656 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 51, NO. 4, AUGUST 2002 A New Method for High-Sensitivity Noise Measurements Carmine Ciofi, Felice Crupi, and Calogero Pace Abstract—A new method for high-sensitivity noise measurement, based on the elaboration in the time domain of the signals coming from two identical amplifiers, is presented in this paper. The most important advantage of this method, which in most cases allows obtaining an equivalent background noise 30 dB below that of the amplifiers which are used in the measurement chain, lies in its simplicity and in the fact that it does not depend on the method used for the estimation of the noise spectra. Index Terms—Correlation, noise measurement, spectral analysis, time domain analysis. Fig. 1. Block diagram of the spectrum analyzer based on the cross-correlation technique. I. INTRODUCTION I N NOISE measurement systems, the sensitivity of the measurement chain is ultimately limited by the equivalent input noise sources of the input preamplifier coupled to the device under test (DUT). The sensitivity that can be obtained by means of standard commercial spectrum analyzers is not sufficient for a large number of applications, especially in the case of the characterization of the low-frequency noise of modern electron devices [1]. In such cases, very low-noise, high-gain preamplifiers must be used between the DUT and the input of the spectrum analyzer in order to conveniently increase the signal level without introducing a significant supplemental noise contribution. The choice of the input preamplifier is rather challenging, especially because of the dependence of the final supplemental noise contribution on the equivalent impedance of the DUT. To reach the ultimate sensitivity levels, one must often resort to a custom design of the input preamplifier. By means of a careful design, the total noise contribution of the preamplifier can be reduced to that introduced by the active element used in the first stage of the amplifying chain. As a general rule, noise measurements are easily performed when the spectral density of the noise produced by the DUT is at least 10 dB higher than the background noise of the measurement chain in the frequency range of interest. However, the lowest background noise levels that can be obtained are still too high for many applications. Therefore, noise measurement methods capable of overcoming the limitation imposed by the background noise of the input preamplifiers are of great interest. The possibility of performing reliable noise measurements when the noise produced by the DUT is comparable to, or even much lower than, the background noise of the input pream- plifier has been demonstrated using different methods [2]–[6]. These methods take advantage of the noncorrelation of the noise sources of the measurement chain with the noise produced by the DUT. A recently proposed approach employs the topology reported in Fig. 1 [3], [4]. It uses two independent amplifiers connected to the DUT. A discrete Fourier transform (DFT) processor follows the output of each amplifier, and the numerical results are processed in order to obtain an estimate of the cross spectrum of the signals at the output of the two amplifiers, with the result of eliminating most of the contribution of the noise introduced by the amplifiers. In this way, under proper conditions, a direct estimate of the sole noise contribution of the DUT is obtained. In this paper, we present another method for performing reliable very low noise measurements that, although in principle equivalent to the one just described, is based on a rather different approach. The measurement principle, the practical issues involved in its actual implementation, and the advantages and limitations of the method we propose, are discussed in the following sections. II. PROPOSED MEASUREMENT METHOD The circuit configuration of the proposed high-sensitivity measurement system is reported in Fig. 2 for the case of voltage is simultanenoise measurements. The DUT voltage noise ously amplified by two different channels with identical gains, and the two amplified signals are fed to the addition and the and can be subtraction blocks. The output voltages written as (1) Manuscript received May 29, 2001; revised April 30, 2002. The authors are with the Dipartimento di Fisica della Materia e Tecnologie, Fisiche Avanzate and INFM, Messina, Italy (e-mail: ciofi@ingegneria.unime.it). Digital Object Identifier 10.1109/TIM.2002.803080 is due to the input signal and to the equivalent input where is due to the current noise (EICN) of the two amplifiers, equivalent input voltage noise (EIVN) of the first amplifier, and 0018-9456/02$17.00 © 2002 IEEE CIOFI et al.: NEW METHOD FOR HIGH-SENSITIVITY NOISE MEASUREMENTS Fig. 2. 657 Schematics of the proposed voltage noise measurement system. Fig. 3. Schematics of the transimpedance noise amplifiers for the case of current noise measurements. is due to the EIVN of the second amplifier. Therefore, at the output of the addition and subtraction blocks we have (2) At least in the case of low-frequency noise measurements, one , and are uncorrelated and, can assume that therefore, at the output of each channel of the spectrum analyzer in Fig. 2 we have (3) is the power spectral density (PSD) of the signal where . By subtracting from , we obtain an estimate alone. It is clear that, in the frequency range in which of the contribution of the EICN of the amplifiers is negligible, we obtain a direct estimate of the PSD of the noise signal we are interested in. Since the contributions of the EIVN of the amplifiers cancel out, the sensitivity of the measurement method is limited by the contributions of the EICN, whose effect depends on the equivalent impedance of the DUT. This fact implies that in order to measure very low voltage noise, the EICN of the amplifier represents the limiting factor rather than the EIVN. Therefore, using the measurement configuration we propose, one may obtain better noise performances from MOSFET input stage amplifiers (high EIVN and low EICN) rather than from those employing BJTs (low EIVN and high EICN), even in the case of low DUT impedances. Another advantage is that this method allows the PSD evaluation of the input signal without requiring the exact evaluation of the single noise sources of the input amplifiers and of the DUT impedance, which can change because of bias or temperature variations or because of drift with time. Moreover, while the PSD of the noise introduced by the preamplifiers is the sum of the PSD of the noise due to each single channel, the PSD of the input signal is proportional to the square of the sum of the signals relative to the two channels. Therefore, the dual-channel configuration is characterized by a doubled signal-to-noise ratio with respect to the case of a single-channel amplifier. It is worth noticing that the method we propose is, in principle, equivalent to the one discussed with reference to Fig. 1. In a sense, we are just performing the evaluation of the cross spectrum by way of a different approach. However, the operations needed for separating the correlated noise contribution from the uncorrelated ones at the output of the amplifiers are performed analogically in the time domain rather than numerically in the frequency domain. Besides resulting in many cases in a simpler implementation, our method does not depend on the properties of the DFT-based spectrum analysis (i.e., using a DFT-based spectrum analyzer for spectrum estimation is not mandatory in our case) and can be successfully performed even by using a single-channel spectrum analyzer. In fact, one can evaluate the PSD of the channel alone, then that of the channel alone, and finally perform the subtraction of the two PSDs thus obtained. By replacing the voltage noise amplifiers of Fig. 2 with the transimpedance amplifiers shown in Fig. 3, it is possible to measure the current noise of a DUT using the same principle described for the case of voltage noise measurements. It can be easily verified that in such a case all the previous considerations are still valid, provided that the roles of the addition block and of the subtraction block and all the statements relative to EIVN and EICN are exchanged. The results obtained by using the new method are reported in the following section, both in the case of the voltage noise measurement configuration and in the case of the current noise measurement configuration. III. EXPERIMENTAL RESULTS The actual circuit used for demonstrating the new measurement method in the case of voltage noise measurements is reported in Fig. 4. The voltage noise amplifiers are based on the TLC2201 by Texas Instruments, which is characterized by Hz) and by an a relatively high EIVN (10 nV/ Hz, excellent level of EICN (0.6 fA/ Hz). The gain of each single 658 Fig. 4. IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 51, NO. 4, AUGUST 2002 Implementation of the schematics of Fig. 2. Fig. 5. Input referred PSDs of the signals at the output of the addition and subtraction blocks of the circuit of Fig. 4 with a 10 resistor at the input, difference between the two spectra and theoretical PSD of the thermal noise of the input resistor. stage is set to 100 by means of 0.1% precision resistors. The dB). resulting bandwidth is from dc up to about 19 kHz ( The EICN of each stage coincides with that of the TLC2201, whereas its EIVN depends on the values of the resistors used for the feedback. Because of the values of the resistors we used, however, the EIVN of each amplifying stage reduces to that of the TLC2201 alone. The U3 unity gain instrumentation amplifier is used both for inverting the output of one of the voltage amplifiers (U2) and for allowing offset compensation in order not to saturate the following stages. The other unity gain instrumentation amplifier could be avoided, but it may be useful, as before, for offset compensation. If no offset compensation is required, the two terminals marked “Offset null” in the figure can be simply grounded. The two INA141 at the end of the amplifying chain realize the addition and subtraction Fig. 6. Input referred PSDs of the signals at the output of the addition and subtraction blocks for the case of the current noise measurements with a 1 G resistor at the input, difference between the two spectra and theoretical PSD of the thermal noise of the input resistor. blocks. The gain of these stages can be set to either 10 or 100, and their accuracy is guaranteed by the manufacturer to be and 750 ppm of the nominal within 500 value. We measured the PSD of the voltage fluctuations at the outputs of the addiction and subtraction blocks by means of a dual-channel FFT spectrum analyzer based on a National of 10 was Instruments PCI-4451 board. A resistance used as DUT. It acted as a source of voltage noise with a power , where is the Boltzmann constant spectral density of and is the absolute temperature. At room temperature, the value of the PSD of the voltage fluctuation to be measured is about 407 pV/ Hz. The contribution of the two EICN of the input amplifiers is negligible (less than 100 fV/ Hz). As can be verified in Fig. 5, the estimation of the PSD of the , as obtained by using voltage fluctuations at the ends of CIOFI et al.: NEW METHOD FOR HIGH-SENSITIVITY NOISE MEASUREMENTS the proposed method, is rather good, with an average value of 409 pV/ Hz and a standard deviation of 270 pV/ Hz after averages with a frequency resolution of 25 Hz, within the investigated bandwidth. It must be noted that the noise produced by the DUT is, in the low-frequency range of the spectrum, almost 30 dB below the EIVN of the input amplifying stages. The maximum sensitivity that can be obtained clearly depends on the mismatch between the addition and subtraction amplifying chains (including the input amplifiers and data conversion stages of the spectrum analyzers) and on the overall time length employed for spectrum estimation. An estimate of the sensitivity that can be obtained for a given setting of the measurement parameters can be obtained by performing a measurement with the input short-circuited. In this case, indeed, the difference between the spectra at the output of the addition and subtraction channels represents the error that will be added to the estimate of the noise produced by the DUT. We have also tested a possible implementation for current noise measurements. In order to obtain the new circuit, we had only to replace the voltage noise amplifier stages in Fig. 4 with a transimpedance amplifier stage realized following the scheme of Fig. 3 with two TLC2201 and 10 M feedback resistors. Since, in this case, the outputs of the two transimpedance amplifiers are inverted, the roles of the addition and subtraction blocks in Fig. 2 are reversed. The equivalent EIVN of each transimpedance stage coincides with the EIVN of the operational amplifier, while the EICN is mostly due to the thermal noise of the feedback resistor, i.e., about 41 fA/ Hz in the flat bandwidth of the transimpedance amplifier. The data reported in Fig. 6 were obtained by using a 1 G resistor as a DUT. Therefore, the input current noise was 20 dB below the EICN of each single transimpedance amplifier. The average value we estimated for the input current noise was 4.1 fA/ Hz, substantially coincident with the theoretically expected value, with a averages with a frestandard deviation of 1.2 fA/ Hz after quency resolution of 50 Hz within the investigated frequency range. IV. CONCLUSION A new method for high-sensitivity noise measurements was presented in this paper. The method is based on the separate estimation of the correlated and uncorrelated noise at the output of two nominally identical voltage (for voltage noise measurements) or transimpedance (for current noise measurements) amplifiers. The method is equivalent, in principle, to other methods based on the cross-correlation technique. However, while in the case of the cross-correlation technique, it is mandatory to employ a two-channel FFT-based spectrum analyzer, the approach we propose is independent of the method used for spectrum estimation and could, in principle, be extended to very high frequency ranges. By using standard commercially available operational amplifiers and without any calibration, we have been able to measure voltage noise levels as low as 400 pV/ Hz superimposed to a noise level 30 dB higher and current noise levels as low as 4.1 fA/ Hz superimposed to a noise level 20 dB higher. Future work will be devoted to the development of calibration procedures for compensating the 659 mismatch in the amplification channels in order to ascertain the ultimate possibilities offered by the method we propose. REFERENCES [1] C. Ciofi and B. Neri, “Low-frequency noise measurements as a characterization tool for degradation phenomena in solid-state devices,” J. Phys. D, vol. 33, 2000. R199. [2] M. Macucci and B. Pellegrini, “Very sensitive measurement method of electron device current noise,” IEEE Trans. Instrum. Meas., vol. 40, pp. 7–12, Feb. 1991. [3] A. van der Ziel, Noise: Sources, Characterization, Measurement. Englewood Cliffs, NJ: Prentice-Hall, 1970, p. 54. [4] M. Sampietro, L. Fasoli, and G. Ferrari, “Spectrum analyzer with noise reduction by cross correlation technique on two channels,” Rev. Sci. Instrum., vol. 70, no. 5, pp. 2520–2525, 1999. [5] M. Sampietro, G. Accomando, L. G. Fasoli, G. Ferrari, and E. C. Gatti, “High sensitivity noise measurement with a correlation spectrum analyzer,” IEEE Trans. Instrum. Meas., vol. 49, pp. 820–822, Aug. 2000. [6] E. Rubiola and V. Giordano, “A correlation-based noise measrurement scheme showing sensitivity below the thermal floor,” in Proc. Int. Conf. Noise Phys. Syst. and 1=f Fluctuations, Hong Kong, 1999, pp. 483–486. Carmine Ciofi was born in Cosenza, Italy, in 1965. He received the degree in electronic engineering in 1989 from the University of Pisa, Pisa, Italy. In 1993, he received the Ph.D. degree in electronic engineering from the Scuola Superiore S. Anna, Pisa. From 1993 to 1998, he was with the Dipartimento di Ingegneria dell’Informazione, University of Pisa. In 1998, he left the University of Pisa and joined the University of Messina, Messina, Italy, where he is currently an Associate Professor of electronics. He has been and is currently involved in research projects on the characterization of electron devices by means of transmission electron microscopy (TEM), on the evaluation of the reliability of metallic interconnections, and on the design and realization of low noise electronic instrumentation. He has published more than 50 papers on international journals and congress proceedings. Felice Crupi was born in Lamezia Terme, Italy, on December 21, 1972. He received the degree in electronic engineering from the University of Messina, Messina, Italy, in 1997 and the Ph.D. degree from the University of Firenze, Firenze, Italy, in 2001. In 2000, he joined the University of Messina as Research Contractor. In 1997, in 1998 and in 2001 he was visitor at the Interuniversity Micro-Electronics Center (IMEC), Leuven, Belgium, and in 2000, he was visitor at the IBM Thomas J. Watson Research Center, Yorktown Heights, NY. His main research interests include the modeling of the degradation and of the breakdown processes in thin oxide layers, the study of the noise in MOSFET devices, and the design of ultra low-noise instrumentation. Calogero Pace was born in Palermo, Italy, in 1965. He received the degree in electronic engineering in 1990 from the University of Palermo. In 1994, he received the Ph.D. degree in electronic engineering from the University of Palermo. In 1996, he joined the University of Messina, Messina, Italy, where he is currently Assistant Professor of electronics. He is currently involved in research projects on the evaluation of the reliability of metallic interconnections, on the design and realization of low-noise electronic instrumentation, and on the design and characterization of electronic gas sensors. He has been involved in research projects supported by Ministero dell’Università e della Ricerca Scientifica e Tecnologica and by the Consiglio Nazionale delle Ricerche (P.F.T.E.O.).