Professional Documents
Culture Documents
Atrenta Inc.
White Paper
Atrenta, Inc.
2077 Gateway Place
Suite 300
San Jose, California 95110
1‐866‐ATRENTA (1‐866‐287‐3682)
atrenta.com
Table of Contents
1 Introduction ............................................................................................................ 3
12 Summary............................................................................................................... 17
1 INTRODUCTION
Production testing for complex chips usually involves multiple test methods. Scan‐based automatic
test pattern generation (ATPG) for the stuck‐at defect model has been the standard for many years,
but experience as well as a number of theoretical analyses have shown that the stuck‐at fault model
is incomplete. Many devices pass very high coverage stuck‐at tests and still fail to operate in system
mode.
Analysis of the defective chips often reveals that speed or timing problems are the culprits. At 90 nm
and smaller processes, the percentage of timing related defects is so high that static testing is no
longer considered sufficient. Functional tests have been used to check for at‐speed operation. But
generating functional at‐speed test patterns is difficult and running this volume of tests on the
automatic test equipment (ATE) is expensive. As an alternative, scan test has been adapted to detect
timing‐related defects. Like standard stuck‐at scan tests, high coverage at‐speed scan test vectors can
be automatically generated by ATPG tools. Manufacturing testing of deep submicron designs now
routinely includes “at‐speed” tests along with stuck‐at tests.
Little has been done so far to make front end designers aware of at‐speed test solutions at the
register transfer language (RTL) level of abstraction. This document is intended to present basic
concepts and issues for at‐speed testing, as well as demonstrate the at‐speed coverage estimation
and diagnosis capability built‐in to the SpyGlass®‐DFT DSM product for RTL designers and test
engineers.
2 DESIGNFORTEST AT RTL
There has been an enormous push for design‐for‐test (DFT) at the register transfer level for the past
several years. RTL designers are now faced with requirements to comply with strict DFT guidelines
to meet ever‐tighter design schedules, even though the designer may not understand all the
complexities of defect models and testability techniques. SpyGlass‐DFT provides the platform and
ease of use to RTL designers to quickly identify and fix testability issues without having to become a
test expert.
Leading chip makers realize the importance of looking at the manufacturing test readiness of their
SoCs early in the design phase to prevent design iterations motivated by improved test coverage.
However, most of this effort has been focused on achieving high stuck‐at fault coverage for designs
with feature sizes of 130nm and above. As the industry moves towards deep submicron technologies
of 90nm and below, speed‐related defects are becoming common and gaining more attention.
Transition fault coverage and path delay fault coverage have become equally important and are an
integral part of any robust DFT strategy today.
The test clocks in traditional stuck‐at testing are designed to run on test equipment at frequencies
lower than the system speed. At‐speed testing requires test clocks to be generated at the system
speed and these clocks are often shared with functional clocks from a phase locked loop (PLL) clock
source. Any additional test clocking circuitry affects functional clock skew and thus the timing
closure of the design. SpyGlass‐DFT DSM is designed to specifically address the problems associated
with timing closure due to at‐speed DFT.
At‐speed tests often result in lower than required fault coverage, even with full‐scan and high
(>99%) stuck‐at coverage. Identifying reasons for low at‐speed coverage at the ATPG stage is too late
to make changes to the design and affects schedules significantly. SpyGlass‐DFT DSM identifies
causes of low at‐speed coverage at RTL and helps achieve quick turnaround times for today’s
aggressive design schedules and time‐to‐market challenges.
Figure 2 illustrates a 1 to 0 transition on the AND gate input that propagates through the OR to the
flip‐flop D‐pin. If the 0 value reaches the D‐pin sufficiently before the system clock then the test
passes.
Destination
Flop
0
1 1 ‐> 0
D Q
1 ‐> 0 1 ‐> 0
system
clock
A path delay fault is a fault associated with a specific combinational path from the Q‐pin of one flip‐
flop to the D‐pin of a flip‐flop (possibly the same flip‐flop) where the two flip‐flops are in the same
test clock domain. Two faults are considered for each path:
• the slow to rise transition on the first node of the path
• the slow to fall transition on the first node of the path
Therefore, the number of possible path delay faults is twice the number of paths.
Since path delay faults are path‐specific (from a source point to a destination point) they are usually
defined by critical paths from a static timing analysis (STA) tool.
Note that a test for a transition fault necessarily involves a test for some path fault and a test for a
path fault necessarily involves one or more transition faults.
Unlike stuck‐at faults, at‐speed tests depend on timing and therefore two or more capture clocks are
required. The first clock initiates or launches a transition at the output of a flip‐flop. The second
clock, fired at system speed with respect to the first clock, is intended to capture the test result.
Suppose that in the example of Figure 5, flip‐flops U1a, U1b, U1c, U2a, U2b and U3 are on the same
scan chain and stitched in this order.
In order to use the last shift method, the scan in state before the last scan shift must be as illustrated
in Figure 5. The last shift will cause U2a to change 0=> 1 (the launch event), U2b to remain at 0 and
U3 to go to 0. The scan enable is then changed to system mode and one capture clock is fired at high
speed with respect to the last shift clock. Note that careful timing is required for the scan enable in
order to use the “last shift” method which is one reason why this technique is difficult and not
commonly used.
Consider the 0=>1 test for top input to the OR gate. The last scan in must have U2a = 1 and U2b = 0
so that on the last shift U2b will transition 0=>1. In order for the off path input to the OR to be at a
non‐controlling value, U2a must be 0 which means that U2a is necessarily changing from 1=>0 as
U2b changes 0=>1. If there is no transition fault then U3 should continue to capture a 1.
Since all flip‐flops in the example in Figure 5 are scanned, they can be initialized to either 0 or 1
through scan shifting. The scan shift clock may be fired at slow speed but the two capture clocks are
fired at system speed. Launched transitions for the OR gate eventually arrive at the d‐input of U3. If
the desired transitions do not arrive at the capture FF (U2) then an incorrect data is captured. The
purpose of at‐speed testing is to verify whether or not correct data or corrupted due to transition
delays.
If a 0 => 1 transition is launched at flip‐flop U2a, then a 0 is scanned into U2a and a 1 is scanned into
both U1a and U1b so that the next state function for U2a is a 1. The first capture clock causes U2a to
change state from 0 to 1. A 0 can also be scanned into U1c so that the OR gate off‐axis input is set to a
non‐controlling value after the first capture clock. The second capture clock will store the value
produced by the OR gate into U3.
6 ATSPEED CLOCKS
At‐speed testing critically depends on the application of the capture clocks. If the two clocks
illustrated in the timing diagram shown in Figure 6 are too far apart, then the extra time allows slow
paths through the logic to pass. If the clocks occur too close together, then properly working paths
may fail. This issue influences which faults are candidates for at‐speed testing and how clocks are
generated.
Figure 7 illustrates use of a PLL and special clock logic to generate the two system speed clocks that
are necessary for at speed testing.
scan_en MUX
contro
lle
scan_clk r
scan_en
pll_clk
clk
shift launch/capture shift
in/
Figure 7: PLL clock timing diagram for atspeed test in/
out out
© Copyright Atrenta Inc., 2010 Page 8 of 17
Facilitating At‐speed Test at RTL
The overall flow for Atrenta’s complete RTL testability solution is shown in Figure 8.
Figure 8: Complete RTL analysis solution for stuckat and atspeed testing
Here are some examples of at‐speed rules. These rules identify issues caused by clocks used in at‐
speed tests for timing closure as well as low coverage.
Notice that transition faults along the path marked in green in Figure 16. are launched and captured
on the same clock and therefore are at‐speed testable.
Figure 15: Clock domain crossing – Red path indicates invalid faults
Stuck‐at coverage depends on the amount of scan in the design. Ignoring such factors as latch
transparency or black boxes, if all flip‐flops are scannable, then nearly 100% test coverage is a
reasonable expectation since tests for stuck‐at faults requires that the fault be controlled and
observed. As scan is added to a circuit, the fraction of logic that can be controlled and observed
increases and therefore test coverage for stuck‐at faults increases. SpyGlass‐DFT treats the Q‐pins of
scannable flip‐flops as controllable and D‐pins as observable so, as the scan‐ability increases, the
estimated test coverage increases.
In launch by first capture at‐speed tests, not only must control and observe of the transition pin be
satisfied (similar to stuck‐at testing) but a third condition is also necessary to have control on the D‐
pin of the scan flop. Scan‐ability of the logic combinationally connected to the transition pin provides
the necessary control and observe so that the first two conditions are satisfied.
The next state of the launch flip‐flop must be controlled to the complement of the scan‐in state. A
rising edge event requires the scan‐in of 0 and a next state on the D‐pin of the launch flip‐flop to be a
1. A falling edge event requires the scan‐in of 1 and a next state on the D‐pin of the launch flip‐flop to
be a 0.
If some flip‐flops in the D‐pin fan‐in cone of a launch flip‐flop are not scannable, then the next state
function of that flip‐flop may not be controllable. Such uncontrollable next state functions prevent
flip‐flops from launching dynamic events. The result is that coverage of at‐speed tests, using the
launch on first capture method, is generally less than the stuck‐at coverage. A fundamental reason is
that the output controllability of scan flip‐flops becomes dependent on the system mode
controllability of their D‐pins.
ASIC vendors have reported that even when the stuck‐at coverage is in the high 90’s, it is often the
case that the at‐speed coverage is much less. Vendors also report that none of their current software
tools give any clue as to how to fix the problem or are even able to predict this coverage drop at the
RTL stage. Test engineers must run at‐speed ATPG to determine coverage. Effort to fix low coverage
during ATPG affects design schedules, or designers compromise on test quality if it is too late to make
design changes.
Faults on multi‐cycle paths will be included in test coverage if the number of registers on a multi‐
cycle path is less than or is equal to the number of capture clocks. Otherwise, such faults will not be
included in test coverage.
10.4 Diagnostics
As described in section 9, an important reason for low of at‐speed coverage is uncontrollability for
the next state functions. The testability analysis of SpyGlass‐DFT DSM determines which flip‐flops
have uncontrollable D‐pins. The fan‐in cones for these D‐pins are then analyzed to identify the cause
of uncontrollability such as non‐scanned flip‐flop, black box, non‐transparent latch or X‐generator.
Figure 17 shows a sample testability report from SpyGlass‐DFT DSM.
11.1 Uncontrollability
As discussed in section 9, incomplete controllability for next state functions in system mode is a
cause for decreased at‐speed coverage. Such cases prevent launch events and thereby prevent
generation of specific tests even though such transitions may be possible in normal functions.
This uncontrollability may be caused by flip‐flops that are not scanned, by black boxes that are not
scanned wrapped or other types of X‐generators.
The flip‐flop colored in red in Figure 18 is not controllable because of the X‐source in its fan‐in cone.
1 > 0 event
cannot be
launched by
this flipflop
Atspeed faults in
this logic cone
cannot be tested
12 SUMMARY
This White Paper discusses at‐speed testing challenges and discusses a solution for facilitating at‐
speed test at the register‐transfer level. The RTL approach is important because designers and test
engineers usually verify the test coverage only at the gate level during the final ATPG stage. This RTL
solution thus saves weeks of effort by fixing potential issues up‐front.
The SpyGlass‐DFT DSM solution is the industry’s first tool which will accelerate design turnaround
times by identifying timing closure issues caused by at‐speed testing – early at RTL. It provides
accurate RTL fault coverage estimation for transition delay testing, together with diagnostics for low
fault coverage early in the design flow, to achieve high test quality with minimum design iterations.
End of Document