Browse Prior Art Database

Guardband Test Process with Interaction Assessment

IP.com Disclosure Number: IPCOM000108575D
Original Publication Date: 1992-Jun-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 5 page(s) / 183K

Publishing Venue

IBM

Related People

Breyfogle III, FW: AUTHOR

Abstract

A process is described to assess the design "safety factor" of multiple factor effects simultaneously. This process is relatively easy to implement utilizing a "small sample size" with automatic test techniques.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 37% of the total text.

Guardband Test Process with Interaction Assessment

       A process is described to assess the design "safety
factor" of multiple factor effects simultaneously.  This process is
relatively easy to implement utilizing a "small sample size" with
automatic test techniques.

      Guardband test techniques are used to assess the "safety
factor" of a product design.  With this technique, a limited number
of products are stressed beyond their normal operating conditions
until either a functional failure occurs or additional stressing will
cause an abnormal failure condition.  These stress conditions can
involve temperature, voltage, oscillator frequencies, etc.

      One previous approach when assessing the guardband of several
factors is to change the stress of one factor while holding the other
factor at nominal conditions.  Fig. 1 shows these measurement levels
at the "*" locations for two factors.

      The above factor effects might be assessed at several level
conditions.  For two factors, Fig. 2 describes this test alternative,
where several levels are noted for factor A:

      Probability plots can be used to graphically evaluate the risk
of having field problems by estimating the percent fail condition at
the specification limits for each of the four test scenarios.
Failure modes that have a significant design exposure will then need
to be "fixed".

      Traditional factorial test design techniques are useful to
efficiently assess the effects of several factors on a response in
one experiment.  Experimental designs can be chosen such that the
primary focus is on just main factor effects or main and interaction
considerations of the factors.  It is desirable from a sample size
and analysis point of view in these designs to consider continuous
responses (as opposed to a binary pass/fail condition). A good
experiment design choice is often 16-trial two-level factorial
experimental design where 5-15 factors can be tested to their limits.
In the five-factor alternative, a 16-trial design can assess five
two-level factors and their two-level interactions.

      For two factors, a factorial experiment can pictorially be
described to have measurements at the A/B "*" locations noted in Fig.
3.  With a 16-trial factorial design, there will be 4 measurements at
each "*"; however, the level states of any other factors (within the
design) will be different for each of the four trials.  The rationale
for testing around the nominal scenario is that normally if the
product performs satisfactorily at tolerance limits, one can often
assume that performance at nominal will be satisfactory.  Note, the
conceptual tolerance limits of factors "A" and "B" are illustrated by
the tick marks next to the axes.

      The process which is next described is more efficient than
considering one factor with other factors held constant (i.e., Figs.
1 and 2 test strategy).  With this process, reduced test time can be
expecte...