A precision noise measurement and analysis method used to estimate reliability of semiconductor devices

In this paper, some problems with previous ultra-low noise measurement methods have been discussed, then a double-channel preamplifier cross-spectrum measurement method has been adopted, different from the previous cross-correlation method [A. van der Ziel, Noise: Sources, Characterization, Measurement, p. 54. Prentice-Hall, Englewood Cliffs, NJ (1970), L. Stor, Experimental techniques in noise measurement with special emphasis on precision measurement, Proc. 10th Int. Conf. on Noise in Physical Systems, pp. 551–560. Budapest, Hungary (1989)] in that an average periodogram using a windowing procedure has been performed. The theoretical analysis shows that the expected value of the cross-spectrum is incoherent with background noise and zero-drift from the preamplifier and power supply system, the average periodogram can decrease the variance of the periodogram and the additional bias of the cross-spectrum periodogram.Experimental results demonstrate that if the equivalent input noise of measuremental set-up is two orders of magnitude lower than the noise of each preamplifier, then an ultra-low noise spectrum can be measured accurately, the low limit is about 0.1 nV/√Hz at 1 kHz, which is 20 dB lower than the noise of each preamplifier. The thermal noise of a small resistance and the shot noise of a diode under forward conditions have been measured, the experimental results are in good agreement with the theoretical value, this means that this method is feasible and accurate for an ultra-low noise spectrum measurement.Finally, the noise spectrum analysis procedure based on the curve fitting method has been presented, which ensures that we obtain an accuracy value of three noise components in the semiconductor, i.e. noise, white noise and g-r noise. This noise spectrum analysis method is a useful tool for investigation of noise mechanism, the diagnosis of defects in semiconductors and reliability estimation.     

Improving reliability of beveled power semiconductor devices passivated by SIPOS

Semi-insulating polycrystalline silicon films were prepared by direct current glow discharge on bevelled power semiconductor devices. Devices passivated by polyimide were also fabricated for comparison studies. The reverse current–voltage characteristic was studied at room temperature and high junction temperature. The differences arose from the remnant unbalance charges in the passivants as well as the varying dielectric strengths. Passivation materials applied to the beveled devices played a key role in achieving the targeted voltage. Results obtained indicated that the performance of power semiconductor devices passivated by SIPOS were superior to those passivated by polyimide, and SIPOS based process was effective and expected to enhance the reliability and yield.     

半导体激光器的低频电噪声谱密度和器件的可靠性

本文给出了半导体激光器的低频电噪声谱密度和器件可靠性关系的实验结果。     

Theory of noise in metal oxide semiconductor devices

Noise in MOS diodes arises from different sources: fluctuations in occupation of surface states, shot noise, and leakage noise. Fluctuations in the occupation of surface states produce changes in the surface space-charge distribution which in turn produce currents. Shot noise is produced by fluctuations of the individual drift and diffusion flows toward the surface. Leakage noise is associated with the small flow of current through the oxide. In MOS triodes these three mechanisms give rise to gate noise and thus input noise in the amplifier, but the first one produces an important indirect effect. Fluctuations in the occupation of interface states result in modulation of the channel conductance. At low frequencies this modulation is the dominant effect, giving rise to a noise power spectrum which resembles1/fnoise. At high frequencies, where only thermal noise in the channel and input noise are of importance, MOS triodes are similar to junction field effect devices from the noise point of view.     

Fluctuations and noise of hot carriers in semiconductor materialsand devices

After recalling the definition of the noise temperature, the macroscopic expressions for noise sources are shown not to be specific to the hot carrier regime, though dependent on the electric field strength. Careful modeling allows one to obtain important information on transport parameters from noise measurements. The microscopic noise source expressions, via the transition rates, give a unified view of the noise sources. In particular, it is clarified that noise sources are intercorrelated, and that there is also space correlations over lengths of a few mean free paths. Recent developments are reviewed, concerning noise modeling using direct numerical methods for solution of the Boltzmann equation. Finally, impedance field methods for modeling noise of devices are briefly evoked     

Sequence test method for reliability evaluation of semiconductor devices

The failure-rate λ, of a device can be determined using Arrhenius model $\text{λ = A}\text{e}{}^{\mathrm{<mtext>?E</mtext><mtext>KT</mtext>}}$. The number of thermal cycles a device can withstand can be postulated using an exponential model, N = g e^?aT. Based on these models, a “sequence” process of thermal-fatigue and life tests is arrived at, which makes use of a mathematical equation giving the rate of decay of MTBF relative to the number of thermal-cycles, $\text{δ = P}\text{exp}\text{[p · ΔT + (}\text{q}\text{T}\text{)]}$. In other words, MTBF of the devices can be reduced by pre thermal cycling.     

Equivalent circuit analysis of noise in bulk semiconductor devices

Equivalent transmission-line analogs may be developed to advantage for the analysis of noise in bulk semiconductor devices. We discuss first a transmission-line analog for the law of conduction and diffusion of a single species of charge carriers that experience small disturbances from equilibrium. Through the use of Nyquist's theorem it is possible to obtain the power spectra of the noise sources of the equivalent transmission-line circuit at thermal equilibrium. Next, an equivalent circuit can be obtained if there is more than one charge carrier. This circuit is generalized to the case of drifted carriers and applied to the metal-semiconductor-metal (MSM) diode and to the small-signal analysis of the Gunn effect. We show how the equivalent circuit for the Gunn effect gives, by inspection, a lower bound on the noise measure achievable with a Gunn diode. We find that the Gunn diode has noise measures exceeding the lower bound. A traveling-wave Gunn device achieves the lower limit.     

A lumped model analysis of noise in semiconductor devices

A clear picture of the noise properties of semiconductor devices may be gained through a study of lumped models. When a semiconductor device is represented by a lumped model it is found that all the noise may be accounted for by associating a noise-current generator with each of the conductances appearing in the lumped model. The expressions describing these noise-current generators are easily found and the determination of the device noise properties involves only lumped model analysis. Such analysis applied to a diode model and a transistor model yields results which are in agreement with the results of other authors and with experimental measurements cited by them. The mechanisms considered in this paper do not, however, account for the so-called 1/f noise.     

Monte Carlo simulation of noise in GaAs semiconductor devices

The Monte Carlo method has been used to calculate the noise in GaAs resistors where finite-size effects are important, being from 0.5 to 1.5 μm in length. The results agree qualitatively with experimental noise measurements performed on similar devices. Monte Carlo noise calculations have also been conducted for Schottky barrier diodes. A method for conducting full-device Monte Carlo simulations of these diodes has been employed. Schottky barrier diodes with both 0.12 and 1.0 μm epitaxial layers were modeled, and the calculated noise is in agreement with experiment throughout a wide range of bias voltages, both when shot noise and when excess noise predominate. Two different methods of calculating the noise current have been compared     

Optimal low-frequency noise criteria used as a reliability test for BJTs and experimental results

Two basic problems in the application of noise criteria to reliability testing are considered. First, the need to look at all noise mechanisms rather than noise at a specific frequency is emphasized, and noise criteria using both 1/f noise and g−r noise are established. Second, the physical mechanisms which explain the low noise level of some failed devices are discussed, and the optimal noise threshold levels are found for minimum error during reliability testing.Reliability testing results for 1000 BJTs (3DG6) show that the failure rate of devices which initially exhibit high noise is about 2–3 times higher than for devices which initially have low noise. This means that the life expectancy of BJTs with a low noise level (using optimal noise criteria) is increased by a factor of 1.5–2.5; thus, noise measurements could be a useful reliability screen classification method for BJTs.     

The influence of X-ray radiation on the low-frequency noise of integrated circuits

The sensitivity of various parameters of integrated circuits developed by various technologies to X-ray radiation and the possibility to predict the level of low-frequency noise are considered.     

Generalizations of the Klaassen-Prins equation for calculating the noise of semiconductor devices

The Klaassen-Prins equation is the standard equation for calculating the drain thermal noise of long-channel MOSFETs. We show that the Klaassen-Prins equation is not always valid, even for MOSFETs. We present generalizations to the Klaassen-Prins equation that include velocity saturation effects of short-channel MOSFETs and that comprise also induced gate noise.     

Characterisation of reliability of compound semiconductor devices using electrical pulses

Transmission line pulses (TLP) with 0–60V amplitude and 100ns pulse width have been applied for accelerated lifetime tests of GaAs devices. 1/f noise, RF noise and I/V measurements are applicable for the characterisation of the reliability of these devices. Different failure mechanisms can be identified by applying square pulses of varying amplitude and different polarity on a variety of samples. Correlation between anomalies in 1/f noise, RF noise and I/V characteristics has been determined. Using this novel method, the determination of failure threshold levels for current density, electric field and charge carrier temperature is possible and critical spots in device design can be ascertained.     

Dielectric breakdown and reliability of MOS microstructures: Traditional characterization and low-frequency noise measurements

Four different types of ultra-thin oxide MOS structures have been analyzed to investigate the dielectric breakdown, by using a combination of techniques: time-zero breakdown, time-dependent breakdown and low-frequency noise measurements. The experiments have shown that different devices presenting lifetime values depending on the fabrication technology are all affected, before breakdown, by sharp fluctuations of the current tunnelling through the oxide in both its components: the main and the substrate current. These new results confirm those previously obtained with a set of homogeneous devices, and show that the presence of current fluctuations is a general experimental evidence in the occurrence of dielectric breakdown. The main physical models proposed in the past to explain the origin of substrate current and breakdown are then briefly reviewed. Thereafter, a comparative analysis showing how these mechanisms could justify the experimental data coming from the noise measurements is presented. The results of this analysis seem to rule out the band-to-band impact ionization mechanism as responsible for the substrate current and breakdown.     

Relation between low-frequency noise and long-term reliability ofsingle AlGaAs/GaAs power HBTs

Self-aligned AlGaAs/GaAs single heterojunction bipolar transistors (HBTs) were fabricated using an advanced processing technology for microwave and millimeter-wave power applications. These devices were processed simultaneously, on different epilayers with similar layer structure design supplied from different vendors. They showed similar dc characteristics (current gain, β=30) and their microwave performance was also identical (f_T=60 GHz, f_max=100 GHz). The HBTs showed different noise and reliability characteristics depending on their epilayer origin. HBT's from the high-reliability wafer showed MTTF of 10⁹ h at junction temperature of 120°C. They also presented very small 1/f noise with corner frequencies in the range of a few hundred Hz. Devices were subjected to bias and temperature stress for testing their noise and reliability characteristics. Stressed and unstressed devices showed generation-recombination noise with activation energies between 120-210 meV. Stress was found to increase the generation-recombination noise intensity but not its activation energy. These HBTs did not show any surface-related noise indicating that processing did not significantly influence noise characteristics. It was found that the base noise spectral density at low frequency can be correlated to the device long term reliability     

Phase noise and a low-frequency noise reduction method in bipolartransistors

A low-frequency amplitude and phase-noise measuring system was set up and both noises were measured. It is noted they were correlated in a bipolar transistor. Both noises were from the same origin: diffusion coefficient fluctuation in the base. A demonstration was carried out on the amplitude noise reduction by using the correlation between both noises     

Low frequency noise measurements as a tool to analyze deep-level impurities in semiconductor devices

An analysis method to extract the pertinent trap parameters of impurities, i.e. energy level, trap density, spin degeneracy and capture cross section, from low-frequency noise measurements is discussed. This method, which is based on the model of traps interacting independently with the conduction band, is applied to generation recombination noise spectra measured on planar n⁺-n-n⁺ Si resistors as a function of temperature, and the trap parameters are obtained. Since the analysis method makes use of some assumptions, its validity is verified by using obtained trap parameters as input for an exact multilevel computer simulation program. This program generates zero-frequency plateau levels and characteristic times in good agreement with the measured data, indicating that the assumptions were warranted. In addition, an alternative explanation of the noise data in terms of a double-donor model is studied. However, no conclusive evidence was found in favor of either one of the two models since both models explained the experimental findings within the experimental errors.     

Skewness and kurtosis of 1/f noise in semiconductor devices

The degree of the Gaussian nature of the white noise present in microwave low-noise devices is experimentally investigated. The chosen experimental technique consists of simultaneously digitizing four versions of the noise which are amplified by four parallel independent amplifiers. The four independent signals are then used to compute the second, and, to a good approximation, the fourth moment of the noise. The ratio of the fourth moment to the square of the second moment is the kurtosis of the noise. Gaussian processes are characterized by a kurtosis equal to 3. A deviation from this value gives an indication about the degree of non-Gaussian nature of the noise. By using this technique, the effect of the additive noise introduced by the amplifiers is strongly reduced. In our experiments, the degree of the Gaussian nature of the white noise of some microwave devices is measured in the frequency range from 100 to 500 MHz. In all the investigated devices, the kurtosis is found to be very close to 3.