PIERS ONLINE, VOL. 3, NO. 2, 27 184 De-embedding Techniques For Passive Components Implemented on a.25 µm Digital CMOS Process Marc D. Rosales, Honee Lyn Tan, Louis P. Alarcon, and Delfin Jay Sabido IX Department of Electrical and Electronics Engineering University of the Philippines, Diliman, Quezon Ctiy 111, Philippines Abstract On wafer measurement and characterization of passive components implemented on CMOS technology is one of the initial activities in implementing RF circuits on CMOS. Using test fixtures is necessary in order to test inductors and capacitors at GHz frequencies. However, these fixtures introduce significant effects on measured parameters. This study focuses on the OPEN and THREE step de-embedding techniques used for on wafer measurement of passive devices. Different values of inductors and capacitors were used as devices under test (DUT). For each component type, a set of structures for the two de-embedding were fabricated. All of these structures were fabricated on a.25 µm Digital CMOS process. DOI: 1.2529/PIERS6962246 1. INTRODUCTION Test fixtures are used to facilitate the on-chip probing and measurement of passive components implemented in CMOS devices. However, these test fixtures have significant effects on the measured parameters of the devices. These effects can be cancelled out by using different de-embedding techniques. These techniques will use several de-embedding structures and offline post processing of the measured data. A typical on wafer characterization follows these series of activities: (a) preparing the measurement setup, (b) calibration, (c) on-chip probing and measurement, (d) de-embedding and (e) parameter extraction. 2. IMPLEMENTED DEVICES To facilitate the study, inductors [1] and capacitors [2] are implemented on a.25 µm Digital CMOS process. Two inductor structures are used for this study: a square 1.8nH inductor and an octagonal 3. nh inductor. Two capacitor structures are also used: a 635 ff horizontal parallel plate capacitor (HPP) and a 584 ff woven capacitor structure. In order to measure the devices under test, a Ground Signal Ground (GSG) fixture is used. This serves as an interface to the device port and probe tips. Different fixtures are used for each type of device under test. The GSG probe tip arrangement is advised for measurement beyond 18 GHz, and required beyond 26 GHz [3]. Some guidelines are followed in the design of fixtures [4]. Figure 1(a) shows the actual GSG fixture used to hold the inductors. The upper ground pads of the fixtures are joined with metal strips from metal layers 5, 4, 3, 2 and 1. The metal strips are joined together by dense via array. The ground structure provided the contact to ground of the substrate from the top. The same arrangement is followed in the lower ground pads. The signal pads are implemented by stacking metal layer 5, 4, 3, 2 and 1 together with the use of dense via array. Another GSG fixture is fabricated to facilitate the on-wafer measurement of the capacitor structures. (a) (b) (c) (d) Figure 1: (a) Open (b) Thru (c) Short 1 (d) Short 2.
PIERS ONLINE, VOL. 3, NO. 2, 27 185 3. MEASUREMENT SETUP A typical setup for doing on-wafer measurements for RF application consists of several equipments: an IC probe station, Probe link arms, GSG probes, semi- rigid cables and Vector Network Analyzer. The laboratory setup consists of an Agilent 8753ES Vector Network Analyzer (3 KHz 6 GHz), Micromanipulator IC Probing Station, Microwave Probe link arms, GSG Picoprobes 4AGSG- 16P with 16 microns pitch. The silicon substrate is also grounded from the back side through the testing chuck. The network analyzer is configured to perform broadband passive component measurements. The power setting is set to dbm and the sweep frequency is set from 3KHz up to 6 GHz. A total of 21 points are used to represent the S-parameters of the whole sweep range. A Touchtone file format in real and imaginary components is chosen to be the output format of the network analyzer. The measurement setup is calibrated using a GGB industries CS-5 calibration substrate. A Short-Open-Load-Thru (SOLT) technique is used for the measurement calibration. 4. DE-EMBEDDING De-embedding is a process of eliminating the influence of the transition region between the probe, probe contact and the device under test. It is mathematically subtracting networks from the measured results. A pre-requisite for a correct deembedding is that certain test structures are available on a wafer together with the DUT itself. Depending on the selected de-embedding method, open short and thru structures are required and must be measured. Figure 1 shows the de-embedding structures used for the inductor measurements. Another set of de-embedding structures is also fabricated for the capacitor measurements. 4.1. Open De-embedding Open de-embedding (OPD) is the most widely used because of its simplicity. Only one set of open pads are required to perform de-embedding. This method assumes that the parasitics leading to the DUT are parallel admittance y p [4]. It is performed by subtracting the Y -Parameter of the open fixture only to the Y -Parameters of the DUT inside the fixture to obtain the Y -parameters of the DUT only. De-embedding the fixture from the DUT inside the fixture was performed by subtracting the admittance y p from the admittance of the DUT attached to the fixture. The Y -parameters of the open fixture is estimated by the admittance y p. (1) is performed for every frequency point. 4.2. Three Step De-embedding Three step de-embedding (TRISD) offers a more general form of de-embedding and calibration. It uses a more complicated model topology for the test fixtures. All parasitics associated with probe pads and interconnect-metal lines can be represented and can be subtracted out of the measurement [5]. This requires the use of an open, thru, short 1, and short 2 structures as can be seen in Figures 1. Three-step de-embedding includes the following operations: (a) Subtract the on-wafer Y -parameters shunting the input and output ports. (b) Subtract the Z-parameters taken from the thru and shorts. (c) Subtract the coupling capacitance between the input and output lines. A detailed discussion of the computation of OPD and TRISD is in [5]. 5. INDUCTOR MEASUREMENTS The inductors inside the fixture are measured following the procedures discussed in the earlier sections. Figure 2 shows the measured S-parameters of the square and octagonal inductors together with the effects of the test fixture. The plot shows how the fixture influences the measured S11 of the inductors. The S-parameters of all the de-embedding structures for the inductor measurement are also measured. The measured S-parameters are processed offline using de-embedding functions written using MATLAB. Figure 3 shows the open de-embedded and three step de-embedded S-parameters of the square inductor. The S-parameter plots of both techniques are similar. In the octagonal inductor, it was also observed that the results of both de-embedding techniques are similar. Parameter extraction is the last stage in the postprocessing of the measured data. Using the deembedded S-parameters, a simple pi model of the inductor can be used to extract its inductance and quality factor.
PIERS ONLINE, VOL. 3, NO. 2, 27 186 ( Im 1 y 12de embeded(ω) L s = ω Q = Im(y 11de embeded(ω)) Re(y 11de embeded(ω) ) ) (1) (2) j.5 j.5 j.2 j.2 -j.5 -j.5 (a) (b) Figure 2: Measured S-Parameters with GSG Fixture (a) square inductor (b) octagonal inductor. j.5 j.5 j.2 j.2 -j.5 -j.5 (a) (b) Figure 3: De-embedded S-parameters of the square inductor (a) OPD (b) TRISD. The square data points are the extracted Q from an open de-embedded measurement while the circle data points are the extracted Q from a three-step deembedded measurement. Initial INDUCTOR Table 1: OPD VS. TRISD OPD TRISD PEAK FREQ (GHZ) PEAK FREQ (GHZ) square 3.15 5.16 3.15 5.16 octagon 3.33 3.96 3.33 3.96
PIERS ONLINE, VOL. 3, NO. 2, 27 187 inspection shows that the extracted Q from both de-embedding methods have similar values. The extracted inductance of the three-step de-embedded measurements exhibits a slightly higher value at higher frequencies. The same trend is also observed for the octagonal inductor. Table 1 shows a summary of the maximum Q of the inductors and the frequency it occurs. The values show the de-embedding technique used does not affect the extracted peak inductor Q. 6. CAPACITOR MEASUREMENTS The S-parameters of the woven capacitor are illustrated in Figure 4. The CADENCE plots of the S11 and S12 parameters are the simulation results acquired from the capacitor without the test fixture. The With Fixture plots represent the measured S11 and S12 parameters before de-embedding. The deembedded S12 resulting from the TRISD procedure follows the Cadence simulator s S12 plot, while the OPD S12 has a small deviation from the simulation result. Both the S11 results of the OPD and the TRISD techniques deviate from that of the Cadence s simulation result. The deviation becomes significant as the frequency increases. Nevertheless, it is seen that de-embedding techniques can eliminate the effect of the fixture in the measured S-parameters and approximate the measurement to the S-parameters from the Cadence simulation. The rest of the S-parameter plots of the implemented capacitors follow the same trend the extracted capacitance of the selected tests is obtained from Im[1/-Y 12] and plotted with respect to frequency. The effect of the two de-embedding techniques in the extraction of the capacitances is observed. The capacitances at 1.2 GHz and 2.4 GHz are extracted for each de-embedding technique as presented in Figure 5 and the resulting capacitance plots obtained from OPD and TRISD showed that the difference is minimal for the small valued capacitors. Figure 4: S-parameters of the woven capacitor. Figure 5: Capacitances at 1.2 GHz and 2.4 GHz. Further analysis on the behavior of the capacitance is done by getting the capacitances at 2.34 GHz and 2.46 GHz and compared to the capacitance at 2.4 GHz. The percent change (% ) is computed, for both de-embedding methods, based on the center frequency of 2.4 GHz, given a bandwidth of 12 MHz. The % metric shows the behavior of the capacitance operating at the frequency of interest within the given bandwidth of 12 MHz. The lower the % the more accurate the capacitor value within the given frequency specification. The capacitances at 2.4 GHz are selected because this is the center frequency used for a Low Noise Amplifier (LNA) developed in the lab. Table 2 enumerates the computation of the % for both OPD and TRISD methods. The capacitor values generally show that all the fabricated capacitors, based on the capacitance at 2.4 GHz, have a % of less than ±1%. The computed % resulting from the OPD data and TRISD data have comparable values implying that any of the two deembedding techniques can be used in extracting the capacitance. The analysis is extended by computing % Cave given by (3). C opd is the capacitance using OPD, C trisd is the capacitance using TRISD, i is the capacitance at the ith frequency and j is the total frequency points from 27 MHz to 3 GHz This is the percent change of the capacitance
PIERS ONLINE, VOL. 3, NO. 2, 27 188 Table 2: OPD VS. TRISD Capacitor OPD % TRISD % 2.34 2.4 GHz 2.4 2.46 GHz 2.34 2.4 GHz 2.4 2.46 GHz % C ave (%) HPP.44 -.56.49 -.6 -.869 Woven.27 -.34.36 -.42-1.46 after the TRISD with respect to the capacitance obtained from the OPD. The frequency range of 27 MHz to 3 GHz is chosen since almost all of the capacitances within this range do not change abruptly as can be seen in the capacitance plots. % C ave = j i=1 ( ) Copd C trisd C opd Equation (3) is applied to all structures and the results are listed also in Table 2. The smallest % Cave for the small capacitor structure is.86% (the negative sign implies that the TRISD measurement is greater than that of OPD for the HPP capacitor structure. Since the highest % Cave is around 2%, it is therefore suggested that any of the two de-embedding techniques may be used to extract the capacitance. 7. CONCLUSION A methodology of doing on wafer measurement of passive devices is presented. The inductor and capacitor test structures illustrated and validated the said methodology. Empirical results show that for measurements up to 6 GHz the variation extracted parameters using open and three step de-embedding method is nominal. The preferred de-embedding technique would be open de-embedding. The advantages OPD offers against the TRISD are the following: (a) OPD mathematical approach is straight forward (b) The concept and implementation is simple compared to TRISD (c) The three-step de-embedding structures required for TRISD can be eliminated, resulting in a smaller chip area f. ACKNOWLEDGMENT The authors wish to acknowledge the assistance and support of Intel Philippines, DOST PCASTRD, IMI and Eazix Inc.. REFERENCES 1. Rosales, M. D., Characterization, comparison and analysis of monolithic inductors in silicon for RF IC s, M.S. Thesis, University of the Philippines, April 23. 2. Tan, H. B., Characterization, comparison and analysis of monolithic capacitors in silicon for RF IC s, M.S. Thesis, University of the Philipines, April 24. 3. Layout rules for GHz-probing, Application Note, Cascade Microtech Inc. 4. Kolding, T. E., On-wafer calibration tecniques for giga-hertz CMOS measrurments, presented at Int l Conf. on Microelectronic Test Structures, 1999. 5. Cho, H. and D. Burk, A Three-step Method for the De-embedding of high frequency S- parameter Measurements, IEEE Transactions of Electron Devices, 1991. j (3)