Optimized uni-traveling carrier photodiode and mushroom-mesa structure for high-power and sub-terahertz bandwidth under zero- and low-bias operation

In this paper, physically-based simulations are carried out to investigate and design uni-traveling carrier photodiode (UTC-PD) for high-power sub-terahertz wave generation at zero- and low-bias operation. The reliability of the physically-based simulation is demonstrated by comparing with our experimental result. Both the bandwidth and RF output power of the proposed UTC-PD is significantly improved by careful design the built-in electric field distribution under high-power input. For the optimized UTC-PD with the mesa diameter of 5 μm, its 3dB bandwidth large than 100 GHz even if the photocurrent reaches 6 mA under zero-bias operation. The device can reach a high bandwidth of 92.4 GHz, 105 GHz, and 119.5 GHz under the reverse bias of 0.5 V, 1 V, and 2 V, respectively, even the input photocurrent as high as 18.2 mA. The peak output-power of the device has enhanced at least 7 dB even at 170 GHz and zero- or low-bias operation. Besides, a novel design of mushroom-mesa UTC-PD (MM-UTC-PD) is proposed which with 4.3% improved high-speed performance. The MM-UTC-PD can trade-off between the external quantum-efficiency and bandwidth when miniaturized junction size is required.


Introduction
High efficiency, high speed, and high output-power photodiodes are critical components for optical communication systems. Typically, there is a trade-off between bandwidth and quantum efficiency for a p-i-n photodiode. The two major speed-limiting factors in p-i-n photodiodes that caused the compromise are the transit time and RC time. The RC time limit can be alleviated either by employing a smaller device area or by increasing the thickness of the depleted photo-absorption layer, thereby decreasing the junction capacitance. However, an increased depletion width consequently increases the transit time. On the other hand, a smaller device area and thinner absorption layer will reduce the external efficiency of p-i-n photodiodes. Unlike conventional p-i-n photodiode, the uni-traveling carrier photodiode (UTC-PD) consists of a p-type neutral absorption layer and a wide-gap depleted collection layer [1,2]. This means that the thickness of the depleted collector can be designed to be independent of the thickness of the absorber. Therefore, the RC delay time is decreased by increasing the thickness of the collector with ignoring the impact on the carrier transit time to a certain extent. The carrier transit time is decreased by downscaling the thickness of the absorber with ignoring the impact on the RC delay time to a certain extent. However, the increased collector thickness will lead to an increased reverse bias to maintain its depleted state. The increased bias voltage results in an increased selfheating effect because the Joule heat is equal to the bias voltage multiplied by the output current [3]. Hence, the more miniaturized size of the active area in UTC-PD will result in the broader bandwidth from dc to subterahertz [4][5][6][7][8]. However, this could lead to difficulty in optical coupling, exhibiting small external efficiency, ease of self-heating, and eventually thermal failure under high-power operation [7][8][9]. Altogether, the main mechanisms limiting the performance of UTC-PD are RC delay time, carrier transit time, space-charge effects, self-heating, etc [10].
One possible solution to mitigate the self-heating is the flip-chip bonding of photodiode onto a heat sink [8,9,11,12]. Another method is to make the UTC-PD worked under zero bias [3]. For zero bias operational UTC-PD, the bias circuit can be omitted, thus optimizing the trade-off between compact photonic integration and high baud rate transmission. This is because the signal crosstalk must be avoided when designing the high data rate transmission and high-density photonic integration technologies for millimeter-wave transmitter [13,14]. Besides, the packaging can be simplified, energy-consumption and dark-current can be reduced, and noise can be restrained, when the device is working at bias-free.
It is well known that the space charge effect can be suppressed by a higher bias voltage. Accordingly, the 3dB bandwidth and RF output-power of photodetector contain some penalty under zero-bias operation. However, for a zero-bias operational UTC-PD, the space-charge screening effect can be restrained to some extent via optimization the doping profile in the absorber, spacer and collector, and the thickness of spacer and collector [15]. In particular, a fixed distribution of background dopants in the collector can be used to repress the effect [16]. In this paper, by introducing and improving a graded doping profile rather than a uniform doping profile [6,17,18] in the depletion region of UTC-PD, the built-in electric field at the location where it will tend toward zero under high photocurrent density operation can be preconditioned to be higher. Consequently, the device can achieve high output-power with a broadband coverage from dc to sub-terahertz under zero-and low-bias operation, without the associated failure caused by increased thermal loading from higher reverse bias. The optimal epitaxial layer parameters of the device are acquired simultaneously.
Utilizing one of the advantages of UTC-PD, we can independently design the thickness of the absorption layer and the collection layer to alleviate the contradiction between quantum efficiency and RC time. In this work, by reducing the area of the collection layer while keeping the size of the active layer unchanged, a novel mushroom-mesa UTC-PD (MM-UTC-PD) structure is formed. Thereby, if it's RC delay time is reduced, then its bandwidth can be improved. Compared with the mushroom-mesa PIN-PD structure [19], the MM-UTC-PD structure does not reduce the area of the absorption layer. Therefore, with the suggested structure, the bandwidth can be enhanced without the external quantum efficiency being affected. However, the proposed MM-UTC-PD structure maybe bring other adverse aspects, if the photogenerated carriers in the suspended part of absorption layer cannot be collected quickly and effectively, a significant disadvantage effect on the bandwidth and responsivity. Consequently, it is necessary to verify the feasibility of this structure.

UTC-PDs under zero-bias operation
We get a modified zero-bias operational UTC-PD (PD1), which has a collector with a uniform doping profile in our previous work [15], and the detailed parameters of the structure is illustrated in table 1. The frequency response (I(ω)) of bias-free operational UTC-PD with large-signal (100% modulation depth) input is calculated by a commercial software of Silvaco Atlas. In this process, we use the Fermi-Dirac carrier statistics model together with the drift-diffusion model for carrier transport. The experimental test parameters of InP and In 0.53 Ga 0.47 As materials in the models of concentration dependent mobility (CONMOB) and parallel electric field mobility (FLDMOB) are available in the report of [20]. The models of carrier recombination in our simulation include Shockley-Read-Hall (SRH), concentration-dependent SRH, and Auger recombination models [21]. Numerical solutions are obtained using the Newton method in a three-dimensional model. Then, the normalized responsivity (t w ( )) and RF output power ( w ( ) P out ) of the device are calculated as follows [22], with the ideal load impedance (R L =50 Ω), the series resistance (R S =15 Ω) and parasitic capacitance (C P =5 fF) of the device, t w w where ω is angular frequency, and w = w= ( ) ( )| I I 0 . 0 The 3-dB bandwidth is defined as the frequencyf dB 3 at which the responsivity drops by 3-dB with respect to the DC value. The junction capacitance (C J )of the photodiode is extracted from the C-V curve via Atlas simulation. The parasitic capacitance and the series resistance respectively caused by the interaction of electrodes and the contact of metal-semiconductor were taken into account [23]. The physical model obtaining the bandwidth and output power of the UTC-PD includes the effects of the series resistance, parasitic capacitance, external load impedance, and the detailed parameters of the structure. In section four, the reliability of the physics-based simulation is demonstrated by compared with our experimental result. Figure 1 shows the built-in electric field distribution of the UTC-PDs with various doping profile in the collection layer under zero-bias operation. As illustrated in table 1, the PD1, PD2, PD3, and PD4 with same epitaxial layer parameters but the doping profile in the collector. The collector of PD1 has a uniform doping profile of 1×10 14 cm −3 , the collector of PD2, PD3, and PD4 with a linear graded doping profile are changing from 1×10 14 (near spacer side) to 1×10 16 cm −3 , 1×10 17 cm −3 , 2×10 17 cm −3 (near N-contact side), respectively. As plotted in the figure 1, the electric field intensity of the spacer and the input-side of the collector decreases at the higher input power [26], and the intensity of the output side of the collector will increase at the same time [25]. Distinctly, due to the linear gradient doping in the UTC-PD, the electric field intensity of the input side of the collector, the spacer and the output side of the absorber is stronger while the output side of the collector decreases. Compared with the uniform doping profile [6,[16][17][18], the built-in field at the location where it will tend toward zero under high input power operation can be preconditioned to be higher by the graded doping profile. In addition, as shown in figure 2, the increasing rate of junction capacitance of the devices with the gradient doped collection layer gets slower with the input power enhancement. Therefore, under zero biased operation, despite the bandwidth of the devices reduces as the input power goes up [27,28], this value decreases much slower, resulting from the large gradient of the doping in the collector, as illustrated in Table 1. Epitaxial layer parameters of the zero-and low-bias operational UTC-PDs a . a The PD1, PD2, PD3, PD4, PD5, and PD6 with same epitaxial layer parameters but the doping profile in the collector.
(2) InP→In 0.53 Ga 0.47 As (Spacer2)/13 means the 13 nm thick Spacer2 is a graded bandgap layer which changes linearly from In 0.53 Ga 0.47 As (near the Spacer1 layer side) to InP (near the cliff layer side).
(3) 1e16→1e14/N, (4) 1e17→1e14/N, and (5) 2e17→1e14/N means the N-type doping concentration varies linearly from 1×10 16 cm −3 , 1×10 17 cm −3 , and 2×10 17 cm −3 (near the subcollector layer side) to 1×10 14 cm −3 (near the cliff layer side) in the collection layer, respectively. (4) The doping profile in the collector of PD5, near the spacer side a uniform doping of 4×10 16 cm −3 with a thickness of 100 nm, and near the N-contact side a uniform doping of 1×10 14 cm −3 with a thickness of 200 nm [24]. (5) The doping profile in the collector of PD6, near the spacer side a uniform doping of 4×10 16 cm −3 with a thickness of 150 nm, and near the N-contact side a uniform doping of 1×10 14 cm −3 with a thickness of 150 nm [25]. figure 2(a). The 3dB bandwidth of the device has a graded doping profile in the collector is less than the structure has the uniform doping profile in the collector under a smaller input photocurrent, but at a bigger input photocurrent, the situation is conversely [25]. As another result, the device has a charge compensated collector will possess a lower and a higher RF output power at sub-terahertz under a smaller and a bigger input photocurrent, respectively, as in plotted figure 2 Although the preconditioned electric field distribution can achieve a greater degree by increasing the doping gradient, the electric field in the output side of the collector tends to zero as the increased doping gradient. Furthermore, as plotted in figure 2(a), the junction capacitance goes up with increasing doping gradient. For the device which the graded doping is changing from spacer side 1×10 14 cm −3 to N-contact side 2×10 17 cm −3 (PD4), its 3dB bandwidth less than 100 GHz when the input photocurrent large than 1.2 mA. The dark current of the device raises as the doping gradient of the collection layer ascends [29]. For n-type InP, when the doping concentration is below 1×10 17 cm −3 , the electron mobility and the overshoot velocity are almost the same as those in un-doped material [30]. Consequently, we suggested that the linear graded doping profile in the collection layer is changing from 1×10 14 (near the spacer side) to 1×10 17 cm −3 (near the N-contact side), corresponding to PD3. Compared with the PD4, the peak output power of PD3 is only cut down 0.8 dB while the energy consumption of PD4 is much higher than that of PD3, as plotted in figure 2(b).
As illustrated in figure 1, for PD3, the small electric field in the output side of the collection layer (−0.3 to −0.17 μm) is exhibited to 1.3 kV cm −1 under a low input power. However, the acquired acceleration of the electron will be higher at the initial segment and then travels through the structure because the electric field of the absorber, the spacer and the input-side of the collector are enhanced by the graded doping [18]. Besides, the electric field at the output side of the collection layer goes up to 5 kV cm −1 under the high input power. That's why, for PD3, as the input photocurrent increased from 1.2 to 10 mA, the 3dB bandwidth is around 100 GHz, and output-power exhibition good linearity, under zero-bias operation. For the device with uniform doping of 1×10 14 cm −3 in the collector (PD1), under the high input power, the electric field in the input-side of the collector and the output-side of the absorber tends to zero and dramatically decreases in the spacer synchronously [26], but the high charge compensated device has a different situation [25]. The junction capacitance strongly depends on the photocurrent for PD1, about three time's increase of the capacitance as the input photocurrent rising from 1.2 to 10 mA (from depletion state to non-depletion state). However, the junction capacitance only increased by 40% for PD3 correspondingly. In reference [6], the capacitance of the photodiode with an active diameter of 5 μm is changed about three times as the reverse voltage descending from 2 to 0 V even when the input power is zero. Therefore, as shown in figure 2(a), when the input photocurrent is above 6 mA, the 3dB bandwidth of PD3 is broader than that of PD1. For PD1, the bandwidth is decayed about 31 GHz (versus 50 GHz [27] and 35 GHz [28]) as the input photocurrent increases from 2 to 5 mA. The outputpower of the device is −4.95 dBm at 100 GHz and the corresponding photocurrent of 3 mA (versus −7 dBm [23]). As illustrated in table 2, for PD1 the peak output-power at 100 GHz is 1.63 dBm (versus −18.6 dBm [14]) and the corresponding photocurrent of~9.79 mA (versus 2.5 mA [14]). Its peak output-power at 160 GHz and 170 GHz is −1.76 dBm and −2.28 dBm (versus −13.9 @ 160 GHz with a 70% modulation depth [27] and −11.3 dBm @ 170 GHz with a 60% modulation depth [28]), respectively, both of the corresponding photocurrent of about 9.2 mA (versus 8 mA [27,28]). Nevertheless, for PD3, the output-power is −5.74 dBm at 100 GHz and 3 mA, and the peak output-power at 100 GHz, 160 GHz, and 170 GHz of 8.8 dBm, 5.25 dBm, and 4.75 dBm, respectively, and the corresponding photocurrent is approximately 15 mA, 27.6 mA, and 28.2 mA. Figure 3 shows the bandwidth of PD3 change with the thickness of the collector under various input power. The 3dB bandwidth of the PD3 is ascended and descended under a low (<6.07 mA) and a high (>18.2 mA) input power, respectively, as the collector becoming thickens (from 300 nm to 800 nm). The device has relatively thin collector, can achieve high linearity, and has the advantage of high speed in the case of high input power, which is consistent with the reported results [7,31,32]. There are four main causes for the above phenomenon. Firstly, the transit time of the photogenerated carrier is inverse to the collector thickness. Secondly, the built-in electric field of the device is collapse significantly under high input power [26]. Thirdly, the intensity of the electric field is reverse to the thickness of the collector due to the built-in potential is constant. Finally, under low input power, the junction capacitance of the device depends on the thickness of the collector and then will be saturated in this region due to the built-in potential with a small constant at zero-bias operation. Hereby, the thickness of the collector is set to 300 nm to obtain the trade-off between the transit time and RC time [32]. According to figure 2(a), when the input photocurrent is over than 1 mA under zero-bias operation, the bandwidth of the UTC-PD with the proposed doping profile will be improved, compared with that of the reported doping profile (PD5 and PD6) [24,25]. Then we obtained epitaxial parameters of the charge compensated zero-bias operational UTC-PD (CC-UTC-PD, PD3), as illustrated in table 1. For the CC-UTC-PD (PD3), the output power in the linear region and the peak RF output power dwindled and enhanced by 0.8 dB and 7.2 dB, respectively, and its bandwidth will be improved when the photocurrent is over 6 mA, compared to that of the UTC-PD with the uniform doping of 1×10 14 cm −3 in the collection layer.  Figure 4 shows the bandwidth and junction capacitance of the devices dependence of input power under low reverse bias operation. Generally, the reverse voltage can enhance the internal electric field of a photodiode, so the space-charge effect and the depletion layer can be suppressed and broadened, respectively. Consequently, the transit time and RC-delay time limited bandwidth is improved, the RF output power is also enhanced. The reverse voltage improves the bandwidth of all types photodiode compared with the zero-bias operation, as described in figures 2(a) and 4. As plotted in those figures, the junction capacitance is reduced obviously as the reverse bias increases [6], especially under high input power. The junction capacitance of the devices with a depleted drift layer is reasonable tends to 7 fF under the photocurrent of 1.2 mA and the reverse bias of 2 V [6,23]. For PD1, the capacitance at 15.7 mA is dwindled by half as the reverse voltage changing from 0.5 to 2 V. However, the electric field around the spacer of the device with a severe collapse under high input power [26], as shown in figure 1, which is not conducive the traveling of electrons from the absorber to the collector. Therefore, the bandwidth of the device is less significant improved as the voltage increases from 0.5 to 2 V when the photocurrent large than 13 mA. For the charge compensated structures, its bandwidth is significantly improved when the voltage increased. The bandwidth of the proposed CC-UTC-PD (PD3) is greater than PD1 when the input photocurrent bigger than 6.0 mA, 4.85 mA, and 2.6 mA under a reverse bias of 0.5 V, 1 V, and 2 V, respectively. The advised device can reach a high bandwidth of 92.4 GHz, 105 GHz, and 119.5 GHz at a reverse bias of 0.5 V, 1 V, and 2 V, respectively, when the input photocurrent is as high as 18.   As plotted in figure 4, the bandwidth of PD6 is greater than CC-UTC-PD (PD3) under a lower input power operation when the bias is above 1 V. Therefore, the RF output-power characteristics of them need a comparison. Figure 5 reveals the RF output-power depends on the input photocurrent for the UTC-PDs under low-bias operation. As shown in figures 2(b) and 5, with the increase of bias, the output power of the devices increases [7,27,28,31] and the output power of CC-UTC-PD is larger than that of PD6 and much larger than that of PD1. Besides, the output power of the devices decreases as the increase of the frequency [6]. In the linear region, PD1 with higher RF output power than CC-UTC-PD (PD3) when the operation frequency around 160 GHz and the reverse bias smaller than 0.5 V. This is due to the bandwidth of PD1 is higher than PD3 under above conditions.

Experimental results
The UTC-PD7 and UTC-PD8 epitaxial layers ware grown on semi-insulating single-side-polished InP substrates by Solid Source Molecular Beam Epitaxy. The epitaxial layer parameters are illustrated in table 4. The UTC-PD7 is a traditional structure, while the UTC-PD8 is a simpler structure which get from PD1. Backilluminated cylindrical mesa structure was fabricated by wet-chemical etching and traditional photolithography procedures. The TiAu layer was deposited on P-and N-contact layers by magnetron sputtering system and liftoff process in different steps. The TiAu P-contact electrode was fully covered the active area and act as a mirror under back-illumination. Afterwards, the wafer was annealed at 420°C for 2 min under a nitrogen atmosphere to form P-type and N-type Ohmic electrode. Then, the wafer covered with polyimide and annealed for prepassivation in nitrogen ambient. Subsequently, the top window was opened for microwave GSG coplanar pads connections to the N-and P-contact electrodes, and anneals it for passivation again. The GSG coplanar pads were deposited on polyimide for high-speed measurements.
The frequency responses of the UTC-PDs were measured using a 40-GHz Agilent network analyzer under 300 K. Light with a wavelength of 1550 nm modulated by Mach-Zehnder modulator was used as input light. The normalized S21 of the UTC-PDs is plotted in figure 6. In the measurement, the UTC-PD8 with 24 μm mesa diameter demonstrated 11.8 GHz and 10 GHz 3dB bandwidth at 0.05 mA and 1 mA photocurrent, respectively, under zero-bias operation. However, the 3dB bandwidth of the UTC-PD7 is only 6.2 GHz under the same conditions.
As shown in figure 7, the 3dB bandwidth of the UTC-PD8 with 20 μm and 14 μm mesa diameter can achieve 20.8 GHz and 40 GHz (versus 13 GHz with 40×5 μm 2 active area [39], 21.5 GHz with 15×4 μm 2 active area [40]), respectively, under zero-bias operation. The bandwidth of the UTC-PDs with reduced mesa diameter is significantly improved because the bandwidth is mainly limited by RC-time. When the reverse bias of 1 V, the 3dB bandwidth of the device with 20 μm mesa diameter can achieve 36.6 GHz, while the 3dB bandwidth of the device with 14 μm mesa diameter is large than 40 GHz.
As illustrated in figures 6 and 8, under bias-free and small input power operation, the bandwidth of the UTC-PD8 almost twice than the UTC-PD7, when both of the structure with same junction area. This is because the built-in electric field in the absorber, spacer and collector layers of the modified UTC-PD8 is large than UTC-PD7 under the same conditions. Therefore, the UTC-PD8 is a benefit structure to achieve high-speed Table 3. The performance of the UTC-PDs under low-bias operation.
Peak output power ( dBm) and photocurrent (mA) under these conditions. This result is consistent with the previously reported zero-bias operational UTC-PD design, the device with lower doping concentration in the collector can achieve high-speed response [14,15,23,28]. Nevertheless, under zero-bias operation, the bandwidth of the UTC-PD8 is significantly decreasing as the increased photocurrent. The main reason for this trend is that with the increase of input power, the electric field decreases and the energy band flattens. The report of [14] also suggest that the UTC-PD with a uniform doping concentration of 3×10 14 cm −3 in the collector result in a small saturated photocurrent. As plotted in figures 7 and 8, the frequency responses of the devices are obviously improved with an applied reverse bias of 1 V. The bandwidth of UTC-PD8 is insensitive to photocurrent under the reverse bias operation compared to zero bias operation. Thus, the major reason for the UTC-PD8 sensitive to photocurrent under zero bias operation is space charge effect. As shown in figures 7 and 8, the simulation results agree well with the experimental results. Therefore, the obtained numerical solutions in this paper are reliable, and it is expected that the proposed CC-UTC-PD can repress the bandwidth attenuated with the increase of photocurrent under zero bias operation.

Mushroom-mesa UTC-PD
The epitaxial parameters of the MM-UTC-PD are same as those of CC-UTC-PD. As plotted in figure 9, the schematic cross-sectional view of the cylindrical mesa of the MM-UTC-PD. By the process of reactive ion etching and selective wet-chemical etching the structure can be fabricated. In the figure, the D A and D C indicate the diameter of the absorption and collection layer of the device, respectively. The beneficial of the structure can be indicated and the optimum ratio between D A and D C can be obtained by changing the diameter of absorber and collector. Figure 10(a) shows the frequency response of the UTC-PDs with various junction area. As plotted in the figure, the transit time limited bandwidth decreases and the RC time limited bandwidth increases as the reduction of the D C /D A . Thus, the bandwidth of the device raises first and then dwindles as the D C /D A changes from 1 to 0.8. When the D C /D A equals 0.9, the 3dB bandwidth of the MM-UTC-PD is 116.25 GHz under zerobias operation, improved by 4.3% compared with the traditional structure. Figure 10(b) shows the RF output power of the UTC-PDs at 100 GHz under zero-bias operation. As illustrated in the figure, for devices with an active diameter of 5 μm, the output power is attenuated as D C /D A diminishes. The peak RF output power of the devices is 8.8 dBm, 7.9 dBm, and 5.76 dBm as the D C /D A of 1.0, 0.9, and 0.8, respectively, the corresponding input power of 92.3 mW, 86.4 mW, and 78.5 mW. The saturated input power descends as the D C /D A decrease, due to the mushroom structure with larger current density under same input power which causing more serious space charge effect. The output power in the linear region of the device faded by 0.7 dB, compared with the traditional structure under the same conditions when the D C /D A equals 0.9.  The transit time limited bandwidth and the output-power attenuated as the D C /D A diminish is major due to the photogenerated carriers in the suspended part of the absorption layer cannot be collected quickly and effectively. The suspended part of the absorber should be less than one diffusion length of the electron for the sake of avoiding the above detrimental effects to some extent. The average minority carrier mobility and lifetime in the Zn-doped absorption layer of~120 cm 2 V −1 s −1 and~260 ps [41], respectively, leading to the corresponding diffusion length only of~280 nm. Thus, for an MM-UTC-PD with an active diameter of 5 μm, the D C /D A should be more than 0.888 is consistent with the above simulation results, and the diffusion length determines the best D C /D A of the MM-UTC-PDs with different active size. Therefore, we proposed MM-UTC-PD with junction diameter (D C ) of 4.5 μm when the active diameter (D A ) of 5 μm. If the proposed structure with the junction diameter of 4.5 μm and 3 μm, then its active area will be 1.26 and 1.4 times the traditional structure,  respectively, when the two type structures have same junction capacitance, so the suggested structure is expected to achieve higher external quantum efficiency and to compensate its smaller loss of the output power.
As shown in figure 10(b), when the active diameter of the traditional UTC-PD is 4.5 μm, the RF output power under 100 GHz peaked at 8.41 dBm with an illumination beam diameter of 5 μm, and the corresponding input power is 98.1 mW. Although the peak output power of the traditional structure is improved by 0.34 dB compared to the suggested structure, the required corresponding input power of the former more than the latter as high as 11.8 mW, which contrary to the advocated of low power consumption. In the linear region, the output power of the proposed structure is 0.6 dB higher than the traditional UTC-PD, when both structures have the same junction diameter of 4.5 μm. Considering that the need to achieve broadband coverage from dc to subterahertz, a miniaturized size of the junction area is typically required [4][5][6][7][8]. Nevertheless, this miniaturization will cause some adverse effects, such as difficulty in optical coupling, exhibiting small external quantumefficiency and larger contact-resistance, ease of self-heating and eventually thermal failure under high-power operation [7][8][9]. In addition, the P-contact resistance is proportional to the metal-semiconductor contact area, and the larger P-contact area facilitates heat dissipation. Therefore, the proposed MM-UTC-PD can be used to trade-off the above contradiction to a certain degree.

Conclusion
The properties of the proposed CC-UTC-PD are simulated, and these simulations indicate it's a benefit structure to high-power sub-terahertz wave generation at zero-and low-bias operation. The physically-based simulation successfully predicted that the performance of the UTC-PD varies with the change of its structural parameters and the external conditions, and the simulation results agree well with our experimental results. This indicates that the physically-based simulation is reliable, and it can be used to design and analysis the UTC-PDs. For the CC-UTC-PD with the mesa diameter of 5 μm, its 3dB bandwidth large than 100 GHz even if the photocurrent reaches 6 mA under zero-bias operation. The device can reach a high bandwidth of 92.4 GHz, 105 GHz, and 119.5 GHz under the reverse bias of 0.5 V, 1 V, and 2 V, respectively, even the input photocurrent as high as 18.2 mA. The attenuation of the device's bandwidth is suppressed when the input photocurrent increased comparing to other structural or experimental results. Besides, the proposed structure has enhanced at least 7 dB in peak output-power compared to the device with a uniformly doped collection layer, even at 170 GHz and zero-or low-bias operation. Although the MM-UTC-PD with a little punishment in peak output power, its bandwidth is improved by 4.3%, compared with the traditional structure. This suggested structure can realize the trade-off between the external quantum-efficiency and bandwidth when miniaturized junction size is required. The above proposed structures may find applications in next-generation optical interconnect or coherent fiber communication systems, where is greatly desirable for broadband high-power photo-receiver with low power consumption and high-density integration. Our future work is to fabricate photodetectors with the proposed structures and further verify the proposed design.