Generalized Fault Modeling for Logic Diagnosis - Models in Hardware Testing - page 136

During test, patterns are applied at the inputs of the circuit and the logic values
at the outputs are observed. As a consequence, logic diagnosis cannot observe the
electrical conditions at the defect sites directly but has to rely on the logic signals
leaving the defect sites towards the outputs. From the perspective of logic diagnosis,
certain internal signals have logic values, which are inconsistent with the fault-free
logic operation of the circuit. By finding for each test a small set of signal lines,
which have to be flipped in order to explain the erroneous outputs, the defect site
itself can be located. This set of flipped lines form the structural aspect of the diag-
nosis result.
With each new pattern, the electrical conditions at the defect sites change. The
downstream logic receives different inputs and provides a different response. Even
the same pattern can lead to different results each time, if the defect site behaves
in a nondeterministic way and the line flips are not always active. The activation
conditions are correlated to the electrical conditions within the defect site. They
form the functional aspect of the diagnosis result.
The combination of the two aspects derived above lead to the notion of a con-
ditional line flip (CLF). Unlike classic fault modeling approaches, which usually
start with certain assumptions about the behavior of the anticipated defects, the con-
struction of CLF does not pose any restrictions on the type of the defects. Therefore,
practically any defective behavior, fault model and diagnosis result can be expressed
in terms of CLFs to some extent.
A CLF is noted by the name of the victim line and an XOR-symbol followed by
a condition clause:
line ˚ Œ condition
The condition clause is evaluated using the original (fault-free) values of the lines
affected by the described defect. A defect may be described by multiple CLFs
on multiple signal lines or even on the same line, if the conditions are mutually
exclusive.
In many approaches the conditions are assumed to be deterministic ( Bhatti and
Blanton 2006 ) or expressible as a Boolean function ( Ubar 2003 ). In recent technolo-
gies, however, such assumptions become more and more restrictive, since defect
behavior is often nondeterministic or timing related. If indeterminism is taken into
account for the conditions, all technology related faults and design related errors can
be expressed by multiple CLFs.
There are two important differences between the CLF calculus and classic fault
models. First, there is no finite set of all possible faults in CLF, since the conditions
can be arbitrary functions over time. Many classical fault models are designed to
provide a finite set of target faults for ATPG. Second, there is no 1-to-1 correspon-
dence between defects and faults. The behavior of a single defect can be described
in many ways, and a single CLF can describe the behavior of many different defects.
Classic fault models on the other hand are motivated by specific defect mechanisms,
where one fault corresponds to a specific defect.
Besides describing the behavior of a defect site, the CLF calculus can also be
used to describe faults. The faults may be of an arbitrary fault model, and by noting
the faults in CLF, the properties and assumptions of the underlying fault model can
be assessed.