Traveling-Wave Photodetector with Asymmetrically ...

Viewer
Transcript

Japanese Journal of Applied Physics Vol. 43, No. 8A, 2004, pp. 5105–5109 #2004 The Japan Society of Applied Physics

Traveling-Wave Photodetector with Asymmetrically Heterostructured Intrinsic Region Joong-Seon C HOE, Yong-Hwan KWON, Kisoo KIM, Jeha K IM, Soon-Cheol K ONG1 y and Young-Wan C HOI1 Electronics and Telecommunications Research Institute, Daejon 305-350, Korea 1 Optoelectronics and Optical Communications Laboratory, Chung-Ang University, Seoul 156-756, Korea (Received January 20, 2004; accepted March 24, 2004; published August 10, 2004)

Bandwidth limitation by carrier transit time was analyzed mathematically when traveling-wave photodetector’s intrinsic region consists of heterostructure. Because of the smaller hole’s velocity than electron’s, the hole transit length is crucial for determining the large bandwidth. In order to compensate the small transit velocity of hole, we propose that the intrinsic region should be designed asymmetrically. In this study we found that appropriate asymmetric structure enhanced bandwidth by about 100% of symmetric structure and proved experimentally the eﬀect of asymmetric intrinsic region. [DOI: 10.1143/JJAP.43.5105] KEYWORDS: photodetector, transit-time-limited bandwidth, heterostructure, TWPD, WGPD, FDTD

1.

Introduction

cladding layer

Recent progress in ﬁber-optic telecommunication technology requires the devices of higher speed and larger capacity in communication systems. To meet these requirements wideband optical communication systems working in the microwave regime photodetectors (PD’s) with both high eﬃciency and broad bandwidth are needed.1) It was found that vertically-illuminated photodetectors (VPD’s) cannot achieve these requirements simultaneously.2,3) On the other hand, side-illuminated photodetectors (SIPD) such as waveguide photodetector (WGPD) and traveling-wave photodetector (TWPD) have drawn much interest since the bandwidth and internal quantum eﬃciency can be speciﬁed almost independently and optimized separately.4) However, there should still be a tradeoﬀ between carrier transit time and junction capacitance because the former can be reduced by using thin intrinsic region while RC time increases with intrinsic region thickness.3) A WGPD is called TWPD if photogenerated RF signal travels along with the incident optical signal, opposed to lumped PD, and when the RF velocity becomes identical to the optical velocity, the device would not be aﬀected by RC time limitation. In spite of overcoming RC time limitation with achieving velocitymatching condition—which is in reality very diﬃcult except for distributed structure5)—carrier transit time limitation still inﬂuences upon the overall bandwidth. Considering that the objective of TWPD design is to obtain not only the large bandwidth but also high RF output,6,7) the absorption layer thickness needs to be thin while the total intrinsic region should be thick enough for small device capacitance per unit length in order to get high RF velocity.8) In the previous work of Eﬀenberger and Joshi,9) the velocity diﬀerence was considered in designing VPD where thick absorbing layer was inevitable for high responsivity. However, they didn’t present mathematical expression that could be useful in determining optimal structures of diﬀerent type such as SIPD. In this study we investigated analytically the transit time limited bandwidth of SIPD’s and present optimum epitaxial structure considering the diﬀerence in hole and electron saturated velocities. By fabricating TWPD’s from two kinds of epitaxial wafers, usual sym

E-main address: [email protected] Present address: Samsung Electro-Mechanics, Suwon, Korea.

y

p−doped region incident beam

+x

Lp

−x

Ln

D

absorbing layer

intrinsic region

n−doped region

Fig. 1. Schematic epitaxial structure used in this analysis. Intrinsic region, containing absorption layer, is sandwiched by doped regions and light incidents through the cleaved edge.

metric and asymmetric structure, we proved the eﬀect of asymmetric structure upon the high speed performance of the devices. 2.

Theory

Lucovsky et al. reported an expression describing frequency response of VPD in 1964.10) Their result has been useful in designing VPD and it also works for WGPD in proper approximation. As crystal growth technology develops, more sophisticated epitaxial structures are adopted for devices, for example WGPD with heterostructure intrinsic region as in Fig. 1 that shows the schematic epitaxial structure used in this analysis. Intrinsic region consists of two barrier layers and absorbing layer sandwiched by them. P- and n-doped regions are located on top and bottom of the intrinsic region, respectively. The separation between absorbing layer and p-doped (n-doped) region is denoted as Lp (Ln ). As in ref. 10, analysis begins from the continuity equations of hole and electron: @p @p ¼gv @t @x @n @n ¼ g þ bv @t @x

ð1Þ ð2Þ

where p, n, g, v, and b indicate hole concentration, electron concentration, generation rate, hole velocity, and ratio of electron to hole velocity, respectively. Before applying the analysis, the carrier generation rate in ref. 10, g ¼ g0 GðtÞ, should be modiﬁed into a function of time and position in order to be applicable to the structure where the part of intrinsic region is transparent to incident light. In the structure depicted in Fig. 1, g can be written as:

5105

Jpn. J. Appl. Phys., Vol. 43, No. 8A (2004)

g ¼ g0 ðxÞð1 þ A expði!tÞÞ D D ¼ G0 x þ x 2 2

J.-S. CHOE et al.

1.0

ð3Þ

ð1 þ A expði!tÞÞ where ðxÞ ¼ 1 if x > 0 and ðxÞ ¼ 0 otherwise. Here the origin is located on the center of absorption layer. !, G0 , and A are constants representing the modulation frequency, the generation rate of absorbing layer, and modulation depth of incident light, respectively. If the eﬀect of transverse mode is to be included, mode proﬁle should be multiplied to eq. (2). In this analysis the eﬀect by mode proﬁle was ignored because it was a minor factor. After some algebra under the boundary conditions of nðD=2 þ Lp Þ ¼ pðD=2 Ln Þ ¼ 0, one can get the solution for the carrier proﬁle as a function of position and time: 8 2G0 A !D i!x > > sin exp expði!tÞ > > > ! 2v v > > > > G D D > > > þ 0 if x > , > > v 2 > > i! v 2 > > > > > G0 ðx þ D=2Þ D > > if jxj ; þ > > > v 2 > > > > D > : 0 if x < 2 8 D > > > 0 if x > ; > > 2 > > > > G A i! D 0 > > 1 exp x expði!tÞ > > > i! bv 2 > > < G0 ðx þ D=2Þ D nðx; tÞ ¼ ð5Þ if jxj ; þ > bv 2 > > > > 2G0 A !D i!x > > > sin exp expði!tÞ > > ! 2bv bv > > > > > > : þ G0 D if x < D : bv 2 Using these results, following unnormalized frequency response of the structure is obtained: 2iv !D i!D i!Lp ~ J ð!Þ ¼ 2 sin exp exp 1 ! 2v 2v v i 2v !D i!D D sin exp ! ! 2v 2v 2ibv !D i!D i!Ln þ 2 sin exp exp 1 ! 2bv 2bv bv i 2bv !D i!D D sin exp : ð6Þ ! ! 2bv 2bv Figure 2 shows the normalized frequency response jJ~ð!Þ= J~ð0Þj2 obtained from eq. (6), assuming v, b, D as 3 106 cm/s, 3, and 0.25 mm, respectively. In (a), Ln ¼ Lp ¼ 0:375 mm was used while (b) structure has Ln ¼ 0:6 mm and Lp ¼ 0:15 mm. For (c) structure Ln and Lp are 0.2 mm and 0.55 mm, respectively. In (a) structure, 3 dB bandwidth was 25.6 GHz but that of (b) asymmetric structure was as small as 17.7 GHz. On the other hand, another asymmetric structure (c) shows 45.1 GHz, which was larger than that

Frequency Response (a.u.)

5106

(a) symmetric (b) asymmetric #1 (c) asymmetric #2

0.8 0.6 0.4 0.2 0.0 0

10

20

30

40

50

60

70

80

Frequency (GHz) Fig. 2. Calculated frequency responses for each structures: (a) symmetric, (b) asymmetric-absorption layer close to n-doped region (c) asymmetricabsorption layer close to p-doped region. For all the structures, intrinsic region and absorption layer are 1 and 0.25-mm-thick, respectively.

of symmetric structure by 19.5 GHz. This is due to the diﬀerence in velocities of hole and electron. As electron moves faster than hole (in this calculation, three times faster), overall transit time reduces if absorption layer is near p-doped layer. If absorption layer is extremely thin; for example, a single quantum well, the optimum position will be determined by the velocity ratio compensating for the retarded transit time. But generally absorption layer has 11) for reasonable thickness exceeding several thousands A responsivity and therefore the device performance is aﬀected by each position where electron–hole pairs are generated. In our previous work, we have already proposed the advantage of asymmetric intrinsic region through ﬁnite-diﬀerence time-domain (FDTD) analysis.12) The analytic solution above, however, is obviously more convenient and eﬃcient in optimizing device structures compared with any numerical methods. Figure 3 shows the behavior of 3 dB bandwidth as a function of the separation between p-doped and absorption layer for various absorption layer thicknesses. As the absorption layer becomes thick, the maximum available bandwidth decreases and the separation between p-doped and absorption layer should be reduced to obtain better frequency response. When the absorption layer is thicker than 0.4 mm, it has to be located right next to p-doped layer for high bandwidth. This is consistent with the dual-depleted region p–i–n photodetector structure,9) which was adopted for vertically illuminated photodetector where relatively thick absorption layer was crucial for high eﬃciency. In the inset of Fig. 3 bandwidths are plotted as a function of center position of absorption layer. It is obvious that when the center of absorption layer is located at a speciﬁc position (in this case, approximately 0.22 mm from the p-doped region edge) the bandwidth has its maximum value. According to eq. (6), the magnitude of frequency response contains v and bv so that the optimum position of center diﬀers from 0.25 mm that corresponds to the same transit time for both carriers when the thickness of the absorption layer is negligible and b equals to 3.

Jpn. J. Appl. Phys., Vol. 43, No. 8A (2004)

45

45

15

InP 0.2 µm

30

InGaAsP

0.25 µm 15

100 Å 500 Å 1000 Å 2000 Å 3000 Å 4000 Å

30

5107

p−doped

60

3dB Bandwidth (GHz)

3dB Bandwidth (GHz)

60

J.-S. C HOE et al.

0.2 µm

InGaAs 0.375 µm

1µm

InGaAsP 0.7075 µm

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

InP

Position of Absorption Layer Center (µm)

symmetric

n−doped

asymmetric

(a)

0 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Separation between p-doped and absorption layer (µm) Fig. 3. Carrier transit-time-limited 3 dB bandwidth as a function of the separation between p-doped region and absorption layer edge. In the inset is shown the bandwidth as a function of absorption layer’s center position.

(b) Fig. 5. (a) Schematic diagrams of epitaxial wafers used in fabricating devices. (b) Plane view of a device after cleaving process.

3dB Bandwidth (GHz)

60

the position of absorption layer plays an important role for the ﬁnal device performance. 50

3. 40

optimally asymmetric symmetric

30

20 0.0

0.1

0.2

0.3

0.4

Absorption Layer Thickness (µm) Fig. 4. Carrier transit-time-limited bandwidth of symmetric and optimally asymmetric structures as a function of absorption layer thickness when total intrinsic region thickness is ﬁxed to 1 mm.

Shown in Fig. 4 is the dependence of 3 dB bandwidth upon the absorption layer thickness. In general, narrow spatial distribution of photogenerated carriers in the thin absorption layer also results in narrow distribution in transit time, which corresponds to sharp impulse response. Optimally asymmetric structure was found to be superior to symmetric structure in the bandwidth by 20.1 GHz to 26.7 GHz. Although the bandwidth is inversely proportional to the absorption layer thickness even in the symmetric structure, the optimum structure provides more improvement. If long device is to be used for high RF output, as in TWPD, thick intrinsic region and thin absorption layer is the design rule for high RF velocity, low capacitance, and low absorption eﬃciency. Thickened intrinsic region, compensating for large junction area, makes carrier transit-time a critical parameter determining the bandwidth. In such cases

Fabrication and Experimental Result

The epitaxial layers of the structures were grown on a semi-insulating InP substrate by low-pressure metallorganic chemical vapor deposition. Among the structures used for calculation in Figs. 2(a) and 2(b) were tried. In0:53 Ga0:47 As and In0:73 Ga0:27 As0:58 P0:42 were adopted as an absorbing layer and cladding layers, respectively. The 3-mm-thick ridge was deﬁned using standard photolithography, followed by reactive ion etching. Exposed sidewall was passivated using polyimide for low leakage current. Ohmic contacts were formed with Ti/Pt/Au for pþ electrode, and Ge/Au/Ni/Au for nþ electrode. Finally, coplanar waveguide(CPW) electrode was deposited with Ti/ Au for microwave probing. Figure 5(b) shows the microscopic plane view image of a device after cleaving process to form the facet for beam incidence. Typical devices showed leakage current of less than 20 nA at the reverse bias of 3 V. The frequency responses of the devices were measured using Lightwave OMS-2010 optical heterodyne source, an HP E4417A RF power meter, and an HP 8487D RF power sensor. Modulated laser beam was focused on the facet through a tapered ﬁber and generated electric signal was observed with a ground–signal–ground (GSG) microwave probe. During the measurement, reverse bias was applied to the devices through a bias-tee in order to get the frequency response saturated. For the asymmetric structured devices, the reverse bias of 3 V was enough for obtaining the saturated response while the symmetrically structured ones required a voltage as high as 5 V due to the heterojunctions within the intrinsic region.2) By applying appropriate calibration, the frequency re-

5108

Jpn. J. Appl. Phys., Vol. 43, No. 8A (2004)

(a)

Length (µm) 0

25

50

75

100 125 150 175 200

3dB bandwidth (GHz)

50 asymmetric structure limit (45.1 GHz)

40

symmetric structure limit (25.6 GHz)

30

J.-S. CHOE et al.

achievable bandwidth in the transit-time-limited regime. The structure suggested in this study adopts thick intrinsic region of 1 mm that might cause poor bandwidth property. If the intrinsic region consists of In0:53 Ga0:47 As single layer, the transit time limited bandwidth is expected to be about 30 GHz.3) As in Fig. 6(b), the bandwidth exceeds 30 GHz by large amount for short asymmetric structured devices, in spite of In0:73 Ga0:27 As0:58 P0:42 layers within intrinsic region whose saturated carrier velocity is smaller than In0:53 Ga0:47 As.15) This is owing to the eﬀect of both the thin absorption layer, as shown in Fig. 4, and asymmetric structure. The measured bandwidth of exceeding 40 GHz for the devices with 1-mm-thick intrinsic region clearly supports the eﬀectiveness of the asymmetric design. The superiority of the asymmetric to symmetric structure is, however, not the result of only the carrier transit time. Within the intrinsic region of both the structure, there exist InP/In0:73 Ga0:27 As0:58 P0:42 heterojunction, which causes not only high working voltage but also carrier pile-up and charge screening resulting in poor high-speed performance.2) While the symmetric structure contains two heterojunctions, the asymmetric structure does only one near the n-doped region which has little eﬀect due to the small eﬀective mass of electron. 4.

20

asymmetric symmetric

10 0

75

150 225 300 375 450 525 600 2

Junction Area (µm ) (b) Fig. 6. (a) Measured frequency responses of two devices with junction area of 150 mm2 . (b) The tendency of 3 dB bandwidth with junction area.

sponse of device were extracted from the data embedding the characteristics of GSG probe, cable, adapters, and bias tee. Calibrated data is shown in Fig. 6(a) for devices with junction area of 150 mm2 . As can be seen clearly, asymmetric structured device showed much better frequency response than symmetric one, which agrees well with the theory developed in §2. The bandwidth of the former is as large as 37.3 GHz but the latter shows bandwidth less than 20.2 GHz. Considering the same physical dimensions of the two devices, this reveals the eﬀect of intrinsic region structure upon the 3 dB bandwidth. Figure 6(b) summarizes the measured 3 dB bandwidth of devices with various junction areas. The bandwidth depends on the device length even if the devices are designed as TWPD. This is due to microwave loss at the ridge region where doped layers are beneath the CPW.12–14) From FDTD calculation, microwave loss was found to be about 6 dB/mm at 40 GHz, which corresponded to loss of 1.2 dB for 200-mmlong devices at that frequency. Therefore, the transit-timedominant regime is the short device length, as in lumped PD where RC time delay is a crucial factor. For both the cases of symmetric and asymmetric structures, the analysis of Section 2 explains well the upper bound of experimentally

Conclusion

We analyzed the eﬀect of absorption layer thickness and position within intrinsic region and conﬁrmed it experimentally. An analytic solution was obtained describing frequency response of photogenerated current in side-illuminated p– i–n PD. According to the result, the bandwidth was enhanced by locating the absorption layer toward p-doped region, which was explained by the slow saturated velocity of hole within the semiconductors. When optimally designed structure was used, the frequency response of photodetector was improved by 20.1 to 26.7 GHz compared to the normal symmetric structure. Through measuring the frequency responses of fabricated devices, we found that the upper bound of bandwidth predicted by the theory agreed well with the measured values in the transit-time-limited regime, and the maximum bandwidth obtained was 23.9 GHz and 45.1 GHz for symmetric and asymmetric structured device, respectively.

1) V. M. Hietala, G. A. Vawter, T. M. Brennan and B. E. Hammons: IEEE Trans. Microwave Theory Tech. 43 (1995) 2291. 2) J. E. Bowers and C. A. Burrus, Jr.: J. Lightwave Technol. 5 (1987) 1339. 3) K. Kato: IEEE Trans. Microwave Theory Tech. 47 (1999) 1265. 4) J. E. Bowers and C. A. Burrus: Electron. Lett. 22 (1986) 905. 5) M. S. Islam, S. Murthy, T. Itho, M. C. Wu, D. Novak, R. B. Waterhouse, D. L. Sivco and A. Y. Cho: IEEE Trans. Microwave Theory Tech. 49 (2001) 1914. 6) K. S. Giboney, M. J. W. Rodwell and J. E. Bowers: IEEE Photon. Technol. Lett. 4 (1992) 1363. 7) R. R. Hayes: U.S. Patent 6 278 820 B1 (2001). 8) K. S. Giboney, M. J. W. Rodwell and J. E. Bowers: IEEE Trans. Microwave Theory Tech. 45 (1997) 1310. 9) F. J. Eﬀenberger and A. M. Joshi: J. Lightwave Technol. 14 (1996) 1859. 10) G. Lucovsky, R. F. Schwarz and R. B. Emmons: J. Appl. Phys. 35 (1964) 622. 11) H. Jiang and P. K. L. Yu: IEEE Trans. Microwave Theory Tech. 48

Jpn. J. Appl. Phys., Vol. 43, No. 8A (2004) (2000) 2604. 12) S.-C. Kong, S.-H. Ok, Y.-W. Choi, J.-S. Choe, Y.-H. Kwon and J.-H. Kim: IEEE J. Sel. Top. Quantum Electron. 9 (2003) 770. 13) S.-C. Kong, S.-J. Lee, J.-H. Lee and Y.-W. Choi: IEEE Trans. Microwave Theory Tech. 50 (2002) 2589.

J.-S. C HOE et al.

5109

14) K. S. Giboney, R. L. Nagarajan, T. E. Reynolds, S. T. Allen, R. P. Mirin, M. J. W. Rodwell and J. E. Bowers: IEEE Photon. Technol. Lett. 7 (1995) 412. 15) T. P. Pearsall, ed.: GaInAsP Alloy Semiconductors (John Wiley & Sons, New York, 1982).