Wednesday, July 25, 2007

Launch-off-shift at-speed test

A comparision of broadside-transition-pattern and launch-off-shift techniques show the latter to be viable for testing a 90-nm wireless baseband device.

Both launch-off-shift (LOS)
and broadside-transition-pattern techniques are finding use in the
at-speed test of devices fabricated in 130-nm processes and below. The
broadside-transition-pattern approach is most commonly used, but our
experiments in applying both techniques to the test of a wireless
baseband device show that LOS can provide advantages.






Learn more about scan technology.

At-speed scan test serves applications for which static testing is not sufficient (Ref. 1).
The basic operation of at-speed scan testing involves loading the scan
chains at a slow clock rate and then applying two clock pulses at the
functional frequency (Figure 1). The first pulse
causes a transition to start propagating from a scan-cell. The second
pulse captures the scan cell value at the end of the path being tested.








Figure 1. At-speed scan
test involves loading scan chains at a slow clock rate and then
applying two clock pulses at the functional frequency.


If
the circuit is operational, then the transition will propagate to the
end of the path in time and the correct value will be captured.
Otherwise, if a delay causes a slow propagation, the transition from
launch to capture cell will be slow, an erroneous value will be
captured, and the defect will be detected.

The most popular at-speed scan pattern is the transition pattern (Ref. 2).
A potential fault of slow to rise (0 to 1) and slow to fall (1 to 0) is
modeled at every gate terminal within the design. Automatic
test-program generation (ATPG) tools target these fault sites and cause
a transition using any launch scan cell and capture results using any
downstream scan cell.

Using PLLs for accurate clocks

A fundamental problem with at-speed scan testing is how to apply
accurate clocking for the at-speed launch and capture pulses.
Traditional stuck-at scan patterns are static. Stuck-at clocking for
loading the scan chain and capturing results is often performed at
frequencies between 10 MHz and 40 MHz. At-speed scan testing can load
the scan chains with a clock frequency that is similar to the one used
for stuck-at tests, but the launch and capture pulses must be applied
at the functional frequency.


Supplying the at-speed clock for launch and capture from a tester
becomes more demanding as the desired frequency increases. A solution
of using some basic programmability around device internal phase-locked
loops (PLLs) provides a nice option (Ref. 3). Providing internal PLL control for at-speed test has become a common practice for at-speed scan testing (Refs. 4 and 5).


The most common technique for applying at-speed transition patterns
is referred to as a broadside or launch-from-capture pattern type (Ref. 6), as shown in Figure 2.
With this pattern type, the scan chain is loaded and then placed in
functional/capture mode by forcing scan_enable (SE) to 0. Sometimes, an
extra cycle is added to the test pattern that has no activity to ensure
that the scan_enable completely settles. Next, two pulses are generated
to launch and capture the transition.









Figure 2. With
broadside transition patterns, a scan chain is loaded and then placed
in functional/capture mode by forcing scan_enable (SE) to 0.


The
broadside pattern launches the transition in the functional mode of
operation, so it is likely to propagate transitions along paths that
are real functional paths. Often, the coverage report from broadside
pattern ATPG can be 10% lower than standard static stuck-at patterns.Launch-off-shift patterns

With LOS patterns (Figure 3),
the launch occurs during the last shift while loading the scan chain.
Next, the circuit is placed into functional/capture mode very quickly
so an at-speed functional clock can be pulsed.


ATPG is much easier with LOS
compared to broadside patterns. It is a simple ATPG activity to load
the starting value for a transition directly to the scan cell one shift
before the last and then load the transition value in the last shift.
Broadside patterns require ATPG to calculate the transition value
through the combinational logic, since it is in functional mode during
the launch pulse. In addition, LOS patterns usually report higher coverage than broadside patterns.


LOS reports
higher coverage and is easier for ATPG, so it results in fewer patterns
and faster ATPG run times compared to broadside patterns. So, why is
broadside transition test more popular than LOS patterns?









Figure 3. In a launch-off-shift transition pattern, the launch occurs during the last shift while loading the scan chain.

There are two primary reasons for the reluctance to use LOS
patterns. The first is the difficulty to make the circuit change from
shift mode to functional/capture mode between the last shift and
functional clock pulse. If standard scan_enable architecture is used,
then the scan_enable must be routed as a clock. Furthermore, since
scan_enable goes to all sequential elements, it is a global clock and
must settle at the system clock frequency. One way to work around this
issue is to add pipelining logic throughout the device for scan_enable (Ref. 7).

Pipelining scan_enable adds additional test logic to the design, but
it removes the difficult task of treating scan_enable as a global
clock. As shown in Figure 4, the clock triggers a change within the local scan_enable.


The other common concern with LOS patterns is that they may test the circuit through paths that are not possible functionally. LOS patterns can shift in a transition that is impossible during normal circuit operation.


An important question to ask is how much of the additional coverage
beyond broadside patterns is due to nonfunctional logic? It’s possible
that testing nonfunctional logic during at-speed tests will falsely
report failures and result in yield loss (Ref. 8).

Accounting for false and multicycle paths

During the design process, many paths are determined to be either
false or multicycle paths. A standard Synopsys Design Constraints (SDC)
file lists false and multicycle paths such that special efforts are not
made for timing closure at these paths. Two types of false paths can
exist. Some false paths cannot be sensitized and are not possible
during functional operation (but may be possible during scan mode). The
other types of false paths are paths that are not intended to operate
at system frequencies. Multicycle paths require more than one
functional clock cycle to propagate.









Figure 4. Pipelining
scan_enable adds additional test logic to the design, but it removes
the difficult task of treating scan_enable like a global clock.

Both
false and multicycle paths must be considered during at-speed scan
testing. Scan has the potential to directly load scan cells into a
circuit state that isn’t possible during functional operation. As a
result, false or multicycle paths may be activated during at-speed scan
testing. If the at-speed scan tests fail due to these paths, then
correctly functioning devices may be falsely discarded. The result
could be yield loss.

To avoid such loss, engineers have often tested for false and
multicycle paths during time-based simulation and test program tester
integration, and they have often performed these tasks manually.


Fortunately, automation has been added to ATPG tools so they can now
directly read standard SDC files and extract the timing-exception path
information (Ref. 9).
With this automation, if a test propagates a signal along a false or
multicycle path that is sensitized during ATPG, then the capture scan
cell will capture an unknown X value.

Baseband-chip case study

Metalink, a company that designs wireless and wireline broadband
communication chips, needed to develop an effective test strategy for
its WLANPlus 802.11n-draft-compliant wireless LAN technology, which is
optimized for the networked home entertainment environment. The
company’s WLANPlus family consists of the MtW8171 baseband device and
the MtW8151 RFIC. The MtW8171 baseband chip is manufactured at a 90-nm
low-power process and implements full at-speed scan-test capability.
For this device, at-speed scan was implemented using both LOS and broadside transition patterns.


To reduce the increased pattern count required to cover transition
faults, we implemented compression logic using Mentor Graphics Embedded
Deterministic Test technology. The bring-up of the scan program for
this chip took only two days from the chip arrival to the moment where
all bring-up patterns were up and running at-speed.


We performed experiments to compare the difference between broadside and LOS coverage. We generated the initial patterns by using procedures that define specific clock sequences that can be used.


















Table 1. Comparison of broadside and launch-off-shift

Broadside


Launch-off-shift


Initial coverage


71.38%


78.57%


With SDC


69.55%


72.88%

Table 1 shows the results of transition pattern generation. Initially, the LOS patterns reported 78.57% test coverage compared to 71.38% for broadside patterns. Thus, LOS
appeared to test >7% more faults. Next, we considered false and
multicycle paths, since these paths are not intended to operate at
functional clock rates.


After accounting for false and multicycle paths (MCPs), the LOS and broadside coverages were reduced to 72.88% and 69.55%, respectively. Therefore, 5.69% of faults reported in the initial LOS
detection were false paths and MCPs. Similarly, 1.83% of broadside
detection was due to false paths and MCPs. Based on these results, we
concluded that a significant portion of the advantage in test coverage
with LOS patterns are due to false-path and MCP testing.


These results imply that when effective false and multicycle paths are considered for LOS patterns, the risk of overtesting is reduced. As a result, LOS
can be an attractive ATPG approach for at-speed test.
Pattern-generation time and pattern count can be significantly smaller
than broadside patterns with similar coverage. The combination of
pipelined scan_enable and false and multicycle path consideration solve
the most common concerns with LOS patterns. Broadside patterns should still be used to top-off coverage since LOS patterns will be unable to detect some faults.

What’s next?

Manufacturers are continuing to look for ways to improve the
effectiveness of at-speed scan testing. One approach, referred to as
“timing-aware” ATPG (Ref. 10),
targets small delay defects. It attempts to test each fault by
propagating the transition down as slow a path (smallest slack) as
possible. In this technique, the at-speed test pattern set is more
likely to detect small defects that could escape a normal transition
test set.


Another new approach to at-speed scan testing is to apply a series
of at-speed shifts just before the launch cycle. Such “BurstMode” ATPG (Ref. 11)
helps the at-speed clock pulses behave more like functional frequency
pulses. It lessens the drooping of the voltage supply caused by the
sudden pulsing of at-speed clocks for launch and capture during normal
transition tests. False and multicycle path handling should be
considered with both of these techniques to avoid the risk overtesting.


Meanwhile, techniques such as pipelined scan_enable make LOS
more feasible, allowing users to evaluate the trade-offs between the
two transition pattern types and determine which is the best solution
for them. Broadside patterns offer less logic insertion and less
nonfunctional path tests, while LOS patterns offer faster pattern generation and fewer patterns. The LOS
approach may also be desirable for companies that are interested in
detecting any type of defect, including those that are nonfunctional.
Fortunately, the common concern with overtesting can be alleviated by
ATPG tool handling of false and multicycle paths through SDC files that
are common in design flows.



Powered by ScribeFire.

No comments: