Documente Academic
Documente Profesional
Documente Cultură
Abstract— This paper deals with multiple soft fault diagnosis of the test equations, which involve the measured voltages and
nonlinear analog circuits and offers a method that allows locating the unknown parameters, are algebraic and usually nonlinear.
the faulty parameters and evaluating their values. The method Most of the works focused on the fault diagnosis of analog
works with a system of nonlinear algebraic test equations, which
are not given in explicit analytical form, and actually may possess circuits address only the case when just one parameter
multiple solutions. The solutions are specified by one or several is faulty. Fewer papers are devoted to the multiple fault
sets of the diagnostic parameter values that meet the test. diagnosis, where several parameters can be faulty. If the
To find the multiple solutions, an extended systematic search parameters are slightly drifted from their nominal values,
method has been developed. The diagnostic algorithm proposed the test equations can be linearized [10]. Otherwise, when
in this paper exploits this method and brings a new concept for
finding the actual solution. As a result, the diagnostic process the parameter deviations are large, the nonlinearities of the
is considerably improved. It can be applied to bipolar junction equations must be considered. These equations may actually
transistor and CMOS circuits manufactured in micrometer and have multiple solutions, which means that several sets of
nanometer technologies. In the last case, however, the transistors parameter values meet the diagnostic test. In such a case,
are characterized by intricate models (PSP 103 or BSIM 4). finding just one specific solution is rarely of interest. Hence,
In consequence, the CPU time increases due to very complex
sensitivity analyses of the circuits required by the diagnostic references [7] and [12]–[15] propose an approach based on
method. In such a case, the proposed approach is useful at the concept of finding multiple solutions of the test equation
the preproduction stage, where the CPU time is offline and not and next selecting from them the actual one. In [12]–[15],
crucial. For illustration, two numerical examples are given. various methods such as the parametric homotopy [12],
Index Terms— Analog circuits, bipolar and CMOS circuits, the simplical homotopy [13], the restart homotopy [15],
fault diagnosis, multiple soft faults, nonlinearity. or the block relaxation method [14] are used to compute
the multiple solutions. To find the actual one, another test is
I. I NTRODUCTION arranged and the procedure is repeated. Next, the common
solution of the ones obtained using both tests is selected.
F AULT diagnosis of analog circuits is an important
problem for design validation of electronic
devices [1], [2]. In general, the fault diagnosis consists
This common solution, being a set of the parameter values,
is considered as the actual one. Many numerical experiments,
of detecting faulty circuits, locating faulty elements, and performed to verify this approach, show its effectiveness.
evaluating their parameters. A fault is called soft when the However, in some cases, the methods applied to solve the test
parameter deviates from its tolerance range, but does not equation fail. Moreover, this approach is time consuming due
produce a short circuit or an open circuit. In some cases, to the necessity of finding multiple solutions of the equations
physical imperfections such as near-opens and near-shorts corresponding to both tests.
may occur [2], [3]. During the last decades, a number of To overcome these drawbacks, a new concept is proposed
methods focused on soft fault diagnosis of analog circuits have in this paper as follows. Two tests of the circuit are performed,
been developed [4]–[15]. If most of the circuit simulations but only one of them is used to find the solutions. The other
take place after any testing, the diagnostic method is classified allows checking whether the obtained solution is the actual
as the simulation-after-test (SAT) approach, otherwise the one. For this purpose, a very efficient method, which is an
simulation-before-test (SBT) approach. Unlike the catastrophic extension of the search idea [18], [19], capable of finding
fault diagnosis, where the faulty parameters tend to infinity multiple solutions is applied. When a solution, being a set
or zero, which usually employs SBT approach [16], [17], of the parameter values, which meet the first test, has been
the soft fault diagnosis commonly uses SAT methods. They computed, a verification procedure is used. The circuit with
are based on the measurements of the voltages at accessible the obtained parameter values driven by the voltage sources
nodes and some analyses of the circuit under test. In dc case, as in the second test is analyzed to find the output voltages.
If they are sufficiently close to the voltages provided by
Manuscript received November 17, 2014; revised February 12, 2015; the second test, the solution is considered as the actual one
accepted February 19, 2015. Date of publication April 27, 2015; date of
current version September 11, 2015. The Associate Editor coordinating the and the computation process is terminated. Otherwise, the
review process was Dr. John Sheppard. method for solving the test equation is continued to find
The authors are with the Department of Electrical, Electronic, Computer and other solution and the approach is repeated. The procedure
Control Engineering, Łódź University of Technology, Łódź 90-924, Poland
(e-mail: michal.tadeusiewicz@p.lodz.pl; stanislaw.halgas@p.lodz.pl). is carried out as long as the solution is obtained, which meets
Digital Object Identifier 10.1109/TIM.2015.2421712 the second test. The advantages of the proposed approach are
0018-9456 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
TADEUSIEWICZ AND HAŁGAS: NEW APPROACH TO MULTIPLE SOFT FAULT DIAGNOSIS OF ANALOG BJT AND CMOS CIRCUITS 2689
avoiding computation of multiple solutions of the second test value problem. For this purpose, the differential equations
equation and possibility of finding the actual solution at the d f i (x(t))
beginning of the computation process. Numerical experiments = − f i (x(t)), f i (x(0)) = 0, i = 1, . . . , n − 1
dt
show that the developed approach considerably improves and d f n (x(t))
speeds up finding the actual set of the parameter values. = ± f n (x(t)), f n (x(0)) = fn0 (2)
dt
In some cases, the CPU time is reduced up to 100 times.
are created. The solution of (2) is given by the equations
II. P RELIMINARIES fi (x(t)) = f i (x(0))e−t = 0, i = 1, . . . , n − 1
This paper deals with multiple soft fault diagnosis of bipolar f n (x(t)) = f n0 e±t . (3)
junction transistor (BJT) and CMOS circuits, considering the
problem of multiple solutions of the test equations. For the The initial condition x(0) is selected in such a way that
circuits comprising n parameters considered as possibly faulty, f i (x(0)) = 0 for i = 1, . . . , n − 1, i.e., x(0) lies on the space
the diagnostic test is arranged similarly as in [12]. The circuit curve l defined by the intersection of surfaces f i (x) = 0,
under test is driven by voltage sources applied to terminals i = 1, . . . , n − 1. The first (n − 1) equations (3) show that
accessible for excitation and the output voltages are read at for any x(0) ∈ l, the trajectory x(t) remains on the curve l.
r nodes accessible for measurement. For m sets of the input Because df/dt = (df/dx)(dx/dt), (2) in x-space have the
voltage values, mr values of the output voltages are measured, form
where mr ≥ n. We take n of the measured voltages values dx
= J−1 (x)[− f 1 (x) · · · − f n−1 (x) ± fn (x)]T (4)
and consider them as elements of the vector u = [u 1 · · · u n ]T , dt
where T means transposition. Each of the voltages is a certain with the initial condition x(0) specified by (2), where
function u i = fˆi (x), i = 1, . . . , n, of the circuit parameters J(x) = df/dx is the Jacobian matrix of f(x). The solution
x 1 , . . . , x n , where x = [x 1 · · · x n ]T . As a result, the equation x(t) of the initial value problem (4) can be obtained by
f̂(x) = u is written, where f̂(x) = [ fˆ1 (x) · · · fˆn (x)]T . This numerical integration. Any point x(tk ) on the curve l for which
equation will be presented in the form f n (x(tk )) = 0 is a solution of (1). In [19], it is proved that the
f(x) = 0 (1) derivative of f n (x) in the direction of l is given by
∂ f n (x) detJ(x)
where f(x) = f̂(x) − u, and called a test equation. Unfortu- f n (x) = = (5)
∂l n
i=1 ni (x)
nately, in real circuits, the nonlinear function f̂(x) is not given 2
in explicit analytical form. To apply the method proposed
in this paper, two tests, labeled A and B, are arranged as where ni (x)(i = 1, . . . , n) are the cofactors of the Jacobian
described above, test A in order to find the solutions and test B matrix, corresponding to the last row of this matrix. If the
to verify whether a solution is the actual one. To differentiate minus sign is assigned to f n (x) in (2), then according to (3),
them, the equation of test B is labeled g(x) = 0, where f n (x) is forced to be attracted to zero along l. Otherwise, it is
g(x) = ĝ(x) − u B . forced to be diverged away from zero. In [19], it is proved
Because in real circuits, the number of nodes accessible that in the course of the searching, the transition in the sign
for measurement and excitation is limited, the information assigned to fn (x) should be made at the points where the
contained in the test equation may not be sufficient to find the Jacobian determinant changes sign and at the solution points.
actual set of the parameters. From mathematical point of view,
it means that the test equation (1) possesses several solutions. III. E XTENSION OF THE S YSTEMATIC S EARCH M ETHOD
To find them, a method that allows obtaining multiple solutions To solve numerically, the initial value problem (4), which
should be applied. An efficient approach for finding multiple can be rewritten in the form dx/dt = J−1 (x)w(x), where
solutions is converting the algebraic problem into the solution w(x) = [− f 1 (x) · · · − f n−1 (x) ± f n (x)]T , we apply the
of differential equations, proposed in [18]. The Branin idea explicit Runge–Kutta method of fourth order [21], [22], which
was employed by Chao et al. [19] to develop a systematic represents an appropriate compromise between the require-
search method, which allows finding multiple dc operating ments of a low truncation error and a low computational
points of nonlinear resistive circuits. The method was then cost. Implicit integration methods are ineffective in this case
extended in [20] to the computation of the input–output because they demand second-order sensitivity analyses of the
characteristics, the analysis of nonlinear resistive circuits circuit.
driven by time-varying sources, the large change sensitivity Because the test equation (1) is not given in explicit
analysis of linear circuits, and the computation of multivariable analytical form, the values of fˆi (x) (i = 1, . . . , n) and their
Nyquist plots. derivatives, with respect to x j ( j = 1, . . . , n), are determined
This paper extends the concept of the systematic search [19] numerically for given values of x 1 , . . . , x n by performing
for multiple solutions of the test equation and employs it in the dc and sensitivity analyses of the circuit. In this way, we find
developed diagnostic algorithm. In the succeeding part of this the required values of w(x) and the Jacobian matrix J(x).
section, the main idea of the systematic search method [19] is However, to determine all the solutions, using the
described. systematic search approach, it is essential that there exists a
To find multiple solutions of the algebraic test simple space curve l. Otherwise, e.g., if l is a multibranched
equation (1), the problem is reformulated in terms of the initial curve, some of the solutions may be lost. To overcome
2690 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 64, NO. 10, OCTOBER 2015
TABLE I
A RRANGED T ESTS A AND B—E XPERIMENTAL A PPROACH
TABLE II
R ESULTS G IVEN BY THE P ROPOSED M ETHOD
TABLE IV
R ESULTS G IVEN BY THE P ROPOSED M ETHOD
TABLE V
A RRANGED T ESTS AND THE O BTAINED R ESULTS
TABLE VI
A RRANGED T ESTS A AND B
TABLE VII
R ESULTS OF E XAMPLE 2 FOR THE M EASUREMENT A CCURACY 0.1 μV
TABLE VIII
R ESULTS OF E XAMPLE 2 FOR THE M EASUREMENT A CCURACY 1 μV
TABLE IX
A RRANGED T ESTS A AND B
R EFERENCES [18] F. H. Branin, “Widely convergent method for finding multiple solutions
of simultaneous nonlinear equations,” IBM J. Res. Develop., vol. 16,
[1] P. Kabisatpathy, A. Barua, and S. Sinha, Fault Diagnosis of Analog no. 5, pp. 504–522, 1972.
Integrated Circuits. Dordrecht, The Netherlands: Springer-Verlag, 2005. [19] K.-S. Chao, D.-K. Liu, and C.-T. Pan, “A systematic search method for
[2] D. Gizopoulos, Ed., Advances in Electronic Testing: Challenges and obtaining multiple solutions of simultaneous nonlinear equations,” IEEE
Methodologies. Dordrecht, The Netherlands: Springer-Verlag, 2006. Trans. Circuits Syst., vol. 22, no. 9, pp. 748–753, Sep. 1975.
[3] B. Kim, M. Swaminathan, A. Chatterjee, and D. Schimmel, “A novel [20] K.-S. Chao and R. Saeks, “Continuation methods in circuit analysis,”
test technique for MCM substrates,” IEEE Trans. Compon., Packag., Proc. IEEE, vol. 65, no. 8, pp. 1187–1194, Aug. 1977.
Manuf. Technol. B, Adv. Packag., vol. 20, no. 1, pp. 2–12, Feb. 1997. [21] L. O. Chua and P.-M. Lin, Computer-Aided Analysis of Electronic
[4] M. Aminian and F. Aminian, “A modular fault-diagnostic system for Circuits: Algorithms and Computational Techniques. Englewood Cliffs,
analog electronic circuits using neural networks with wavelet transform NJ, USA: Prentice-Hall, 1975.
as a preprocessor,” IEEE Trans. Instrum. Meas., vol. 56, no. 5, [22] E. Kreyszig, Advanced Engineering Mathematics. New York, NY, USA:
pp. 1546–1554, Oct. 2007. Wiley, 1999.
[5] S. Bhunia, A. Raychowdhury, and K. Roy, “Defect oriented testing [23] B. Kaminska et al., “Analog and mixed-signal benchmark circuits-first
of analog circuits using wavelet analysis of dynamic supply current,” release,” in Proc. Int. Test Conf., Nov. 1997, pp. 183–190.
J. Electron. Test., vol. 21, no. 2, pp. 147–159, 2005. [24] T. Quarles, A. R. Newton, D. O. Pederson, and
[6] M. Catelani and A. Fort, “Soft fault detection and isolation in analog A. Sangiovanni-Vincentelli, SPICE 3 Version 3F5 User’s Manual.
circuits: Some results and a comparison between a fuzzy approach and Berkeley, CA, USA: Univ. California Press, 1994.
radial basis function networks,” IEEE Trans. Instrum. Meas., vol. 51, [25] ICAP4. Working With Model Libraries, Intusoft, San Pedro, CA, USA,
no. 2, pp. 196–202, Apr. 2002. 2000.
[7] G. Fedi, R. Giomi, A. Luchetta, S. Manetti, and M. C. Piccirilli, [26] G. Gildenblat, Ed., Compact Modeling: Principles, Techniques and
“On the application of symbolic techniques to the multiple fault location Applications. Dordrecht, The Netherlands: Springer-Verlag, 2010.
in low testability analog circuits,” IEEE Trans. Circuits Syst. II, Analog [27] PSP Model Documentation. (2014). [Online]. Available: http://
Digit. Signal Process, vol. 45, no. 10, pp. 1383–1388, Oct. 1998. pspmodel.asu.edu/psp_documentation.htm, accessed May 8, 2014.
[8] A. D. Spyronasios, M. G. Dimopoulos, and A. A. Hatzopoulos, “Wavelet
analysis for the detection of parametric and catastrophic faults in
mixed-signal circuits,” IEEE Trans. Instrum. Meas., vol. 60, no. 6,
pp. 2025–2038, Jun. 2011.
[9] M. Tadeusiewicz and M. Korzybski, “A method for fault diagnosis in Michał Tadeusiewicz (SM’92) received the
linear electronic circuits,” Int. J. Circuit Theory Appl., vol. 28, no. 3, Ph.D. and D.Sc. degrees in electronic and
pp. 245–262, 2000. electrical engineering from the Lodz University of
[10] M. Tadeusiewicz, S. Hałgas, and M. Korzybski, “An algorithm for Technology, Łódź, Poland.
soft-fault diagnosis of linear and nonlinear circuits,” IEEE Trans. He is currently a Full Professor with the
Circuits Syst. I, Fundam. Theory Appl., vol. 49, no. 11, pp. 1648–1653, Department of Electrical, Electronic, Computer,
Nov. 2002. and Control Engineering, Lodz University of
[11] M. Tadeusiewicz and S. Hałgas, “An algorithm for multiple fault Technology, and the Head of the Nonlinear Circuits
diagnosis in analogue circuits,” Int. J. Circuit Theory Appl., vol. 34, and Systems Division. He has authored and
no. 6, pp. 607–615, 2006. co-authored about 194 technical papers, two books,
[12] M. Tadeusiewicz and S. Hałgas, “Multiple soft fault diagnosis of and 15 textbooks. His current research interests
nonlinear circuits using the continuation method,” J. Electron. Test., include theory and analysis of nonlinear circuits and fault diagnosis of
vol. 28, no. 4, pp. 487–493, 2012. analog circuits.
[13] M. Tadeusiewicz and S. Hałgas, “Global and local parametric diagnosis
of analog short-channel CMOS circuits using homotopy-simplicial
algorithm,” Int. J. Circuit Theory Appl., vol. 42, no. 10, pp. 1051–1068,
2014.
[14] M. Tadeusiewicz and S. Hałgas, “Multiple soft fault diagnosis of BJT Stanisław Hałgas received the Ph.D. and
circuits,” Metrol. Meas. Syst., vol. 21, no. 4, pp. 663–674, 2014. D.Sc. degrees in electronic and electrical engineering
[15] M. Tadeusiewicz and S. Hałgas, “Multiple soft fault diagnosis of from the Lodz University of Technology, Łódź,
analog circuits using restart homotopy method,” Elektron.-Konstrukcje, Poland.
Technol., Zastosowania, vol. 54, no. 12, pp. 87–91, 2013. He has authored or co-authored about one
[16] M. Tadeusiewicz, S. Hałgas, and M. Korzybski, “Multiple catastrophic monograph, one textbook, and 100 technical papers.
fault diagnosis of analog circuits considering the component tolerances,” His current research interests include nonlinear
Int. J. Circuit Theory Appl., vol. 40, no. 10, pp. 1041–1052, 2012. circuits analysis and fault diagnosis of analog
[17] M. Tadeusiewicz, A. Kuczyński, and S. Hałgas, “Catastrophic fault circuits.
diagnosis of a certain class of nonlinear analog circuits,” Circuits, Syst.,
Signal Process., vol. 34, no. 2, pp. 353–375, 2015.