 22 shows two different equivalent circuits and the fabricated micrographs for the photonic MMW generator (PWG), where each PWG includes a UTC–PD, a short–stub matching circuit (impedance transformer), and a metal–insulator–metal (MIM) capacitor, ${\mathrm{C}}_{\mathrm{MIM}}$ (i.e., $\mathrm{C}$ in Figure 22d,e) [196,199]. The calculated relative output power curves are given for comparisons as shown in Figure 22c, in which the higher power is obtained by PMG–A at ~100 GHz (W–band). Meanwhile, the PMG works at the J–band as shown in Figure 22f, where a WR–3 waveguide output port is used for practical use. It is noticed that a typical case is a coupler integrated with UTC–PD packaged with WR–1.5, which attains a record output power of −19 dBm at 500 GHz with a 9 mA photocurrent while showing a wide range 3–dB bandwidth of 340 GHz, i.e., from 470 GHz to 810 GHz [201]. The detailed results of PDs integrated with WR ports and short stubs [90,197,198,200,201] are shown in

Table 3. The comparisons of PDs integrated with WR ports and short stubs.

Device Type	Coupling Type	Si Lens	${\mathit{I}}_{\mathit{p}\mathit{h}}$ (mA)	Output Power (dBm)	Ref.
UTC–PD	WR10 port (W–band)	No	25	$13.2 (20.8 \mathrm{mW}$) @ 100 GHz	[90]
UTC–PD	WR3 port (J–band)	No	26	$-8.73 (134 \mathsf{\mu }\mathrm{W}$) @ 264 GHz	[190]
UTC–PD	WR8 port (F–band)	No	25	$12.3 (17 \mathrm{mW}$) @ 120 GHz	[197,198]
UTC–PD	WR6 port (D–band)	No	7	$>10 (>10 \mathrm{mW}$) @ 104~185 GHz	[200]
Coupler integrated UTC–PD	WR1.5 port	No	9	$-18.96 (12.7 \mathsf{\mu }\mathrm{W}$) @ 500 GHz	[201]
Coupler integrated UTC–PD	WR1.5 port	No	9	$>-22 (>6.31 \mathsf{\mu }\mathrm{W}$) @ 470~810 GHz	[201]
Coupler integrated UTC–PD	WR1.5 port	No	6	$-22 (6.31 \mathsf{\mu }\mathrm{W}$) @ 500 GHz	[201]
Coupler integrated UTC–PD	WR1.5 port	No	6	$>-25 (>3.16 \mathsf{\mu }\mathrm{W}$) @ 470~810 GHz	[201]

.

4.2. Photodiodes Integrated with Antennas

Another photonic–integrated device is PD with antennas for increasing output power for practical use. The general schematic diagram of the photonics–based transmitter and receiver is shown in Figure 23. One of the unique performances of UTC–PDs [44,54] is both a large bandwidth and high saturation output power, which can be used as an MMW signal emitter [123] for wireless applications.

To obtain V–band–to–G–band (50 GHz~220 GHz) photonically driven emitters and extend applications to MMW and THz signal wireless links, high–power photonic–integrated emitters with three types of planar antennas have been proposed [22,202], as shown in Figure 24. The high–speed and high–power broadband CC–MUTC–PD [78,101] with 0.2 A/W responsivity and 12 dBm output power is used for the MMW and THz emitter design. The fabricated emitter with the sinuous antenna and 8 μm PD shows a record output power of 20 dBm at 90 GHz with a photocurrent of 20 mA while achieving an output power of 9.9 dBm at 70 GHz, 11.5 dBm at 100 GHz, all with 7 mA photocurrent, for log–periodic and spiral antennas, respectively [22].

Using flip–chip technology, UTC–PD can be mounted on a planar slot antenna on the Si substrate [123,124,203], as shown in Figure 25a, and then boned on a hemispherical Si–lens to collimate the MMW signal direction, which generates a larger than 0.5 mW output power from the PD chip while radiating a bigger than 0.3 mW power into the free space, at a 120 GHz bandwidth [123,124]. Meanwhile, this design indicates the potential applications in future wireless communication, sensing, radar, measurement, and imaging systems based on photonic–assisted MMW generation and transmission techniques. Beyond the F–band, the electrical transmission line will bring loss and reflection, so the miniaturized PD–integrated antenna with a quasi–optical configuration (Figure 25b) is necessary for sub–MMW signal handling [204]. As shown in Figure 25c, a UTC–PD integrated with a log–periodic antenna achieves a saturation output power of 300 μW at a photocurrent of 20 mA (Figure 25d) [205]. Using a similar configuration, the UTC–PD integrated with a log–periodic antenna exhibits a high output power of 2.6 μW at 1.04 THz with a saturation current of 13 mA [204].

However, a single PD integrated with an antenna suffers some limitations, including the limited emitted power resulting from the saturation of PD and the limited antenna directivity due to the finite effective aperture of the planar antenna [206]. Here, four InP–based photodiodes operating at 300 GHz, which are integrated with a linear bowtie antenna array mounting on the Si–lens, are proposed, as shown in Figure 26, to increase the radiation power. Compared to the straight radiation of a single emitter, the array emitter realizes a 10.6 dB higher power (theoretical gain: 12 dB), while the 3–dB beamwidth is reduced by 8.5° and up to 22.5°, which achieves the output power increase and higher directivity. These photonic–integrated antenna arrays provide new solutions to wireless communications beyond 5G and will further exploit combinations of other photonic devices for application to future complex MMW and THz transceivers.

In addition, a resonant antenna integrated with UTC–PD is another excellent candidate to be employed for output power radiation in the THz bands [207,208,209]. Figure 27a shows the THz UTC–PD chip integrated with a twin–diploe antenna [209] using an impedance matching circuit [90] to enlarge the output power and the resonant behaviors are found exhibiting the peak output powers of 5.9, 3.8, and 1.1 μW at 0.78, 1.04 and 1.53 THz (Figure 27b), while generating a record output power of 10.9 μW at 1 THz with a photocurrent of 14 mA [209].

Moreover, the THz wave and THz beam steering provide great opportunities for free–space applications, including imaging, sensing, radar, and wireless communications in the 6G era and beyond [15,135,210]. Figure 27c shows a 4 × 1 UTC–PD array integrated with a 4×4 unidirectional slot antenna to generate THz signals based on coupled fiber beats techniques [211]. The adaptive genetic algorithm is used to control the produced THz beam steering and the experiment achieves a precise pointing towards the intended direction of the 300 GHz beam, which greatly improves the THz power efficiency by beam steering and paves the way for THz wireless communication using large–scale UTC–PD array [211,212,213].

Figure 27. (a) THz UTC–PD chip integrated with twin–diploe antenna using impedance matching circuit and (b) the output power versus different frequency [209]. (c) Adaptive THz beam steering in fiber–coupled UTC–PD array chip to generate THz wave [211,212,213].

Figure 27. (a) THz UTC–PD chip integrated with twin–diploe antenna using impedance matching circuit and (b) the output power versus different frequency [209]. (c) Adaptive THz beam steering in fiber–coupled UTC–PD array chip to generate THz wave [211,212,213].

With the development of communication techniques, the frequency range (MMW to THz: 30 GHz~3 THz) between microwave and far–infrared is set to meet the growing demand. Exploiting and making the most of the MMW and THz wireless signals [7,214,215] with high–radiated power have long been studied and generated by the electronic system [26,27,28] or photonic system [6,8,9,15,16,17,18,19,20,21,22,23] or hybrid electronic–photonic systems [216,217], which have shown a very diverse range of application in information and communications, security screening, food and agriculture, biometrics, medical and medicine, drug inspection, semiconductor wafer inspection, and air pollution [218,219,220].

Given the photonic–generated THz signal, ultrafast broadband, and high–power PDs, such as UTC–PDs [44,54] and traveling–wave PD (TW–UTC–PD) [221], are necessary. By combining the TW–PD structure for high responsivity and the UTC–PD structure for high output power, a new TW–UTC–PD is proposed for integration with antennas for MMW and THz signal radiation [21,222]. The TW–UTC–PD separately integrates with the resonant, bowtie, and log–periodic antennas for generating THz signals, covering frequencies up to 1 THz. The radiated power at different photocurrent levels is shown in Figure 28, where the 4 × 15 μm² PD integrated with a resonant antenna achieves an RF output power of 148 μW at 457 GHz and 24 μW at 914 GHz with a 10 mA photocurrent (Figure 28a). Figure 28b shows that the RF output power of 105 μW at 255 GHz and 10 μW at 612 GHz with a 13 mA photocurrent are attained by a 2.5 × 50μm² PD integrated with a bowtie antenna while exhibiting 70 μW at 150 GHz with a 9 mA photocurrent for integrating with a log–periodic antenna.

Another type of PD for THz applications is related to the design, usage, and optimization of the hybrid absorber UTC–PD (i.e., modified UTC–PD, MUTC–PD) [72], as shown in Figure 29. The band diagram for this type of MUTC–PD is shown in Figure 29a, where the absorber consists of a non–depleted absorber with p–doping and a depleted absorber without doping. The maximum 3–dB bandwidth can be calculated while just comprehensively analyzing the numerical relation of thickness among depletion (${\mathrm{W}}_{\mathrm{dep}}$), non–depleted (${\mathrm{W}}_{\mathrm{an}}$), and depleted (${\mathrm{W}}_{\mathrm{ad}}$) absorbers. Figure 29d shows the frequency response at different quasi–fields in the non–depleted absorber, while the ${\mathrm{W}}_{\mathrm{an}},$ ${\mathrm{W}}_{\mathrm{ad}}$ and ${\mathrm{W}}_{\mathrm{dep}}$ are separately 0.12 μm, 0.06 μm, and 0.18 μm. The 3–dB bandwidth is enlarged from 603 to 1087 GHz with the increase in quasi–field from 10 to 30 kV/cm.

The MUTC–PD is integrated with a bowtie antenna shown in Figure 29b, where Si–lens (Figure 29c) is employed to increase the output power. Being accompanied by a well–designed matching circuit and the rectangular waveguide output port, the quasi–optical antenna–integrated photomixer module can be completed as shown in the inset of Figure 29e. It can be seen that the measured output power gradually decreases with the increase in the frequency while the typical values are larger than −20 dBm at 300 GHz, −30 dBm at 1 THz, and approximately −40 dBm at 2 THz. Furthermore, the calculated results which changed with the different effective electron velocities are also shown in Figure 20e, as comparisons with the measured values. Parts of PDs photonic–integrated antennas [18,21,22,25,72,91,98,123,124,134,142,204,205,206,208,209,223,224,225,226,227] are summarized in

Table 4. The comparisons of emitted power for PDs integrated with antennas.

Device Type *	Coupling Type	Si Lens	${\mathit{I}}_{\mathit{p}\mathit{h}}$ (mA)	Output Power (dBm)	Ref.
NBUTC–PD	Yagi antenna	No	30	–14.1 (25.7 mW) @ 100 GHz	[18,223]
TW–UTC–PD	Log–periodic antenna	No	9	$-11.55 (70 \mathsf{\mu }\mathrm{W}$) @ 150 GHz	[21]
TW–UTC–PD	Bowtie antenna	No	13	$-9.79 (105 \mathsf{\mu }\mathrm{W}$) @ 255 GHz	[21]
TW–UTC–PD	Resonant antenna	No	10	$-8.3 (148 \mathsf{\mu }\mathrm{W}$) @ 457 GHz	[21]
TW–UTC–PD	Bowtie antenna	No	13	$-20 (10 \mathsf{\mu }\mathrm{W}$) @ 612 GHz	[21]
TW–UTC–PD	Resonant antenna	No	10	$-16.2 (24 \mathsf{\mu }\mathrm{W}$) @ 914 GHz	[21]
CC–MUTC–PD	Log–periodic antenna	Yes	7	9.9 (9.77 mW) @ 70 GHz	[22]
CC–MUTC–PD	Sinuous antenna	Yes	7	11.5 (14.13 mW) @ 90 GHz	[22]
CC–MUTC–PD	Sinuous antenna	Yes	20	20 (100 mW) @ 90 GHz	[22]
CC–MUTC–PD	Spiral antenna	Yes	7	8.8 (7.59 mW) @ 100 GHz	[22]
MUTC–PD	Vivaldi antennas	No	10	5 (3.2 mW) @ 110 GHz	[25]
Hybrid absorber UTC–PD	Bowtie antenna	Yes	6	$-20 (10 \mathsf{\mu }\mathrm{W}$) @ 300 GHz	[72]
Hybrid absorber UTC–PD	Bowtie antenna	Yes	6	$-30 (1 \mathsf{\mu }\mathrm{W}$) @ 1 THz	[72]
Hybrid absorber UTC–PD	Bowtie antenna	Yes	6	$-35.2 (0.3 \mathsf{\mu }\mathrm{W}$) @ 1.5 THz	[72]
Hybrid absorber UTC–PD	Bowtie antenna	Yes	6	$-40 (0.1 \mathsf{\mu }\mathrm{W}$) @ 2 THz	[72]
Hybrid absorber UTC–PD	Bowtie antenna	Yes	6	$-44 (0.04 \mathsf{\mu }\mathrm{W}$) @ 2.5 THz	[72]
MUTC–PD	Patch antenna	No	45	20 (100 mW) @ 60 GHz	[98]
TWPD	Bowtie antenna	Yes	6.3	$-33 (0.5 \mathsf{\mu }\mathrm{W}$) @ 40 GHz	[91]
UTC–PD	Slot antenna	Yes	>20	–5.23 (0.3 mW) @ 120 GHz	[123,124]
UTC–PD on SiC	Tapered slot antenna	No	14.5	1.59 (1.44 mW) @ 273 GHz, −1 V	[134]
UTC–PD on SiC	Tapered slot antenna	No	18	4.04 (2.53 mW) @ 273 GHz, bias optimized	[134]
WG–UTC–PD	Bowtie antenna	Yes	13.5	−12.22 (60 μW) @ 250 GHz	[142]
UTC–PD	Log–periodic antenna	Yes	13	−25.9 (2.6 μW) @ 1.04 THz	[204]
UTC–PD	Log–periodic antenna	Yes	20	−5.23 (300 μW) @ 300 GHz	[205]
PIN–PD array	Bowtie antenna	Yes	NA	>10 dB vs. an emitter @ 300 GHz	[206]
UTC–PD	Log–periodic antenna	Yes	15	−11.2 (75 μW) @ 414 GHz	[208]
UTC–PD	Log–periodic antenna	Yes	15	−17.4 (18 μW) @ 609 GHz	[208]
UTC–PD	Log–periodic antenna	Yes	15	−20.8 (8.3 μW) @ 804 GHz	[208]
UTC–PD	Twin dipole antenna	Yes	14	−19.6 (10.9 μW) @ 1.04 THz	[209]
UTC–PD	Log–periodic antenna	Yes	20	−9.2 (120 μW) @ 300 GHz	[224]
Type–II UTC–PD	Dual–ridge horn antenna	No	NA	−15 (31.6 μW) @ 240 GHz	[225]
UTC–PD	TEM–horn antenna	Yes	2.75	−29.5 (1.13 μW) @ 940 GHz	[226]
UTC–PD	TEM–horn antenna	Yes	2.5	−33.4 (0.46 μW) @ 1.04 THz	[226]
UTC–PD	TEM–horn antenna	Yes	2.88	−35.1 (0.31 μW) @ 1.365 THz	[226]
UTC–PD	slot antenna	Yes	10	−15.53 (28 μW) @ 700 GHz	[227]
UTC–PD	slot antenna	Yes	10	−24.56 (3.5 μW) @ 1.25 THz	[227]

* Waveguide– and traveling–wave–type PDs are denoted as ‘WGPD’ and TWPD, while others are vertical–mesa PDs.

.

5. Discussion

During the design of high–speed and high–output power photodiodes, the compatible photodiode–integrated devices, the measurement setups, and applications for communication systems, there are lots of considerations and challenges that need to be noticed. They are mainly reflected as follows: (1) The efficient optical coupling and precise alignment between the photodiode and other optical components. (2) The wider bandwidth with high response for photodiodes. (3) The wider dynamic range and linearity for higher output power without small saturation or signal distortion. (4) The effects resulting from temperature and heat from high–power conditions. (5) The reliability and robustness of photodiodes, and the scalability and modularity of photodiode–integrated devices. (6) The potential limitations at the 300 GHz band, such as the coverage range with high propagation loss, the penetrating obstacles from the walls, buildings, and vegetation, and the complexity and cost of exploiting and deploying photodiodes to the MMW and THz bands. (7) The design of corresponding other components matching with photodiodes in THz links and systems.

Though achieving high–speed, high–power photodiodes and photodiode–integrated devices with over 2 mW output power in the THz band has been realized, it is still a big challenge to apply photodiodes to 6G technology. Compared with existing communication systems, 6G requires ultra–high data rates (such as 200~1000 Gbit/s) and also requires the matching terahertz sources, photodiodes, and components in THz links and systems. However, considering the wider applications of photodiodes in transparent materials, including cell phone screens, smart windows, and automobile/aircraft win screens, is also a big challenge. A possible solution could be to implement the integration of photodiodes with waveguide–based Bragg grating devices in glass inscribed using femtosecond laser in the future.

6. Conclusions

This paper reviews the advances in high–speed, high–power photodiodes from fundamentals to applications. (1) For the overall design and consideration of photodiodes, this paper discusses the principles of PIN–PD and UTC–PD, while reporting the detailed calculation of the 3–dB bandwidth for the two typical photodiodes and summarizes the energy band diagrams for several kinds of photodiodes. Given the equivalent circuits, this paper shows the general ones and RF small–signal ones together, and also considers the different designs for the RC–delay time part and the external circuit parts, including the air bridge, CPW pad, and load. Additionally, the DC saturation and RF output power are also analyzed in detail and a table for comparison is also included. (2) The solutions of bandwidth–responsivity trade–off for photodiodes are discussed in detail, which range from the use of high–reflected mirrors to the making of microholes and extending the reflected mirrors for group–IV photodiodes. (3) The photodiode photonic–integrated applications are presented, including the integration with short stubs and various planar antennas for high output power at high–speed conditions, and showing the parameter comparisons in tables, which provides a reference for the practice use of photonics–based devices operating at the system–level with the radiation power up to the mW–level and beyond in the THz band.