Solid-State Electronics 45 (2001) 359±364
Determining non-quasi-static small-signal equivalent circuit of a RF silicon MOSFET Seonghearn Lee a,*, Hyun Kyu Yu b a
b
Department of Electronic Engineering, Hankuk University of Foreign Studies, San 89, Wangsan-ri, Mohyun-myun, Yongin, Kyungki-do 449-791, South Korea Micro-Electronics Technology Laboratory, Electronics and Telecommunications Research Institute, Yusong-gu, Taejon 305-606, South Korea Received 6 January 2000; received in revised form 18 July 2000
Abstract We present an accurate non-quasi-static small-signal MOSFET model incorporating distributed channel and substrate resistances to extend the frequency limit of the model validity to higher frequencies. The model parameters are accurately determined using a new semi-analytical extraction technique combining analytical and optimization approaches as an eective way to attain a global minimum. The validity of the model and parameter extraction method is justi®ed by observing excellent agreements between the measured and modeled S-parameters in the wide range of frequency. Ó 2001 Published by Elsevier Science Ltd. Keywords: MOSFET; Parameter extraction; Non-quasi-static; Modeling; S-parameters
1. Introduction Recently, RF communication systems are going toward the wider bandwidth and higher-frequency range. In RF/microwave IC applications [1], a low-cost silicon MOSFET is rapidly emerging as a core device [2]. Accurate small-signal equivalent circuit modeling of a silicon MOSFET becomes very important for designing RF ICs and characterizing processes and devices. In most papers [3±6], a simple quasi-static small-signal equivalent circuit in Fig. 1 has been widely used for MOSFET modeling and parameter extraction. However, it was recently pointed out that the connection of drainbulk junction capacitance Cds to the internal source in Fig. 1 is not physically acceptable and produces severe errors in extracting model parameters at high frequencies [7]. In order to eliminate this problem, an improved
*
Corresponding author. Tel.: +82-31-330-4117; fax: +82-31330-4120. E-mail address:
[email protected] (S. Lee).
model in Fig. 2 has been proposed and its physical accuracy has been demonstrated [7]. In general, the Si well/substrate region that possesses the lossy dielectric property acts as the distributed resistive network. This substrate eect becomes increasingly important to predict RF performance in the highfrequency region [8±16], but is not accounted for the conventional small-signal models discussed above. Recently, several substrate models with a distributed resistive network [8±10] have been developed for a SPICE BSIM3v3 model [17]. As a simpli®ed version, a compact model with a single series resistance is also proposed [12±16]. This simple model is more useful to enhance the accuracy of parameter extraction than the distributed network [15]. On the other hand, the accuracy of these previous quasi-static models is considerably limited to lower frequencies below the cuto frequency (fT ) [18±20]. Physically, the inversion channel region is distributed in the lateral direction under the gate oxide, and this distributed nature results in the channel propagation delay generated by RC transmission line eect [18±21]. Recently, this channel delay eect has been successfully
0038-1101/01/$ - see front matter Ó 2001 Published by Elsevier Science Ltd. PII: S 0 0 3 8 - 1 1 0 1 ( 0 1 ) 0 0 0 0 6 - 5
360
S. Lee, H.K. Yu / Solid-State Electronics 45 (2001) 359±364
Fig. 1. A simple quasi-static small-signal equivalent circuit model for a RF Si MOSFET gm gmo exp
jxs.
Fig. 2. A physically acceptable quasi-static small-signal equivalent circuit model for a RF Si MOSFET.
modeled as a bias-dependent RC series element [18±23]. This non-quasi-static eect due to the channel delay plays an important role when the operating frequency gets close to fT . Thus, the non-quasi-static eect should be incorporated to extend the frequency limit of the model validity to higher frequencies. Together with this model development, intense research of the accurate parameter extraction of the models should be carried out as an essential step in RF MOSFET modeling. Therefore, in this paper, we present an accurate non-quasi-static small-signal model accounting for the distributed channel and substrate eects in RF Si MOSFET, and develop a new semi-analytical parameter extraction technique combining analytical and optimization approaches for enhancing the extraction accuracy.
2. A non-quasi-static model Fig. 3 shows a non-quasi-static small-signal MOSFET model that is constructed by adding the distributed channel and substrate resistances into Fig. 2. In this model, bias-dependent RC distributed channel eects
Fig. 3. A non-quasi-static small-signal equivalent circuit model for a RF Si MOSFET.
are considered by connecting non-quasi-static channel resistances
Rgsi and Rgdi in series with channel capacitances
Cgsi and Cgdi , respectively [18±22]. In addition, the high-frequency substrate region is modeled by connecting the series bulk resistance
Rbk to Cds [12±16]. For the purpose of enhancing the model accuracy, overlap and fringing components
Cgso and Cgdo are separated from channel ones [23]. The parameter of Rg models the eective gate resistance [24] accounting for the distributed transmission line eect [25]. The model parameters in a conventional quasi-static model are usually determined from the measured Sparameters using several direct methods [3±6], but these methods are not applicable to extract the non-quasistatic model due to the large number of unknown parameters as well as dierent circuit topology. In order to remove this drawback, a global optimization technique is traditionally used to ®t the model to the measured Sparameters, but this optimization may suer uncertainties in ®nding an unique solution due to the large number of unknowns. Therefore, in this paper, the following semi-analytical procedure combining analytical and optimization techniques has been developed as an ecient way to reduce unknown parameters.
3. Model parameter extraction N-MOSFETs with multiple n poly-gate ®ngers with 0.8 lm length and 10 10 lm total width were fabricated on p-type 2 kX cm high-resistivity Si wafers using a standard twin-well CMOS process [26]. S-parameters were measured under the grounded bulk con®guration and probe pad parasitics are eliminated from the measured ones using an ``open'' test structure [27,28]. Since the non-quasi-static model in Fig. 3 can be simpli®ed as a quasi-static model in Fig. 2 at low frequencies well below fT , all quasi-static parameters can be directly determined. Then, non-quasi-static parameters are extracted using the numerical optimization while
S. Lee, H.K. Yu / Solid-State Electronics 45 (2001) 359±364
361
keeping all quasi-static parameters constant. As the ®rst step of this semi-analytical extraction technique, a direct extraction technique is developed to determine the quasi-static parameters as follows. Because gm and Rds are neglected in Fig. 2 at zero bias, the following equation is derived [3]: Rd Re
Z22 Z12 . Thus, Rd is extracted to be 24 X from Fig. 4 by taking a frequency-independent value in the low-frequency range. Also, the value of Rs 24 X is determined by plotting the frequency response of Re
Z22 Z12 for a test MOSFET where source and drain are interchanged. As shown in Fig. 5, Ld is determined to be 112 pH by using the curve ®tting of the following equation [4,5] in the wide range of frequency: 1 Im
Z22 x
Z12 Ld
Ed x2 B
1
where B, Ad and Ed are expressed as functions of intrinsic parameters, and are constant values at ®xed bias because the intrinsic ones are independent of frequency. Eq. (1) is originally derived from the quasi-static model, but is not strongly aected by non-quasi-static and bulk elements in Fig. 3. Thus, the Ld extraction using Eq. (1) is likely to be valid for the non-quasi-static model. The value of Cds is approximated by the following simple expression: Cds
1 Im
Y22A Y12A x
2
where the Y A parameters are obtained by subtracting parasitic resistances from the measured Z-parameters at low frequencies where inductances can be neglected. In this subtraction, we used the value of Rg that was ex-
Fig. 5. The measured data (s, n) and ®tted curves (Ð) versus frequency for measured equation (1) and corrected equation (5).
tracted from the following simple equation [3] derived at zero bias: Rg Re
Z11 Z12 . Using Eq. (2), Cds is directly extracted to be 35 fF. In order to determine the rest of parameters more accurately, corrected Z c -parameters are obtained by subtracting extracted values of Rd , Ld , and Cds sequentially from the measured Z-parameters. Using the corrected Z c -parameters, the new value of Rg is extracted to be 16 X by ®nding a constant value in Fig. 4 after abrupt fall-o of the frequency dependent term in the following equation [4,5]: c Re
Z11
c Z12 Rg
Ag x2 B
3
The values of Ls 50 pH, and Lg 86 pH are extracted by ®nding constant terms through the curve-®tting process of the following equations (4) and (5) in the wide frequency range, respectively [4,5]: 1 c Ls Im
Z12 x 1 c Im
Z11 x
Fig. 4. The frequency response curves of measured Re
Z22 Z12 at zero bias and corrected Re
Z11 Z12 at VGS 2 V and VDS 3 V.
Es x2 B
c Z12 Lg
4 Eg B
x2
Fg 2 x
x2
B
5
where Ag , Es , Eg , and Fg are functions of the intrinsic parameters, and are independent of frequency. As shown in Fig. 5, Lg is extracted using the measured data in the high-frequency region where Fg =x2
x2 B is neglected compared to Eg =
x2 B: This high-frequency ®tting process results in the correct extraction of Lg , because Eq. (5) is approximately valid for a non-quasistatic model in Fig. 3. Fig. 5 shows a good correspondence between the measured data and ®tted curves of
362
S. Lee, H.K. Yu / Solid-State Electronics 45 (2001) 359±364
Eqs. (1) and (5) in the wide range of frequencies up to 39.5 GHz, which demonstrates the accuracy of the extracted values. After Rs , Ls , Rg , and Lg are sequentially subtracted from the corrected Zc -parameters, intrinsic parameters are extracted using the following Yi -parameter equations [3,5] for an intrinsic MOSFET in a dotted box of Fig. 2: Cgs
1 Im
Y11i Y12i x
6
Cgd
1 Im
Y12i x
7 Fig. 7. The s and Rds versus frequency.
1 Rds Re
Y22i
8
gmo jY21i
9
s
Y12i j
1 phase
Y21i x
Y12i
10
After these intrinsic parameters are plotted as a function of frequency in Figs. 6 and 7, uniform data in the low-frequency range are taken as their quasi-static parameter values. After all, it was found that Cgs 126 fF, Cgd 22:5 fF, Rds 745 X, gmo 12:5 mS, and s 3:8 ps. In order to verify the inadequacy of the quasi-static model in the high-frequency region, the modeled S-parameters for Fig. 2 with the above extracted parameters are compared with the measured ones in Fig. 8. The deviation between the measured and modeled S-parameters starts to be noticeable from about 7 GHz, and becomes larger at higher frequencies. Fig. 8. Comparison of measured S-parameters (}) with modeled ones using quasi-static equivalent circuit of Fig. 2 ( ) and non-quasi-static equivalent circuit of Fig. 3 (Ð) from 0.5 to 32.5 GHz.
Fig. 6. The frequency response curves of Cgs , Cgso , Cgd and gmo .
This discrepancy is originated from non-quasi-static channel and bulk eects. In order to remove the discrepancy, a non-quasistatic model in Fig. 3 is used in this work. At zero bias
VGS VDS 0 V, Cgs is reduced to Cgso owing to the absence of channel capacitance. Using
1=x Im
Y11 Y12 , Cgso of 45 fF is obtained with higher accuracy from the measured S-parameters at zero bias. Thus, Cgsi is extracted by Cgsi Cgs Cgso 81 fF. Since Cgdi disappears in the saturation region
VGS 2 V; VDS 3 V, Cgdo Cgd 22:5 fF. However, when a device operates in the linear region, Cgdi should be determined by Cgdi Cgd Cgdo . Next, Rgsi and Rbk were optimized to obtain the closest possible ®t to be measured
S. Lee, H.K. Yu / Solid-State Electronics 45 (2001) 359±364
S-parameters, while all quasi-static parameters were ®xed at previously extracted values. From this optimization, it was obtained that Rgsi 25 X and Rbk 25 X. In Fig. 8, a good agreement is observed between the measured and modeled S-parameters using Fig. 3, demonstrating the accuracy of the non-quasi-static and semianalytical extraction method. 4. Conclusions An accurate small-signal model accounting for nonquasi-static and substrate eects has been proposed for high-frequency circuit simulation. This non-quasi-static model is developed to solve the serious problem that a quasi-static model is inaccurate in the high-frequency region. In order to identify the model parameters, all quasi-static parameters are directly extracted from derived formulations, and the non-quasi-static channel and bulk resistances are next obtained using optimization. This new semi-analytical approach provides an ecient way to prevent the optimization from trapping into a local minimum. The validity of this model and extraction method is veri®ed over the wide range of frequency. The non-quasi-static model will be successfully used to design RF ICs operating at higher frequencies.
Acknowledgements The authors would like to thank Dr. Cheon Soo Kim at ETRI for his valuable discussions. This work is supported by the Ministry of Information & Communication of Korea (``1999 Support Project of University Foundation Research'' supervised by IITA). References [1] Camilleri N, Costa J, Lovelace D, Ngo D. Silicon MOSFETs, the microwave device technology for the 90s. IEEE MTT-S-Int Microwave Symp Dig 1993:545±8. [2] Voinigescu SP, Tarasewicz SW, MacElwee T, Ilowski J. An assessment of the state-of-the-art 0.5 lm bulk CMOS technology for RF applications. Tech Dig IEEE Int Electron Dev Meet 1995:721±4. [3] Lovelace D, Costa J, Camilleri N. Extracting small-signal model parameters of silicon MOSFET transistors. IEEE MTT-S Int Microwave Symp Dig 1994:865±8. [4] Lee S, Yu HK, Kim CS, Koo JG, Nam KS. A novel approach to extracting small-signal model parameters of silicon MOSFETs. IEEE Microwave and Guided Wave Lett 1997;7(9):75±7. [5] Lee S, Yu HK. Parameter extraction technique for the small-signal equivalent circuit model of microwave silicon MOSFETs. Proc IEEE/Cornell Conf Adv Concepts in High Speed Semicond Dev and Circuits. 1997:182±91.
363
[6] Raskin JP, Dambrine G, Gillon R. Direct extraction of the series equivalent circuit parameters for the small-signal model of SOI MOSFET's. IEEE Microwave and Guided Wave Lett 1997;7(12):408±10. [7] Lee S, Yu HK. Accurate high-frequency equivalent circuit modeling of silicon MOSFETs. Electron Lett 1999;35(17):1406±8. [8] Liu W, Gharpurey R, Chang MC, Erdogan U, Aggarwal R, Mattia JP. R.F. MOSFET modeling accounting for distributed substrate and channel resistances with emphasis on the BSIM3v3 SPICE model. Tech Dig IEEE Int Electron Dev Meet 1997:309±12. [9] Jen SH-M, Enz CC, Pehlke DR, Schroter M, Sheu BJ. Accurate modeling and parameter extraction for MOS transistors valid up to 10 GHz. IEEE Trans Electron Dev 1999;46(11):2217±27. [10] Enz CC, Cheng Y. MOS transistor modeling for RF IC design. IEEE J Solid-State Circuits 2000;35(2):186±201. [11] Ou J-J, Jin X., Ma I, Hu C, Gray PR. CMOS RF modeling for GHz communication IC's. Proc 1998 Symp VLSI Technol. Dig Tech Papers 1998:94±95. [12] Vanoppen RRJ, Geelen JAM, Klaassen DBM. The highfrequency analogue performance of MOSFETs. Tech Dig IEEE Int Electron Dev Meet 1994:173±6. [13] Tin SF, Osman AA, Mayaram K, Hu C. A simple subcircuit extension of the BSIM3v3 model for CMOS RF design. IEEE J Solid-State Circuits 2000;35(4):612±24. [14] Tin SF, Mayaram K. Substrate network modeling for CMOS RF circuit simulation. Proc IEEE Custom Integrated Circuits Conf 1999:583±6. [15] Lee S, Yu HK. A semianalytical parameter extraction of a SPICE BSIM3v3 for RF MOSFETs using S-parameters. IEEE Trans Microwave Theory Technol 2000;48(3):412±6. [16] Chan Y-J, Huang C-H, Weng C-C, Liew B-K. Characteristics of deep-submicrometer MOSFET and its empirical nonlinear RF model. IEEE Trans Microwave Theory Tech 1998;46(5):611±5. [17] BSIM3v3 manual, Department of Electrical Engineering and Computer Science, University of California, Berkeley, 1995. [18] Bagheri M, Tsividis Y. A small-signal DC-to-high frequency nonquasistatic model for the four-terminal MOSFET valid in all regions of operation. IEEE Trans Electron Dev 1985;32(11):2383±91. [19] Tsividis YP. Operation and Modeling of the MOS Transistor. New York: McGraw-Hill; 1987. [20] Vandeloo PJV, Sansen WMC. Modeling of the MOS transistor for high frequency anlog design. IEEE Trans Comput-Aided Des 1989;8(7):713±23. [21] Chan M, Hui KY, Hu C, Ko PK. A robust and physical BSIM3 non-quasi-static transient and AC small-signal model for circuit simulation. IEEE Trans Electron Dev 1998;45(4):834±41. [22] Biber CE, Schmatz ML, Morf T, Lott U, Bachtold W. A nonlinear microwave MOSFET model for spice simulators. IEEE Trans Microwave Theory Tech 1998;46(5):604±10. [23] Raskin JP, Gillon R, Chen J, Vanhoenacker-Janvier D, Colinge J-P. Accurate SOI MOSFET characterization at microwave frequencies for device performance optimization and analog modeling. IEEE Trans Electron Dev 1998;45(5):1017±25.
364
S. Lee, H.K. Yu / Solid-State Electronics 45 (2001) 359±364
[24] Manku T. Microwave CMOS-device physics and design. IEEE J Solid-State Circuits 1999;34(3):277±85. [25] Abou-Allam E, Manku T. A small-signal MOSFET model for radio frequency IC applications. IEEE Trans ComputAided Des 1997;16(5):437±47. [26] Kim CS, Yu HK, Cho H, Lee S, Nam KS. CMOS layout and bias optimization for RF IC design applications. IEEE MTT-S Inter Microwave Symp Dig 1997:945±8.
[27] Van Wijnen PJ, Claessen HR, Wolsheimer EA. A new straightforward calibration and correction procedure for ``on wafer'' high-frequency S-parameter measurements (45 MHz±18 GHz). Proc IEEE Bipolar Circuits and Technol Meet 1987;70±3. [28] Lee S. Eects of pad and interconnection parasitics on forward transit time in HBT's. IEEE Trans Electron Dev 1999;46(2):275±80.