·Table of Contents
·Materials Characterization and testing
A Methodological Approach to Fault Location of IC Modules
Qian Zhihong, Ma Yukuan, Gao Minghui, Dai Yisong
Department of Electronics Engineering, Jilin University of Technology, Changchun, Jilin 130025, P.R.China
The Second Aeronautical College of the Air force, Changchun, Jilin 130025, P.R.China
To locate the faults of integrated circuit (IC) modules, while still keeping the modules non-destruction, a methodology has been proposed. The analysis and fault location requires a noise parameter estimate of the integrated operational amplifier and even the multistage integrated operational amplifiers of an IC module. It is shown that the equivalent input noise voltage (En) and equivalent input noise current (In) are the reasonable expression for the noise characteristics of a multistage amplifier. The novel approaches are based on the integer of all the noise sources and the noise relativity, so that the accuracy and efficiency are greatly improved and the algorithms are guaranteed to be more canonical. A practical implementation for the fault location of an IC module is described in this paper. The faults previously set has been found by comparison of noise spectrum models and the test data measured at various test points. Experimental results show the effectiveness of the proposed methodology. Thus, one may apply spectral parametric modeling to this power spectral density for the estimation of noise figure and for the fault location of electronic systems.
An electronic system contains a large number of integrated circuit (IC) modules, which are used to amplify or deal with a big variety of signals in the system. Study on representative IC modules has universal import for fault location.
Several yield equations for defect-tolerant ICs can be found in the literature -. Most previous approaches to component-level fault diagnosis have relied on a priori computation of effect form cause . Given pre-computed fault effects or symptoms for each plausible fault in a circuit, diagnosis consists of matching the observed symptoms of a faulty circuit with the computed effects of a particular fault. The techniques may suffer form the shortcoming that they can only diagnose faults for which symptoms have been computed or can be computed during diagnosis. This implies that the fault to be diagnosed must have been modeled. The work reported here was aimed at the diagnosis of non-classical faults of unknown type, definitely, the fault-free circuits must be modeled, and the fault circuits to be diagnosed may be not.
Clearly, no method can diagnose a fault and generate a result of the type 'the fault is a type x fault' unless the method has built-in knowledge of the given fault type . Therefore any general method that aims to diagnose non-classical novel faults must resort to production of a rather more vague result that is nevertheless useful for repair or detailed elucidation of the underlying cause of the fault. The work presented in this paper is a methodological approach towards this aim, and it has shown how faulty regions of a circuit can be localized even when multiple or non-classical faults are present, using the results of tests on the faulty circuit.
In many applications, particularly in military and space applications, reliability assessment techniques are used to identify possible hazardous failure modes of equipment. When complexity increases, the chance of failure goes up exponentially. Therefore, complex electronic equipment has low reliability, and it is difficult and expensive to locate the faults of the equipment for the reason, partly, that the cause of a fault may be obscure because a malfunction corrected by replacing one unit (such as a printed-circuit board) with another does not prove that the replaced board was defective.
The methodology to locate the faults of integrated circuit modules has been proposed here. The analysis and fault location requires a noise parameter estimate of the integrated operational amplifier and even the multistage integrated operational amplifiers of an IC module. It is shown that the equivalent input noise voltage (En) and equivalent input noise current (In) are the reasonable expression for the noise characteristics of a multistage amplifier. The En-In noise model for IC modules is derived by the noise correlation matrix, which represents the feedback network of an electronic system.
By conventional method for noise analysis of an electronic network, the output noise power is calculated with the sum of individual noises, and then the parameter is translated into the equivalent input noise voltage and input noise current. The novel approaches are based on the integer of all the noise sources and the noise relativity, so that the accuracy and efficiency are greatly improved and the algorithms are guaranteed to be more canonical.
2. IC MODULES, NOISE AND MATRIX ALGORITHMS
The output noise spectrum of an IC module depends on its input noise and many circuit parameters, such as nominal circuit values, loads, parasitic capacitance, frequency etc. This output noise spectrum could be expressed as a synthetic of the branch noise of the circuit.
Every fault condition can be considered as a change in some branch noise and it will cause a more or less significant change in the output noise spectrum of an IC module. It is available to measure the output noise spectrum and compare the difference between the noise spectrum and that calculated using derivatives.
Noise is generated in any material that is above absolute zero. All electrical devices and their controls also generate noise. Even if there were no external noise, there would still be noise in the output voltage caused by the IC modules themselves. In fact, noise consideration is more complex than that we thought of. We are, in general, still far from having a complete process for the design of an IC module or an electronic system. Although manufacturers supply adequate parameters on manuals for users, they are still faced with some challenging problem like noise estimation of an IC op amp (operational amplifier).
Assuming that the noise characteristics can be somehow determined empirically, a model still need choosing for describing the noise so that it leads to a more accurate fault diagnosis of the IC module. This model is referred to as En-In noise model, where En and In denote equivalent input noise voltage and equivalent input noise current, respectively -. In this way, all the noise sources in the IC module are assumed to be lumped into a single equivalent input noise voltage and equivalent input noise current. Figure 1 shows the circuit placement of the noise sources in front of a noiseless IC module.
Fig 1: Placement of the equivalent input noise voltage and equivalent input noise current generators in front of a noiseless IC module.
An En-In model consists of an ideal noiseless amplifier, a noise voltage generator in series with either of the inputs, and a noise current generator in parallel with each input.
What is required is a structured approach to the problem, through which the effects of noise sources can be superposed. If we look at a single element of the ensemble for the time interval 0 £ t £ T we can make up the time average á X(t) ñ of the function X(t)of a random variable x(t) by the definition
If this time average approaches the ensemble average in the limit when T goes to infinity, the noise processes under investigation are said to be ergodic so that the voltage spectral density of a two-port network is given as
Here, denotes the n-order random harmonic components of equivalent input noise voltage E(t) , and is the complex conjugate of. Similarly, the current spectral density of an IC network is
The equation is valid if the sufficient condition, is absolutely integrable, is satisfied .
Applying (2) and (3) yields
and cross-spectral density
Here, Y2m is called short-circuited forward transfer admittance, something like a transfer function; and H2m is called open-circuited forward transfer admittance. The procedure was described in detail in Literature  and .
Equation (4) and (5) can be expressed by the matrix relationship existing between output-input ports of IC op amps
It follows from (7) that the En-In noise model matrix of an IC op amp is given by
Here,A-1 is the inverse matrix of A, and represents the transposed matrix, or Hermitian conjugate matrix , of matrix Ke-i.
These are noise spectral matrices, which are useful if an IC module has to be evaluated repeatedly for several values of the components or the frequency. By a software package, it takes less time to evaluate the same analytical expression with one or a few parameters or the frequency varying over some range than to perform a full analysis each time. It gets easier to perform the analysis with a computer program.
3. PRACTICAL IMPLEMENTATION
The method was developed for the analysis of rocket designs to identify single point failure modes, that is single component failures, or multistage integrated operational amplifier modes, which fail the system. The analytical expressions do not only allow us to avoid the determination of noise parameter difference between good or bad IC models with time-consuming, but also enable us to design better systems. The matrix format aids the multi-level analysis of very large systems and generates concise results.
The work presented above is a first step towards the aim, fault location. The second is in development of the practical implementation for ICs to establish the common reasons for failure so that corrective action can be taken.
The output noise spectrum of the system to be detected is derived from those of the noise sources of all the components of the system and their transfer function to the output so that measuring terminals should be determined first. The noise that is contributed by resistors in the circuit should be taken into account. The spectral density of resistor noise is calculated by
where k is Boltzmann's constant, k=1.374´10-23(J/K)
and T represents temperature (K).R is the resistor value.
The IC module as shown in Figure 2 provides the example of fault location, the circuit of which is shown in simplified form.
Fig 2: An IC module as an active filter.
The circuit illustrated in Fig.2 can be regarded as a typical two-port-network so that its normal output noise spectrum and the tolerance can be obtained by noise matrix approach, i.e. calculate the noise spectrum at the outputs of the IC module to be diagnosed. Details of doing this are given in Literature . In fact, the amplifier noise current spectral density is customarily neglected compared with the noise voltage spectral density . This is justifiable for almost all practical impedance levels associated with both bipolar and MOS realizations.
Fig 3: Output noise spectrum of the active filter
During the study, we tested two types of failure sample circuits, circuit 1 and circuit 2. Measured with a FFT analyzer, the two circuits turn out to be faulty circuits. Fig. 3 illustrates the normal noise spectrum, its tolerance (dotted line) and failure conditions of circuit 1 and circuit 2. Comparing with the normal noise spectrum, we could make the decision that the two sample circuit are failures.
The amount of analysis required for a particular system or sub-system should be determined by the accuracy required, the consequences of failure and the nature of the system. The output noise spectrum gives only limited information, and further measurements and calculations for some stages are necessary to locate the failures in the system. Place a short circuit around the input terminals and the input capacitor of the sample circuit 1, and have the circuit from A2 to A3 opened. Measure the output noise spectrum of A1, the first stage of the IC. We found that the measured spectrum is quite different from the spectrum calculated by En-In model at the normal condition of the system. Hence, the first stage of the fault circuit 1 makes prove to be the failure, which was true.
Diagnosing the fault circuit 2 performs the same procedures as that mentioned above. Placing a short circuit across the output terminals of the first stage of the system, measuring the output noise spectrum of A2, we found that the measured spectrum is quite different from the spectrum calculated by En-In model at the normal condition of the system. Thus, the second stage of the fault circuit 2 has a fault.
It is advisable to select some internal and external noise measurement points to be used for measurements if some of terminals of the IC are beyond reach. In this case, the output noise spectrum is the sum of the overall noise contribution of some devices in the IC module. The arguments as to whether the correlation coefficient needs to be taken into account for circuit design and simulation still exists. Some researchers like A. V. D. Ziel  favor consideration of the correlation. However others, such as Pospieszalski , Froelich , and Podell , do not. In fact, the decomposition of a noise source into two uncorrelated sources greatly simplifies the procedure of predictions of device noise performance, and the quality of the predictions of obtained thereafter.
This paper presents the methodology for fault diagnosis and location with noise models of IC modules. The matrix expression of a noise model provides a shortcut to predict the property of an IC module. It has been shown that the use of noise measurement can simplify the fault diagnosis process significantly. Next, a software package should be made up, which will prove itself to be a valuable design aid for, experienced, and novice designers, complementary to numerical simulator. Future work should consist of investigating the potentiality of widely application.
It is becoming more difficult to locate the different faults of IC modules because of the complexity of the electronic systems. At the same time, it is becoming increasingly possible to design systems with very few unrevealed dangerous failure modes. So, while the reliability of such systems may be as good as or better than earlier designs, it is may be necessary to analyze parts of these systems on a functional basis. Similarly, it may not be possible to assess all parts of a system quantitatively whereas a more qualitative approach may be required. The noise measurement methodology is equally valuable in the achievement of non-destructive testing.
- I. Koren and C. Stapper, "Yield Models for defect-tolerant VLSI circuits: A review," Defect and Fault-Tolerance in VLSI Systems. New York: Plenum Press, 1989, pp.1-21.
- C. Stapper, F.Armstrong, and K.Saji, "Integrated circuits yield statistics, " Proc. IEEE, vol.71, pp.453-470, Apr. 1983.
- I. Koren and D. Pradhan, "Modeling the effect of redundancy on yield and performance of VLSI systems," IEEE Trans. Computers, vol. 37, no. 3, pp.344-355, Mar 1987.
- C. Thibeault, Y.Savaria, and J.L.Houle, "Equivalence proofs of some yield modeling methods for defect-tolerant integrated circuits," IEEE Trans. Computers, vol. 44, no. 5, pp.724-728, May 1995.
- S.J. Sangwine, "Diagnosis of multiple faults in combinational digital circuits by modeling of transition propagation along critical paths," IEE Proc.-G, Vol. 139, No. 5, Oct. 1992.
- Y.S. Dai, "Design of low-noise test instruments," Chinese Journal of Scientific Instrument, vol.1, no.1, 1980, pp.123-133.
- Y.S. Dai, Noise in Electronic Systems and Design of low-noise Systems (in Chinese). Changchun, China: Jilin People's Publising House, 1984.
- T. Luo, Y.S. Dai, "Noise circuit theory and noise calculation of integrated circuits," Acta Electronica Sinica (in Chinese), vol.18, no.6, 1990, pp.79-83.
- A.V.D.Ziel, Noise: Sources, Characterization, and Measurements. Prentice Hall, Englewood Cliffs, N.J., 1970.
- R.A. Horn and C.R. Johnson, Matrix Analysis, Cambridge: Cambridge University Press, 1990.
- Qian Zhihong, Zhao Xiaoming and Dai Yisong. "En-In models study of automotive IC modules." Proc. IEEE IVEC'99, Changchun, China, Sept. 1999, pp.188-192
- P. Bowron and K.A. Mezher, 'Noise and sensitivity optimisation in the design of second-order single-amplifier filters," Int. J. Circuit Theory Appl., vol. 19, 1991, pp.389-402.
- D.P.E. Dale, "The LinCMOS design manual," Texas Instruments, 1985, pp. 315-320.
- M.W.Pospieszalski, "Modeling of noise parameters of MESFETs and MODFETs and their frequency and temperature dependence," IEEE Trans, vol. MTT-37, no.9, 1989, pp. 1340-1350.
- R.K.Froelich, "An improved model for noise characterization of microwave GaAs FETs," IEEE Trans., vol.MTT-38, no.6, 1990, pp. 703-706.
- A. Podell, "A functional GaAs FET noise model," IEEE Trans., vol. ED-28, no.5, 1981, pp. 511-517.
- Y.S. Dai, Electronics on Noises (in Chinese). Shandong, China: Shandong Science & Technology Publishing House, 1997.