Failures by Design
By Jose Ramon Novoa, Future Electronics Spain
While design for manufacture has come a long way
from the days when prototypes were ‘thrown over the
wall’ into production, there is still much more to be
done. And there is a huge incentive to get it right. In
today’s ultra competitive market, the product that can
be manufactured and tested in a cost effective and
timely manner – while maintaining high yields – will
reach the consumer sooner and reap greater profits.
Lack of due consideration to Design for Manufacture
(DFM) and Design for Test (DFT) can mean poorly
assembled and insufficiently tested products reaching
the customer. This inevitably leads to a high field
failure rate, expensive recalls and damage to a company’s
reputation.
The key to successful DFM and DFT is to include
manufacturing engineers early in the design process
and to understand their problems with issues such as
component obsolescence. Similarly, the test department
often suffers from insufficient test coverage due
to the designer’s lack of foresight when moving the
product from the research and development (R&D)
laboratory to the factory.
In designers’ defense, they face tougher challenges
than ever before, in making their products factory
friendly. Smaller boards with finer pitch components
and more densely packed interconnects can make it
very difficult to add test access points.
Fortunately, test and component vendors have been
working hard to resolve many of these issues. For example,
suppliers have eased the move to Restriction
of Hazardous Substances (RoHS)-compliant products
by offering services such as Bill-of-Material (BOM)
audits.
And designers can take advantage of functional test
techniques such as Built-in Self Test (BIST) and
boundary scan (also named JTAG and based on the
IEEE 1149.1 standard) to dramatically improve test
coverage. Moreover, forward thinking manufacturers
such as National Semiconductor are helping by incorporating
BIST and boundary scan directly into their
components.
Building in Quality
Designing a product with due consideration to test
throughout its life will improve the product’s total
quality. But in the R&D lab, long term test considerations
are often neglected in the rush to meet project
deadlines.
While a designer continually debugs and analyzes
his product throughout the development cycle, the
lab’s test regimes will not stand up to the rigors of
a production environment that has as its top priority
maximizing output. Functional ‘go/no go’ tests are
substituted for detailed tests using oscilloscopes and
logic analyzers.
It is very important that these functional tests fully
exercise the product, otherwise its integrity cannot
be guaranteed. These test procedures should be continually
improved using feedback from factory and
field failures to enhance coverage and improve test
yields.
However, it is important to realize that comprehensive
test coverage cannot compensate for sub-standard
manufacturing. In fact, a well regulated manufacturing
process allied to a test system with reasonable
coverage is better than a poorly controlled process
that relies on comprehensive test coverage to pick up
faulty products.
Consider a simple example: a manufacturing facility
with 90% manufacturing yield and fault coverage of
99.99% will allow one faulty product into the field per
hundred thousand. That is the same as a system with
a 99.99% manufacturing yield with 90% fault coverage.
But the cost of repair of a faulty product at test
is ten times lower than at the assembly stage. So it is
always better to get it right in production.
Aiding Production
The designer can help his manufacturing colleague
by understanding his challenges. A good example is
RoHS legislation. The designer must specify RoHS-compliant
components throughout, because just one
non-compliant chip will cause the product to fail to
meet the legislation, and even worse, can contaminate
the whole soldering process.
Unfortunately, component suppliers have taken different
views regarding part numbering, which can cause
confusion on the factory floor. Some have retained the
same part number for Pb-inclusive and Pb-free components,
differentiating them only with a date code.
Others have retained the original part number for the
Pb-inclusive version and applied a new number to the
Pb-free item.
The good news is that Future Electronics can supply
comprehensive documentation on a supplier’s RoHS
strategy and specifications to the designer. In addition,
Future can help with component labeling systems
and support the transition to RoHS-compliance.
Electronic components have a limited market life,
and sooner or later the manufacturer makes the device
obsolete. Consequently, designers need to take
into account the status of the devices they are using,
matching the likely component lifetime with the
planned lifetime of their design.
Again, distributors are the best source of information
on this, as they maintain direct contact with the
manufacturers. Future Electronics informs buyers of
impending end-of-life notices and offers alternatives
or ‘last time buy’ information.
Nonetheless, designers should always consider second
sources from the earliest stages of the design,
approving alternatives with full part numbers and
minimum order quantities.
Designers should also assist their manufacturing colleagues
with the use of in-system programmable devices.
For example, the programming pins for these
types of devices should be easily accessible and not
connected to loads such as large capacitors. This is
straightforward if it is considered early during PCB
layout, or even at the pin assignment stage.
Improving Test
During manufacturing test, two concepts are important:
observability and controllability. Observability is
a measure of the ease with which circuit signals can
be observed at the PCB or silicon level. Controllability
is a measure of the ease with which test signals can
be applied at the PCB or silicon level.
The more complex and densely packed a system is,
the more difficult it is to optimize observability and
controllability because it is harder to get access to
test points.
To overcome this difficulty, several new test techniques
have been introduced to increase test coverage
in complex products. These are:
- Physical and functional partitioning
- Increasing effective coverage of existing test points
- BIST
- Structured test systems
Functional partitioning isolates different functions of
a circuit in functionally independent and complete
modules and can be used for hardware and software.
For example, even though the clock path connects
to most parts of the circuit (and can be controlled by
some parts), it is important to isolate it in order to
define problems that may be directly due to clock malfunction.
The same technique is applicable to other
circuit elements such as power supply circuitry and
memory functions.
A second partitioning technique is physical partitioning.
For example, it is good practice to group analog
circuitry in a different area from digital devices, primarily
to avoid interference, but also to apply a different
test procedure to each part. CAD tools used
to generate PCB schematics can sometimes conflict
with this requirement, as their primary function is
usually to minimize PCB footprint and net lengths. In
doing so, they can mix analog and digital circuitry, and
high- and low-voltage paths. Designers will need to
take note of this when reviewing the output from their
CAD tools.
As electronic designs make increasing use of finepitch
devices, or those with hidden leads such as Ball
Grid Arrays (BGA), and PCBs with many layers, many
circuit nets are inaccessible to the probes of bed-of-nails
testers. Consequently, the test points that are
available must offer maximum observability and controllability.
Passive test points provide analog signals
for oscilloscopes, or digital signals for digital analyzers.
Active test points allow input signals to increase
the controllability of the circuit, such as initializing
digital circuits, bypassing internal clock sources with
customer defined clock signals, or defining feedback
loops.
Designers are the best people to define these points
because they know the critical nets that affect most of
the design. There are programs that can simulate the
complete board and its failure coverage;
these are especially useful if a designer
has simulation models for all devices on
the board. However, fault coverage can
usually be improved by manual intervention
and heuristic techniques.
BIST is primarily based on signature analysis
and consists of self generation of many
pseudo-random test vectors that generate
a signature at the output of the tested device.
For a digital circuit with known initial
conditions, the generated signature should
be unique.
BIST is not only useful for auto testing
ASICs or complex blocks, but also transmission
lines that otherwise would be very
difficult to test. An example is an LVDS
transmission line that due to its differential
and high-speed characteristics is difficult
to check using structural test procedures.
However, by using BIST the emitter can
generate a long test pattern that is sent to
a receiver, which then compares received
values against expected ones. The receiver
then indicates any errors. This system
tests the emitter, receiver, connectors and
cables.
BIST is not widely used in board designs, as there is
no standard procedure for integration into larger systems.
It is mainly used to enhance testability of single
ICs prior to board assembly. However, thanks to the
recent introduction of structured testing techniques, it
is now possible to add BIST capabilities to assembled
boards, for in-field testing, using the same test vectors
employed during factory test.

Figure 1: Elements of a boundary scan device
Structured Test Using Boundary Scan
A structured test is one that closely defines the methodology,
allowing different designs to be tested with
the same equipment and procedures. The most widely
used structured test is based on boundary scan.
Boundary scan provides a means to test connections
between ICs on an assembled board without using
physical test probes, via a Test Access Port (TAP). The
technique adds a boundary scan cell that includes a
multiplexer and latches to each pin on the device (see
Figure 1). Boundary scan cells in a device can capture
data from pin or core logic signals, or input data via
the pins. The captured data is then serially shifted out
and compared to the expected results.
The method is controlled via a serial data path called
the ‘scan path’ or ‘scan chain.’ By allowing direct access
to nets, boundary scan eliminates the need for a
large number of test vectors, which are typically required
to initialize sequential logic. Tens or hundreds
of vectors may be sufficient to conduct a test that had
previously required thousands of vectors.
Potential benefits of boundary scan are shorter test
times, higher test coverage, increased diagnostic capability
and lower capital equipment cost. A further
advantage is that it can be used for programming
devices after manufacturing, or for emulation with
the final board during the software design phase.
Moreover, boundary scan remains accessible for test
in the field for maintenance purposes, verification and
software updates.
Unfortunately, many designs include few or no devices
with boundary scan capability. Fairchild Semiconductor,
however, offers its SCAN family of logic devices
such as the SCAN18373. This is a transparent latch
with a three-state output, and includes JTAG boundary
scan. The device also provides JTAG functionality
to the nets connected to it and if a designer needs
a transceiver, flip-flop or line driver, there are other
members of this family that will offer JTAG functionality
to these functions.
While boundary scan chains can be cascaded, it
makes sense to partition chains for programmable
logic, microprocessors and ASICs in the same way as
analog and digital elements are partitioned during traditional
test. This allows independent control of each
area.
National Semiconductor supplies the SCANSTA111
for chain partition. The SCANSTA111 extends the test
bus into a multi-drop test-bus environment. The advantage
of a multi-drop approach over a single serial
scan chain is improved test throughput and the ability
to remove a board from a multi-board system while
retaining test access to the remaining modules. An
improved version of the SCANSTA111 is the SCANSTA112,
which supports up to seven local boundary
scan chains that can be accessed individually or combined
serially.
In addition, the SCANSTA101 converts
parallel commands to serial JTAG vectors
that can be introduced in the field to test
the boards using the same procedures as
in manufacturing, permitting in-field test
and reprogramming for systems with limited
access.
Several high speed LVDS products from
National Semiconductor, such as the
SCAN921023 or SCAN921224, include
BIST to verify at high speed not only the
boards, but also the LVDS transmission
lines communicating at full speed.
Cooperation is Key
While education, supplier initiatives and innovative
test techniques such as BIST and
boundary scan have significantly eased the
designer’s DFM and DFT responsibilities, it
is internal communication that is by far the
best tool.
It is amazing what a designer can learn
from simply touring the factory floor and
understanding the challenges faced by
manufacturing. This one action can lead a designer
to select, for example, one component over another
because it has a more printer friendly profile, or because
its packaging is able to withstand higher reflow
thermal profiles.
Manufacturing engineers could also ease the designer’s
life by reviewing proposed BOMs, PCB profiles
and test regimes to offer feedback and advice. Finally,
do not forget your component supplier. Future Electronics
employs talented engineers who are willing to
offer advice on DFM and DFT founded on their vast
experience dealing with hundreds of customers.