IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 53, NO. 2, APRIL 2004
581
A Novel System for Systematic Microwave Noise and DC Characterization of Terahertz Schottky Diodes Stephan Biber, Oleg Cojocari, Günther Rehm, Bastian Mottet, Manuel Rodríguez-Gironés, Lorenz-Peter Schmidt, Member, IEEE, and Hans L. Hartnagel
Abstract—An automated system is developed to evaluate a large number Schottky diodes for terahertz applications with respect to their dc and noise characteristics using a highly sensitive noise measurement technique for one port devices. An extensive RF switching matrix allows noise characterization of one port devices at selected frequency points over a bandwidth from 2 to 8 GHz. The measurement principle also accounts for the impedance mismatch between the system and the device under test (DUT). Furthermore, the setup includes an automated three-axis nanopositioning system capable of consecutively contacting many Schottky diodes arranged in a honeycomb array. The highly accurate positioning of the DUT allows to create reproducible contacts with the diodes using electrochemically etched whisker tips. The smooth contacting procedure enables several hundred contacts with the same whisker tip. With this system, we evaluate the statistical distribution of dc and noise parameters of Schottky diodes with an anode diameter of 1 m within one honeycomb chip. The system helps in optimizing the production parameters of Schottky diodes for terahertz frequencies. Index Terms—Terahertz (THz), Schottky diodes, noise.
I. INTRODUCTION
F
OR FREQUENCIES exceeding 1 THz, the development of Schottky diodes with submicron diameter has become an important issue. One of the most critical parameters of these diodes is their junction capacitance, which has to be in the femto-Farad range in order to make the diode an effective mixing element. The diode size must be reduced in order to keep the junction area as small as possible [1]. Diodes for frequencies between 500 and 1000 GHz usually have diameters in the range from 0.8–1.2 m and capacitances in the fF range. Higher frequencies require anode diameters in the 0.5–0.25- m range [2], [3]. These diodes are usually arranged in a honeycomb array as shown in Fig. 1. The diodes in a honeycomb structure have a common cathode on the rear side, while the circular anodes are structured on the front side with a Pt-GaAs Schottky contact which is covered by a thin layer of gold. The anodes are etched into the SiON passivation layer so that small holes of about 300 nm in depth are created to guide the whisker Manuscript received August 1, 2002; revised August 31, 2003. This work was supported in part by the Deutsche Forschungsgemeinschaft. S. Biber and L.-P. Schmidt are with the Lehrstuhl für Hochfrequenztechnik, University of Erlangen-Nuremberg, 91058 Erlangen, Germany. O. Cojocari, B. Mottet, M. Rodríguez-Gironés, and H. L. Hartnagel are with the Institut für Hochfrequenztechnik, Technical University of Darmstadt, Darmsdadt, Germany. G. Rehm was with the Lehrstuhl für Hochfrequenztechnik, University of Erlangen-Nuremberg, 91058 Erlangen, Germany. He is now with Diamond Light Source Ltd., Didcot, U.K. Digital Object Identifier 10.1109/TIM.2003.820487
tips. The anodes are contacted with electrochemically etched whisker tips. Pushing the upper limit for these devices toward higher frequencies is inevitably connected with reducing the diode size. This mainly causes three different problems. The production of the diodes has to be optimized for very small anode diameters in order to achieve diodes with acceptable mixer performance. The excess noise [4], [5] emitted from the diodes increases with increasing current density and contributes the major part to the overall noise [6] of the diode. Contacting diodes with anode diameters of 0.5 m or less becomes increasingly difficult. In order to optimize the production parameters of the Schottky diodes for minimum noise and ideal dc characteristics, a systematical evaluation of these parameters is needed. Although earlier studies have investigated these parameters, no data is available on the statistical distribution of the quality of Schottky diodes within one honeycomb array. The contacting procedure in a real terahertz-mixer system does not allow to choose a specially selected diode. Usually, the whisker is blindly placed somewhere on the chip enabling an electric contact with a randomly chosen diode. Considering this procedure, it becomes even more important to get insight into the statistical variations of diode parameters within one chip, instead of characterizing only a few randomly chosen devices. The system introduced here combines sensitive noise characterization at IF frequencies with dc measurements and accurate nanopositioning. The three-axis positioning system moves the diode samples relative to the whisker tips and contacts one diode after another. To our knowledge, this is the first system capable of automatically contacting Schottky diodes with anode diameters as small as 1 m and measuring their dc and noise parameters at selected frequency points distributed over a bandwidth of more than one octave. The whole contacting and measurement procedure is fully automated and all instruments are controlled from one computer. The positioning unit can be moved with a minimum step width of 50 nm. The very carefully controlled contacting procedure makes it possible to use the same whisker tip for contacting many hundred diodes. II. THEORY OF NOISE MEASUREMENT As opposed to two-port devices such as amplifiers, whisker contacted diodes are one-port devices which can not be characfactor method. We, therefore, terized with the well-known present an extended theoretical derivation of a measurement method proposed by [7]. The basic circuit diagram is shown in Fig. 2. The noise of an electronic noise source (NS) is coupled
0018-9456/04$20.00 © 2004 IEEE
582
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 53, NO. 2, APRIL 2004
Fig. 2. Block diagram of measurement setup. Fig. 1. Scanning electron microscope (SEM) image of the anodes of ahoneycomb diode array. Backscatter electrons (BSE) were detected in order to achieve a tomographic image of the surface.
by a circulator to the device under test (DUT). The noise generated by the DUT itself and the part of the noise which is reflected from the DUT due to impedance mismatch between the DUT and the measurement system is coupled to a low noise amplifier (LNA). The four unknown parameters which have to be evaluated during the measurement procedure are as follows: equivalent noise of low noise amplifier in K; gain of amplifier; reflection coefficient of the DUT; noise temperature of DUT in K; bandwidth of the receiver The four measurement steps necessary to solve for the unknown parameters are carried out under the following conditions: 1) case a: , NS on: ; , NS off: ; 2) case b: , NS on: ; 3) case c: , NS off: . 4) case d: A simplified model, which does not account for the losses in the passive circuit, gives better understanding of the relationship between the different parameters (1) (2) (3) (4)
noise power measurements the diode noise temperature can be derived from the simplified model as follows: (5) Precise results can only be achieved if the complete three-port network is characterized in terms of scattering-parameters. A comprehensive model including the attenuation and the reflection coefficients of the whole circuit and the noise emitted from the lossy components within the system is needed. The analytical model described below only includes one simplification in order to reduce the analytical effort. If a wave traveling toward one port is reflected, it will couple back into the circulator and will be completely absorbed before it comes back to the same port again. That means that this model as opposed to [7] also accounts for reflections of first order at all of the ports. The following equations express the power measured by the power detector as a function of the parameters, the noise temperature of the noise source , the bandwidth , the physical temperature of the system , and the above mentioned quantities , and (6) (7) being the incident power at port two and with noise temperature seen by the LNA. Assuming follows:
being the , it (8) (9) (10)
noise temperature of the noise source at state “hot”; noise temperature of the noise source at state “cold”; reflection coefficient of the DUT; Boltzmann’s constant. For and , it is assumed that the power coming from the noise source is completely reflected at the diode (the unbiased diode behaves like an open circuit). Using these four
Inserting the correct parameters for the cases a-d with and for case b and d gives a unique solution for the noise temperature of the device under test. then . can easily be derived becomes equivalent to from the set of linear equations given above. It is important to emphasize that the whole system is based on the measurement of the magnitude of the reflection coefficient.
BIBER et al.: NOVEL SYSTEM FOR SYSTEMATIC MICROWAVE NOISE AND DC CHARACTERIZATION
583
In order to avoid effects caused by phase shifts and noise correlation, it is very important to determine the frequencies for which the isolation for all of the paths is maximal. III. EXTRACTION OF DC PARAMETERS To describe the dc performance of the diode, the reverse curis rent , the ideality factor , and the series resistance derived from (8) (11) As opposed to other techniques (e.g., four-point measurement technique), , and are derived by a curve-fitting algorithm which fits the measured data to the analytical model. The advantage of this method is that the entire dc characteristic (about 100 measurement points) is included in the calculation , and . The – curve is recorded up to a maximum of current of 3 mA to prevent diode burnout. IV. AUTOMATED MEASUREMENT SYSTEM A block diagram of the completely automated system is shown in Fig. 2. Additional to the system shown in Fig. 2, the real system consists of a second circulator for each frequency band. This circulator in front of the amplifier acts as an isolator. It ensures high backward isolation between the amplifiers and the DUT and matches the LNA to 50 . During the development of the system, it turned out that accurate noise temperatures independent of the reflection coefficient of the DUT can only be acquired if isolation between the LNA and the DUT is very high. Otherwise, the noise emitted from the LNA and from the noise source travels to port three on at least two different paths and can cause undesired effects due to noise correlation. The amplifiers are connected to a spectrum analyzer, which 10 MHz. integrates the noise over 3 s in a bandwidth of The spectrum analyzer is tuned to the frequencies, for which the above mentioned conditions are fulfilled and, finally, acts as a tunable narrow-band filter with power detector. All RF connections are realized in semirigid-technology to ensure mechanical stability. Fig. 3 shows a typical set of parameters of the system. It is evident that suitable frequencies for accurate measurements have to be carefully selected. In this case, 2.17 GHz, 2.85 GHz, and 3.85 GHz are chosen because isolation for all 45 dB, 20 dB, the paths is very high ( and 25 dB). The frequencies for the second band from 4–8 GHz are chosen with similar criteria. Computer-controlled coax switches are used to change from the 2–4 GHz to the 4–8 GHz band. The coax switches are tested with respect to their repeatability and it turned out that even after several switching operations the change in the parameters of the switches is negligible. All other instruments such as the precision voltage source, the ampere meter, the motor controller for the nanopositioning system and the spectrum analyzer are controlled by the same computer via a GPIB bus interface. The contacting procedure can be monitored by a long-distance microscope, which is connected to an analog CCD camera. The chip with the diodes is soldered on a gold plated
Fig. 3. Selected s parameters for 2–4 GHz path of the measurement system.
copper substrate. The substrate is then mounted on a chuck, which is moved by the positioning system. Whisker wires are soldered to the inner and outer conductor of a semirigid cable in order to contact the diodes. The setup is tilted so that the contacting procedure can be observed through microscope. As shown in Fig. 2, the whisker wires are not included in the parameter measurements used to calibrate the system because no suitable calibration standard exists which allows an accurate connection to the whisker tips. In order to minimize the influence of the whisker wires, their length is kept smaller than 2 mm, which is significantly smaller than ( /10) at IF frequencies. The automated measurement procedure consists of three consecutive parts. 1) The chip with the diodes is moved toward the fixed whisker tips, until an electrical contact is determined by the ampere meter. The diodes are then moved another 10 m toward the whisker tips in order to press the whisker on the anode and ensure good ohmic contact. 2) The dc parameters are measured by sweeping the voltage until a threshold current of 3 mA is reached. 3) The noise parameters are measured for the desired frequency points and bias currents. One of the major issues related to the performance of the system is the question whether it is generally possible to use the same whisker contact for many diodes or if the blind contacting procedure will damage the whisker after a few contacts because of the unevenness of the surface and the high mechanical pressure which is applied to the fine tip. The used whisker is made of 25 m AuNi wire which was electrochemically etched to form a sharp tip. The porous surface is caused by removing the wire from the acid during the etching process. Fig. 4 shows a scanning electron microscope (SEM) image of the same whisker before and after more than 200 contacts. The tip appears to be slightly flatened but it still is sharp enough to contact diodes with 1- m diameter.
V. REPEATABILITY OF THE CONTACTING PROCEDURE The extremely small size of the diodes makes the contacting procedure an important question. The repeatability of these measurements is checked by contacting the same diode many times, and by repeating the measurement without recontacting
584
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 53, NO. 2, APRIL 2004
Fig. 5. Ideality factor n and series resistance R . The same diode was contacted over 40 times. std (n) 0.0072; std (R )=0.957.
=
Fig. 4. SEM images of whisker wires. top: overview image (1000 left: after etching (9000 ); bottom right: after 200 contacts (9000
2
2); bottom 2).
the diode. Fig. 5 shows the measured dc parameters and of a diode which was contacted over 40 times and characterized with respect to its dc performance. Fig. 6 shows another diode measured 150 times without recontacting. In both experiments, the diode characteristics converge after a number of measurements, which proves the repeatability of the developed dc characterization procedure. The major uncertainty is induced by recontacting the diode. A standard deviation for the series resistance and standard deviation of 0.118 of 0.0021 for the ideality factor characterizes the accuracy of the system. The measurements demonstrate that the setup is capable of recontacting the same diode with reproducible contacts. The noticeable but small variations observed during the first measurements can be ascribed to actual changes in diode characteristics in a sort of burn-in effect that should be investigated more precisely. Considering the SEM images together with these measurements, a significant degradation of measurement accuracy with increasing number of contacts due to mechanical deformation of the whisker tip is not to be expected.
Fig. 6. Ideality factor n and series resistance R . The diode was contacted only once and then the dc measurements were repeated 150 times.
TABLE I
1T IN K VERSUS FREQUENCY
VI. VERIFICATION OF THE NOISE MEASUREMENTS In order to determine the absolute accuracy of the measurement system and the temporal variations of the acquired noise temperatures, a noise source with known noise temperature is needed. Conventional noise sources are not suitable because the calibration data are not accurate enough to determine differences below 20 K. We, therefore, use two different methods for the verification of the system. The first method uses, instead of the DUT, a 50- SMA load cooled down to liquid nitrogen temperature (77 K). At each frequency, 36 measurements are performed within 5 min. Table I shows the measured mean dif77 K. ference The extremely small deviations show that the system is very between the expected temstable over time. The difference perature of 77 K and the measured temperature is less than 40 K for all frequencies. A typical plot of 120 measurements at
Fig. 7. A 12-min record of the measured noise temperature at 2.1 GHz.
2.1 GHz within 12 min is shown in Fig. 7. The standard deviation of 0.25 K demonstrates the extreme stability of the system over time. The second verification procedure uses a calibrated noise diode as DUT and compare the calibration data with our measurements. For convenience the measured noise temperature is plotted in effective noise ratio (ENR). ENR
K
(12)
In order to get more than one plot, a precisely calibrated variable attenuator (0 dB, 3 dB, 6 dB, 10 dB, 16 dB, 20 dB) is connected
BIBER et al.: NOVEL SYSTEM FOR SYSTEMATIC MICROWAVE NOISE AND DC CHARACTERIZATION
Fig. 8. Comparison of calibration data of a noise source with different attenuators versus measurements from our new system.
585
Fig. 9. Comparison of two different diode samples with respect to their dc parameters.
to the 20-dB ENR noise source. The results shown in Fig. 8 take the noise contribution from the attenuator into account. This calibration procedure shows excellent agreement between the calibration data and our measurement system. The maximum error between the results from the system and the calibration data detected for all the 48 measurements in Fig. 8 is 4.6%. The mean error is less than 2%. More interesting in this verification procedure is the investigation of the relative error defined as
(13) A similar definition holds for higher attenuations. For all the six different noise temperatures, which are available from the calibrated noise source, the difference between one noise temperature and the next lower noise temperature is compared with the measured difference. This procedure is performed for all eight frequencies listed in Table I, which gives 40 different values for the relative error as defined above. A mean relative error of only 2% gives evidence of the high accuracy of the system. The first calibration procedure shows that the absolute noise temperature can be measured with an accuracy better than 20 K excluding the third and eighth frequency. The second calibration shows that the mean relative accuracy of the measurements is within 2%. VII. RESULTS Fig. 9 shows a scatterplot of the dc parameters for Schottky diodes that were fabricated using slightly different processes. The diodes under investigation have anode diameters of 1 m. The Schottky contacts are located on top of a GaAs wafer with a thickness of 70 m. It can clearly be seen that the diodes from sample 2 have better dc characteristics in terms of the series resistance and the ideality factor. Their parameters are also less scattered than those for sample 1. The differences in the mean ideality factor and the mean series resistance from both sam-
Fig. 10. Noise measurement for 69 different diodes as a function of frequency and bias current.
ples are significantly higher than the derived measurement uncertainty. This demonstrates that changes in the production parameters of the diodes can be monitored statistically with the new measurement system. In addition to the dc measurements, systematic noise measurements are made for Schottky diodes with an anode diameter of 1 m. Fig. 10 shows the results for 69 different diodes taken from one sample chip. The noise is measured as a function of frequency and bias current. The vertical bars indicate the standard deviation of the noise temperature for a certain frequency and bias current. Two different trends can be observed. First, there is a clear increase of noise temperature with bias current. This is due to the increasing current densities which make excess noise become the predominant noise source in the diode. Second, there is a less-pronounced, but still visible, decrease of noise temperatures with increasing frequencies. This is again due to excess noise which is stronger at low frequencies. Both phenomena are consistent with the noise mechanisms described by [7] and [9]. We have no physical interpretation for the increase of noise temperatures at 5.6 GHz compared to the lower
586
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 53, NO. 2, APRIL 2004
frequencies. The slightly higher noise temperatures there might be due to small errors in the calibration.
VIII. CONCLUSION It is shown that a very sensitive contacting procedure allows reproducible contacts to the 1 m small devices. The error in the dc characteristics, which is due to the contacting uncertainty, is negligible. A three-port theory for the noise measurement system including reflections of first order together with a well-calibrated measurement system gives precise and stable results for the noise temperature of the diode. Two different verification procedures show good agreement between theory and measurement. The system accounts for the impedance mismatch between the DUT and the measurement system. The contacting procedure and the measurement system are fully automated and controllable from a single computer. Typical dc and noise measurements have demonstrated the capability of the system to contact many diodes and to perform consistent dc and noise measurements.
ACKNOWLEDGMENT The authors would like to thank Mrs. Weiss and Mrs. Koch at the Institute of General Materials Properties, Erlangen, Germany, for helping with the SEM images, as well as the mechanical workshop at the Lehrstuhl für Hochfrequenztechnik, especially Mr. Wick and Mr. Bauer, for helping with the assembly of the semirigid cables and the mechanics.
REFERENCES [1] T. W. Crowe, “GaAs Schottky barrier mixer diodes for the frequency range 1–10 THz,” Int. J. Infrared Millim. Waves, vol. 10, no. 7, pp. 765–777, 1989. [2] K. Huber, R. Engelbrecht, R. Hocke, M. Raum, H. Brand, S. Martius, and L.-P. Schmidt, “A broadband, low noise heterodyne receiver for stratospheric measurements at 2.5 THz,” in Proc. IGARSS, Hamburg, Germany, June 1999. [3] H. W. Hübers, H. P. Röser, and G. W. Schwaab, “A heterodyne receiver for the frequency range 1–6 THz,” in Proc. 30th ESLAB Symp. Submillimeter and Far-Infrared Space Instrumentation, Noordwijk, The Netherlands, Sept., 24–26 1996, pp. 159–162. [4] G. Rehm, K. Huber, and S. Martius, “Excess noise in Schottky diodes for THz applications,” in Proc. 6th IEEE Int. Conf. Terahertz Electronics, Leeds, U.K., Sept. 1998, pp. 131–134. [5] K. M. Kattmann, T. W. Crowe, and R. J. Mattauch, “Noise reduction in GaAs schotky barrier mixer diodes,” IEEE Trans. Microwave Theory Tech., vol. 35, pp. 212–214, Feb. 1987. [6] T. J. Viola and R. J. Mattauch, “Unified theory of high frequency noise in Schottky barriers,” J. Appl. Phys., vol. 44, pp. 2805–2808, 1973. [7] S. Palczewski, A. Jelenski, A. Grüb, and H. L. Hartnagel, “Noise characterization of Schottky barrier diodes for high-frequency mixing applications,” IEEE Microwave Guided Wave Lett., vol. 2, pp. 442–444, Nov. 1992. [8] S. M. Sze, Physics of Semiconductor Devices, 2nd ed. New York: Wiley, 1981. [9] A. Jelenski, E. L. Kollberg, and H. H. G. Zirath, “Broad-band noise mechanisms and noise measurements of metal-semiconductor junctions,” IEEE Trans. Microwave Theory Tech., vol. 34, pp. 1193–1201, Nov. 1986.
Stephan Biber received the M.Sc. degree in electrical engineering from the University of Colorado, Boulder, in 2000, where his research was focused on microwave technology and the application of microwave remote sensing for oceanography and climate-change monitoring. Since 2001, he has been a Research Assistant at the Institute of Microwave Technology, University of Erlangen-Nuremberg, Erlangen, Germany, where he is involved with millimeter and submillimeter wave technology. His work includes the improvement of Schottky-diode mixers for frequencies up to 2.5 THz, silicon-micromachining of quasioptical devices, and electrooptical effects in silicon.
Oleg Cojocari received the degree from the Baltsy Radio-technical College, Moldova, and the degree of Engineer in microelectronics and computer science from the Technical University of Moldova, Chisinau, in 1990 and 1995, respectively. From 1996 to 2000, he was with the Department of Microelectronics and the Laboratory of Low Dimensional Semiconductor Structures, Technical University of Moldova. Since 2000, he has been with the Department of Microwave Engineering, Darmstadt University of Technology, Darmsdadt, Germany, where he is involved in technology development of the Schottky-based terahertz devices. His research interests include the low-noise Schottky contacts and device fabrication technology for terahertz frequency operation based on GaAs and GaN material systems.
Günther Rehm received the diploma degree and the Ph.D. degree in engineering from the Friedrich-Alexander University of Erlangen-Nuremberg, Erlangen, Germany, in 1996 and 2002, respectively. Between 1996 and 2002, he was with the Institute for Microwave Technology at the Friedrich-Alexander University of Erlangen-Nuremberg. His research interests were included imaging using radiometry at submillimeter wavelengths. Since 2003, he has been with Diamond Light Source, Ltd., Didcot, UK, where he is responsible for electron beam diagnostics at a 3 GeV synchrotron.
Bastian Mottet was born in Krefeld, Germany, in 1974. He received the Dipl.-Ing. degree in electrical engineering from the Technical University of Darmstadt (TUD), Darmstadt, Germany, in 1999. He is currently pursuing the Ph.D. degree as a Research Associate at the Institute of Microwave Engineering, TUD. His research concerns reliability aspects of integrated III-V compound semiconductors and their contact systems and the development of devices for terahertz applications.
Manuel Rodríguez-Gironés, photograph and biography not available at the time of publication.
Lorenz-Peter Schmidt (M’87) received the Dipl.-Ing. degree from the Technical University of Aachen, Aachen, Germany, in 1974 and the Ph.D. degree on the analysis of transverse discontinuities in microstrip lines from the High Frequency Technology Institute, RWTH, Aachen. Between 1974 and 1979, he was with the High Frequency Technology Institute, RWTH, as a Research Assistant. In 1979, he was a PostdoctoralResearch Associate at the University of Texas, Austin. From 1980 on, he was with AEG-Telefunken (later DASA, now EADS), where he later became the Head of several microwave and millimeter wave research groups. Since 1998, he has been Full Professor and Head of the Institute for Microwave Technology at the University of Erlangen-Nuremberg, Erlangen, Germany. Dr. Schmidt is member of the German IEEE MTT/AP-Chapter Commission.
BIBER et al.: NOVEL SYSTEM FOR SYSTEMATIC MICROWAVE NOISE AND DC CHARACTERIZATION
Hans L. Hartnagel received the Dipl.-Ing. degree from the Technical University of Aachen, Aachen, Germany, and the Ph.D. and Dr. Eng. degrees from the University of Sheffield, Sheffield, U.K., in 1960, 1964, and 1971, respectively. He received the Dr. h.c. from the University of Rome Tor Vergata, Rome, Italy, and the Dr. h.c. from the Technical University of Moldova, Kishinev, in 1994 and 1999, respectively. After being with Telefunken, Ulm, Germany, he joined the Institute National des Sciences Appliquées, Villeurbanne, Rhône, France, and then the Department of Electronic and Electrical Engineering, University of Sheffield, as a member of staff. In January 1971, he became Professor of electronic engineering at the University of Newcastle upon Tyne, U.K. Since October 1978, he has been a Professor of high frequency electronics at the Technical University of Darmstadt, Darmstadt, Germany. He is the author of several books and numerous scientific papers on microwave semiconductor devices and their technology and circuits. He has held many consulting positions, partly while on temporary leave of absence from his university positions.
587