Oscillation based analog testing - a case study

Miljana Milid, Miona Andrejevic StoSovic, and VanCo Litovski*

Faculty of Electronic Engineering, University of NiS, Serbia
* {miljana.milic; miona.andrejevic; vanco.litovski}@elfak.ni.ac.rs

Abstract— The Oscillation Based Testing (OBT) method The implementation of OBT depends on the very
represents an effective and simple solution to the testing
circuit under test. The main reason for that is the ne-
problem of discrete continuous time analog electronic cessity to create an oscillator out of a given circuit.
niters. Its implementation, however, still imposes need
That is one of the difficulties related to the implemen-
for knowledge of the circuit under test (CUT) behaviour
tation of the method. In addition, some of the faults,
and the simulation algorithms of oscillators in the time
especially the soft ones, create fault effects that are
domain. In this paper we describe the implementation of
not easily discernible. Moreover, while mainly the
the OBT method to a second order notch cell realized
number of test points is reduced to one, the oscilla-
with one operational amplifier. For simulation we used
tor's output, the problem of measurement is still not
LTspice. A realistic model for simulation of the ope-
fully solved since one has to decide which and how
rational amplifier in time domain was used. The results
obtained confirm the hypothesis of usefulness of OBT. many parameters of the response are to be extracted.
Finally, the method needs to bring the oscillator into a
Single soft and catastrophic faults are considered with
steady state what frequently slows down the testing
more detail while double soft faults are exemplified only.
Ideas for future work are suggested. process. It is worth mentioning that there are rare si-
tuations where no fault effect may be observed when
using one testing point (output voltage) so additional
Key word - active filter, testing, fault dictionary,
oscillator, simulation, BIST, design for testability. measurements are needed such as /DD [2], [3].

In this paper we describe our experience on imple-

I. INTRODUCTION
mentation of the OBT for testing the second order
One of the fundamental problems in analog testing notch cell of discrete analog active filters. This cell is
is the synthesis of the test signal. One has to choose the most frequently used one when cascade synthesis
among virtually unlimited possible variants. Analog of active filters is considered. Hence the importance
testing may be done by DC signals, in the frequency of the case study. It will be shown that the opamp mo-
domain, and in the time domain. O f course, if neces- del used in simulation is of crucial importance for the
sary, test signals from several domains may be used implementation of the method what is the main result
simultaneously. The DC signals are usually checking of the paper.
for fault effects related to the quiescent conditions and A fault dictionary will be created by simulation.
noniinearities, while in the frequency domain one has Discrete operational amplifiers available on the mar-
to find the spectrum of the testing signal in order to ket will be implemented. This is of crucial importance
activate the fault effect. In the time domain one is to to allow for distinction between analog and linear cir-
search for a signal waveform or several of them that cuit testing what is frequently groundless mixed in the
will enable testing in the shortest possible time so literature. A n exhaustive list of single faults and a list
optimizing the production and decreasing the price of of most probable double faults will be used. The
the product. problem of feed-back circuit synthesis will be solved
On the other side one may observe the problem of in the simplest manner. As the only response parame-
response measurement. Namely, it is always a ques- ter the oscillation fi-equency will be selected to em-
tion as to how many test (measurement) points are phasize the difference between various fault effects.
needed and which quantities are to be extracted as a In the next we will first give a short review of the
favourable measure of the state of the circuit. OBT. Then the notch biquad cell will be considered as
There exist, however, a technique that needs no such. Next, the simplest oscillator circuit obtained
test signal. It is known as the oscillation based testing from the notch cell will be analysed theoretically using
(OBT) [1]. The basic idea behind this powerful me- the ideal opamp model. After fault insertion the fault
thod is to create a redundant feed-back loop that is to dictionary will be created and analysed for soft and
be activated during testing only. By measurement of catastrophic faults.
the output signal and by comparison with the response
of the fault-free circuit, one may conclude whether II. T H E OSCILLATION B A S E D TESTING METHOD
there are defects in the circuit or not. Note, while
usually one creates test signals targeting specific When proper feed-back circuit is added to an ana-
faults, here all possible defects are targeted with only log building block one may create an oscillator. By
one measurement that makes OBT very effective. measuring the output voltage (or some other respon-

18 MIPRO 2011/MEET
 ses) of that circuit one may extract quantities, such as proper structuring o f the analog subsystem on the chip,
the frequency, the amplitude o f the first and other har- as shown in Fig. 1, one may create possibilities to con-
monics, the D C value o f the output voltage, etc. that trol the testing process from outside o f the chip. In mo-
may contain information on the presence o f a fault in re complex mixed- signal architectures, however, ba-
the circuit. M a i n advantage o f this method is avoidan- sed on O B T , one may implement the built in self test
ce o f the search for input stimuli and appropriate mode (BIST) concept, too. In that case one supposes that a
of operation o f the circuit; independence of the method digital signal processing (DSP) unit is implemented
on the type o f faults present i n the circuit (soft or somewhere on the chip and used while the device is in
catastrophic); and simplified selection o f the test point test mode to evaluate the output-signal parameters
and measured quantity at the output [1], [4], [5], [6]. what is then used for qualification o f the circuit as
fault-free or faulty.
From the design for testability ( D F T ) point o f view
this method offers exceptional advantages. Namely, by
^Circuit under test (CUT)

Analog
Inpiits|_ block 1
Analog
block 2
1 _ Analog
block N-1
, Analog
block N Outputs

Testing model - _ " '!-

Additional Control I Output
AMUX
circuitry logic 1 Oscilations
Scan test
signals

Test logic

Fig 1. Architecture of the O B T

III. T H E N O T C H B I Q U A D C E L L The circuit o f F i g . 3 is a band-stop filter with a

notch frequency atf^ =l/(2nRC). It is the basic buil-
As a case study for implementation o f the oscillati- ding block for the active cell o f Fig. 4 which w i l l be
on based method the Sallen-Key non-inverting second here considered as circuit under test ( C U T ) .
order notch filter cell w i l l be used. The importance o f
this cell comes from the fact that it is the most fre- I V . T E S T I N G A N O T C H - C E L L BY T H E
quently used cell within every analog filter implemen- OSCILLATION-BASED METHOD
ted as a cascade, no matter o f the position i f its pas-
sband on the frequency axis as exemplified in Fig. 2. The oscillator circuit so obtained is depicted in Fig.
5. The fault free circuit w i l l oscillate at fo. When de-
fects are present one may observe three different situ-
ations: a) The faulty circuit oscillates at fo, b) the faulty
circuit oscillates at some other frequency /Q , and c)
there are no oscillations i n the circuit. The testing will
be successful i f there is any difference between the os-
cillating frequencies and i f there are no oscillations.
Fig. 2. Amplitude characteristic of a tenth order band-pass filter
realized by four notch and one band-pass cell Unfortunately, i f nominal frequency is observed i n a
faulty circuit, i n general, there is no confirmation that
The schematic o f the cell is based on twin-T circuit no faults are present in the circuit.
as shown i n Fig. 3, while the very cell is depicted i n
Fig. 4 [7].

R 2C R

c c
R/2

I
Fig. 3. Twin-T circuit Fig. 4. Sallen-Key (SK) non-inverting second order notch filter cell

MIPRO 2011/MEET
 V. F A U L T SIMULATION

c Our goal is to create a fault dictionary that is a tab-

le containing faults and fault effects. To get it a large
^2
number (As large as the number of defects conceived
in advance is.) of repetitive simulations is to be perfor-
/?2 med. For every simulation a fault is to be inserted in
the original oscillator circuit so creating a new oscilla-
'out

in Us
tor. Note that simulation of an oscillator is not a stra-
ightforward task since one usually implements an A-
stable integration rule (such as Euler-backward) for
solving the differential equations of the oscillator [8].
F i g . 5. R C oscillator based on the N o t c h filter cell Here however, since the circuit is unstable, one needs
to use an integration rule that is not A-stable (trapezo-
The fault list used contains both parametric (soft) idal, for example).
and catastrophic faults.
When parametric faults are considered, the oscilla- For correct simulation one needs to use a complete
tor is treated as a linear circuit so one may use the usu- schematic of the operational amplifier or a qualified
al oscillator analysis method-to get the expected oscil- model that performs well from the phase-shift point of
lation frequencies. Namely, by writing the modified view. To shorten the computation time we used a
nodal equations [8] for the circuit of Fig. 5, and equa- model as shown in Fig. 6 which is redrawn based on
ting the system determinant to zero, after separating the SPICE code given in [3] with the parameters ac-
the real and imaginary part, one obtains the following commodated to Linear Technologies LTlOOl [10]. To
two expressions for the possible oscillation frequen- illustrate how successful the modelling of the phase
cies: difference inserted by the operational amplifier is, the
2 A-l simulation results for a fault free circuit (oscillator) are
wo = depicted in Fig. 7. Two signals are drawn, the operati-
onal amplifier input (difference of voltages at the input
(1) terminals) and its output (the operational amplifiers
and output voltage). One can deduce from the drawing a
time delay of 0.05 ms what is equivalent to a
significant phase shift of 1.018 rad or 58.3°.
0)0^ =

(2)
where:
/( = A =1 + ^ 4 / ^ 5 . (3)
One of these expressions is usually required for
frequency calculations while the other is needed to
find the necessary value of A for sustained oscillation.
The above expressions were derived under conditi-
on that the operational amplifiers are ideal with infinite
gain which is not true in a real circuit. This is impor-
tant since the closed-loop-gain of the oscillator circuit
is characterized by both modulus and phase. The latter
is frequency dependent and fiindamentally determines
the oscillation frequency. The operational amplifier's
phase shift becomes of importance in this situation and
has to be taken into account. So, the expressions given
by (1) or (2), may be used as reference only. This will
(!) 0
be shown after experiments with simulation are per-
formed. It is worth mentioning that the need to include
the operational amplifiers phase shift was earlier men-
tioned in the literature [9], The idea was, however, im- F i g . 6. C i r c u i t diagram o f the op-amp's model
plemented in the frequency domain and led to con-
clusions quite different than the ones we are reporting A. Defect models
here. The defects here are categorized in several groups:
• Catastrophic defects within the RC circuit.

120 MIPRO 2011/MEET

 • Parametric defects within the RC circuit. most perfect fault coverage of catastrophic faults.
• Separate examples of multiple parametric defects
T A B L E I. CATASTROPHIC DEFECTS (SC stands for; short circuit. O P
stands for: open circuit. No stands for: no oscillations)
I-'./5.75
Measured Computed
V's:
Defect Element
/o [kHz] / fomz]
No. relative
type value
increment
U^-^ -i4.4---t--ii 44 -.4 4 r%i
(1) (2)

Fault

f T +r \
Mr 1
2
free
3.30/0.0

No
3.30

0.389
3.30

No
I.-.1-LAI i . ; SC: /J|

• rf J-.; 3 SC:/?2 No No No
;4 MM 4 SC:^3 ^3=0 n No 0.389 No

-2
1 ll 5 OP: /J, /?i=oon 3.40/3.0 No No
t:1-

-2.5 6 OP:/?2 R2=<x>n No No 0.389

0 001
lime (s) 7 a?:Ri /f3=oon 3.50/6.1 No 0.55
Fig. 7. Responses of the fault free oscillator of Fig. 4 deinonstrating
the phase shift introduced by the op-amp. 8 SC: C, C,=<x)F 3.50/6.1 No No
For every passive element within the RC circuit 9 SC:C2 C2=ooF No No 0.389
short- and open circuit is considered as a catastrophic 10 No No 0.674
SC:C3 C3=ooF
fault. After insertion, the resulting schematic was si-
mulated and the results are shown in Table I. We de- 11 OP: C, C,=OF No 0.275 No
note ..measured/o" the oscillation frequency obtained 12 OP: C2 C2=0F No No No
after circuit simulation. The ..computed" frequency is
obtained either from (1) or from (2) depending which 13 OP:C3 C3=0F No 0.550 No
is positive. The resistors in the negative feed-back cir-
cuit were chosen to be /?4=l-4kn, and Rs=2.lk£l,
The oscillating frequencies obtained from (1) and
leading to a gain A=\.661. (2) are placed in the last two collumns of Table I, res-
pectively. If negative frequency is obtained from the
formula, a "No" is placed in the appropriate cell of the
B. Fault dictionaries Table.
A table containing a set of faults and the corres- By observing the last two columns of Table I one
ponding responses of the system is referred to as fault may easily conclude that implementation of ideal op-
dictionary. These are important not only for testing but amp model can't be used as reliable information from
also for diagnostic purposes. Since the number of pos- testing point of view. As can be seen neither the fre-
sible faults in a system may be very large, when quencies of „measured" and „computed"yb are equal
creating the fault dictionary, one generally chooses a
nor the claim about the existence of the oscillations
set of most probable faults. If so, one claims that the
may always be considered as true. Special problem re-
structured approach to testing is applied.
presents the fact that one is always due to evaluate
Search trough the fault dictionary, in general, both expressions to check for fo.
enables fault coverage to be established and the test
Wheii considering parametric defects one is to de-
signal to be qualified.
cide as to what extent a change of the element value is
The first experiment was creation of a fault dictio- to be treated as a defect having in mind that small
nary for the catastrophic faults. To get the „measured" changes may be looked up as tolerances coming from
fo, simulation of the oscillator with the fault inserted the technology process. Here a parametric defect will
was performed while the „computed"yb was obtained be seen when the element value within the RC-circuit
by substituting the element values in (I) and (2). The is changed for 20% in comparison to its nominal va-
results are shown in Table I. A "No" was placed for lue. Both changes are taken into account positive and
the "measured" frequency when no oscillations were negative. As a resuh Table II is produced. It contains
the faults and the fault effects for all conceived soft
observed from the simulator. One may deduce from
faults in the RC-circuit.
the fourth column of Table I that only in four out of
twelve cases the circuit keeps oscillating. In eleven out By inspection of Table II one may easily conclude
of twelve cases the fault effect is clearly recognizable. that in seven out of twelve cases there is difference
In one of the cases (OP; Ri), the oscillation frequency between the behaviour of the fault-free and the faulty
is incremented for 3% only. In general, however, one circuit. In one case the change in the frequency value
may state that by implementation of OBT we got al- is so small (row no. 6 of Table II) that can't be used as

MIPRO 2011/MEET
 a reference for making decision about the existence of o f extracted parameters o f the output signal what is
the fault. Here again we may conclude that the fauU considered as a task for future work.
coverage is almost perfect. A s for the computed fre- T A B L E III. T w o DEFECTS P R E S E N T S I M U L T A N E O U S L Y
quency fo, one may state that it is i n discrepancy with
the measured one even for the fault-free circuit what Measured Computedfo
was the case in Table I too. /o[kHz]/ [kHz]
No. Defect type
relative
increment [%] (1) (2)
T A B L E II. P A R A M E T R I C DEFECTS
1 Fault free 3.30/0.0 3.30 3.30
Measured Computedfo
/o[kHz]/ IkHz] 2 1.2-C2; I.2-C3 2.90/12.1 0.562 0.589
No, Defect type
relative
(2) 3 O.8C2; O.8C3 3.70/12.1 No 0.634
increment [%] (I)
1 Fault free 3.30/0.0 3.3 3.3 4 I . 2 C 2 ; O.8C3 3.50/6.1 No 0.575

2 1.2/?i 3.30/0.0 No 0.561 5 O.8C2; 1.2Cj 3.30/0.0 No 1.123

3 0.8/?i 3.30/0.0 0.891 0.701 6 1.2«,; 1.2c, 3.40/ 3.0 No 0.502

4 1.2/?2 3.00/0.0 0.615 0.584 7 0.8-.R,; 0.8C, NO 0.540 0.784

5 0.8/Jj 3.60/9.1 No 0.660 8 l.2/?i;0.8C, 3.25/3.1 0.735 0.660

6 \.2Ri 3.30/0.0 No 0.605 9 O.SiJ,; 1.2C| 3.35/1.5 No 0.610

7 O.SRi 3.20/3.0 0.809 0.630 10 1.2/J2; I . 2 C 2 2.90/12.1 0.529 0.552

8 I.2C, 3.35/1.5 No 0.527 11 O.S/Jj; O.8C2 3.80/15.1 No 0.705

9 0.8C| NO 0.615 0.728 12 \.1R^, O.8C2 3.20/3.0 0.763 0.628

10 1.2-Cj 3.20/3.0 1.23 0.584 •3 0AR2\.2C2 3.50/6.1 No 0.628

11 O.8C2 3.50/6.1 No 0.660 B y inspection o f Tables I, II, and III, one may find
12 1,2C3 3.10/6.1 0.710 0.625 that some o f the fault effects while different from the
fault-free response, are identical among them. One re-
13 0.8-C3 3.60/9.1 No 0.600 fers to this type o f faults as ambiguity groups or func-
tionally equivalent faults (FEF). These groups are not
The possible number o f double soft defects is affecting the testing since the go-no-go concept is
much larger than i n the case o f the single faults. O f applied i.e. the information that there is a fault in the
course there is much smaller probability for double system is enough to decide for the circuit to be discar-
faults to occur. This is why a reduced set o f pairs o f ded. In cases, however, when diagnosis is to be perfor-
soft faults was considered as shown in Table III. One med F E F is limiting the diagnostic resolution. Finding
may observe that in eleven out o f twelve cases the altematives that lead to resolution between F E F is,
again, a task for future work.
faulty circuit exhibits new value for f^. In nine out o f
ten cases the decision making is possible.
VI. C O N C L U S I O N
C. Overview of the results It was shown by an example that the Oscillation
When evaluating the O B T approach implemented Based Testing ( O B T ) method represents an effective
to the notch filter cell we have to consider the follo- and simple solution to the testing problem o f discrete
wing. First o f all, no test signal was needed to be continuous time analog electronic filters. Its imple-
found. That is a big advantage for the test engineer. mentation, however, still imposes need for knowledge
Second, only one test point was observed i.e. the o f the circuit under test ( C U T ) behaviour and the si-
circuits output, what is the most natural way of access mulation algorithms o f oscillators i n the time domain.
for measurement. Finally, only one quantity was ex- In this paper we describe the implementation o f the
tracted as a measure o f fault coverage: the oscillation O B T method to a second order notch cell realized
frequency. with one operational amplifier. Single soft and cata-
strophic faults were considered with more detail while
With such a simple procedure a waste number o f double soft faults were exemplified only. The most
faults were covered that lead to a conclusion that the simplified approach was implemented meaning that
O B T method, for this example, is an excellent testing only one test point was used while for the creation o f
concept. the fault dictionary only one parameter o f the output
voltage was extracted. Ideas for future work were
Further improvement o f the fault coverage may be
suggested.
expected to be achieved by simple broadening the list

122 MIPRO2011/MEET
 Besides the validation of the O B T method for tes- [4] K . Arabi, and B. Kaminska, "EfTicienl and accurate testing of
ting of analog filters we here, for the first time, analog-to-digital converters using oscillation-test method,"
Proc. of the European Design and Test Conference (ED&TC
implement a realistic model of the operational amp- 97), Paris, France, Marchl997, pp. 348-352.
lifier that leads to more dependable conclusions rela-
[S] K . Arabi, and B. Kaminska, "Oscillation-test methodology for
ted to the presence of a fault in the circuit. We expect low-cost testing o f active filters," IEEE Trans, on
that this approach will be more appreciated when Instrumentation and Measurements, Vol. 48, No. 4, August
testing of circuits containing larger number of ope- 1999, pp. 798-806.
rational amplifiers is considered. [6] S. Das, el al., "Testing analog and mixed-signal circuits with
built-in hardware - A new approach," IEEE Trans, on
Instrumentation and Measurement, Vol. 56, No. 3, June 2007,
pp. 840-855.
REFERENCES
[7] Huelsman, L.P., "Active and Passive Analog Filter Design:
[I] K. Arabi, and B. Kaminska, "OsQillation-test strategy for A n Introduction", Mcgraw-Hill College, 1993.
analog and mixed-signal integrated circuits," Proc. of the
I4the IEEE VLSI Test Symposium, (VTS 96), Princeton, [8] Litovski, v., Zwolinski, M . : ..VLSI circuits simulation and
New Jersey, April/May 1996, pp. 476-482. optimization", Chapman and Hall, London, 1997
[2] Alii, P.K., "Testing A C M O S Operational Amplifier Circuit [9] Zamik, M.S., Novak, F. and Maiek, S., "Design of
Using A Combination O f Oscillation And Iddq Test oscillation-based test structures for active RC filters", l E E
Methods", M.Sc. Thesis, Louisiana State University, USA, Proc.-Circuits Devices Syst., Vol. 147, No. 5. October 2000,
2004. pp. 297-302.
[3] G. Hu, H . Wang, M . Hu, and S. Yang, "Oscillation test [ 10] http://www.linear.com/pc/downIoadDocument.do?navId=HO,
strategy for analog filters by monitoring output voltage and C1 ,C 1154.C 1009,C 1021 ,P 1206,D 11418.
supply current," Thinghua Science and Technology, Vol. 12,
No. SI, July 2007, pp. 78-82.

MIPRO 2011/MEET