A Kelvin contacting system commonly is used on devices that require
precise measurements such as power management devices, precision op
amps, and high bit-count ADCs or DACs. Because most Kelvin solutions
have higher acquisition costs, they are only used when other test
methods are not appropriate. In fact, many Kelvin test systems also can
be less reliable and require a more difficult load board layout.
However, Kelvin-ready technology overcomes these obstacles through its
highly configurable, optimized design.
In developing a Kelvin-ready contacting system, the goal was to
design a robust system that improved on existing Kelvin test sockets or
contactors. Based on industry interviews, here are the most important
* Force and sense contacts must hit device I/O.
* Load board layout does not require tight tolerances.
* Force line can handle high current.
* Can add, move, or remove Kelvin measurement capabilities easily,
onsite, and for any given I/O.
* Can work for small devices and small pads and leads.
* Can use Kelvin measurements even for RF devices.
* Must be standardized and cost-effective system.
* Must have reduced maintenance, long contact life, and reusable
* Contacts should be easy to replace.
[FIGURE 1 OMITTED]
The variances in package dimensions require that the force contact
be designed to hit the center of the device pad. To ensure that the
sense contact also will connect with the device pad or lead, it has two
tines, one on each side of the force contact tip. By keeping the force
contact a solid one-piece design, it can have reliable low resistance
and handle greater current with than multipiece contacts. Monte Carlo
analysis of the design, including the manufacturing tolerances, shows
that the force and sense contacts do not have the potential for shorting
or missing the device pad and creating an open condition.
The tight routing and subsequent shorting problems often associated
with Kelvin solutions were addressed by using two different contact
technologies. This is actually an optimized solution since the force and
sense lines have different performance criteria to best measure the true
parameters of the DUT. For both the non-leaded pad and leaded
Kelvin-ready solutions, the gap between load board pads for the force
and sense lines is greater than 5 mm, which leaves plenty of room to
route signals. A 4-wire Kelvin connection schematic is shown in Figure
In a Kelvin system, the sense line is monitoring the voltage at the
DUT and connected to a voltmeter with a 10-MQ to 100-MQ input resistance
so only picoamps pass through the sense contact. Any sense resistance
basically has no effect on the sense measurement because it is in series
with and small in comparison to the voltmeter's resistance.
If a Kelvin measurement is not needed when using Kelvin-ready, the
sense contact and the sense line load board pad can be removed for that
device I/O. This allows RF or highspeed signals to be routed straight
from the device to RF connectors for optimized performance. For RF
testing, this also removes the stub, and the normal Kelvin-ready force
contact now can test devices with frequencies above 20 GHz.
The feedback from the sense line is used by the tester's pin
electronics' parametric measurement unit to increase or decrease
the voltage at the pad or lead of the DUT. This can be very important in
improving yields of RF devices where the RF output power fluctuates
depending on the DC power. If the DC input power is always known at the
device pad or lead, an easy setup when using Kelvin-ready, all failed
devices are truly failed devices and do not require retesting.
Some devices require both RF and Kelvin testing to adequately test
RF and precision analog chips within the same package. In addition,
there is a focus on making devices small with larger numbers of I/Os,
which in turn leads to devices with smaller pads and finer pitches.
To be reliable, a Kelvin system must ensure that the sense contact
always touches the device pad, so there needs to be some built-in
redundancy in the sense contact. With this, you are assured of making
contact to the device pad no matter how the device is aligned in the
contactor or socket.
The resulting error in measurement for the non-Kelvin case in
Figure 2 is dependent on the contact resistance of the force contact.
The error can be large if the device resistance is very low or if the
contact resistance of the force contact is fairly high.
On the other hand, Figure 3 shows that, when using Kelvin, the
resulting error is not dependent on the contact resistance of the force
contact and only slightly dependent on the sense resistance. Because of
these factors, a Kelvin test system can make more precise measurements,
allowing test limits to be tighter and lot-to-lot variations more easily
Both the pad and leaded Kelvin-ready schemes use a modified, rigid,
one-piece contactor (in this case, the Johnstech ROL [TM] 200). The gap
between the force and sense contacts was optimized by finite element
analysis (FEA) to maximize the sense contact life. Monte Carlo analysis
with manufacturing tolerances was used to prevent shorting.
The force contact has an existing rigid design to provide the best
current-carrying capability, highest RF bandwidth, and the best
self-cleaning wipe function to break through oxide layers present on
many device pads. The flexible sense contact design features a smaller
wipe function with a more pointed contact probe for accurate voltage
measurements. Also, it routes sense signals away from the force signal
to facilitate easier board layout and decoupling part placement.
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
The design uses the alignment plate, or cover, to keep the sense
contacts in place. Removing the alignment plate results in access to the
sense contacts, making it easy to replace them.
To show the relationship between the sense and force contact,
Figure 4 depicts an isometric CAD view along with a high-frequency
structure simu lator (HFSS) top view of the Kelvin-ready contactor
without the housing and alignment plate. The bottom of the figure
illustrates the side view of the Kelvin-ready system showing the
housing, cover for handlers that have precise placement capabilities,
and the alignment plate.
The sense contact color was changed to red in the upper right HFSS
view to better show contrast to the force contact at the device
interface in the figure. The HFSS side view on the bottom depicts both
contacts: in normal test mode and in the compressed state. The force
contact's self-cleaning wipe function is left to right, and the
sense contact's wipe function is up and down.
These two different wiping motions tend to move away any debris
that might form, such as tin oxide from the device pad or molding debris
from sawed packages. Both mechanical FEA simulations and measured data
from extensive lab experiments and customer beta site testing showed
that the life of the product was extended. Sense contacts routinely
lasted more than 1,000,000 insertions, even when contacting devices
using NiPdAu plating, which is much harder than the matte tin plating
found on less expensive packages.
Kelvin measurements aren't always needed, so for normal RF,
analog, or mixed-signal devices only the ROL [TM]200 force contact needs
to be installed into the housing. This handles any scenario up to 20
If precision measurements are required or a feedback loop is needed
to reduce false failures, the Kelvin-ready sense contacts then could be
installed on one, a few, or all of the signal lines.
Not all device I/Os require measuring very small voltages and
currents. With a configurable and reusable Kelvin-ready housing, the
customer only needs to change the sense contact configuration and
replace the load board with the one that corresponds to the DUT. If the
type of handler is known when the test contactors are ordered, the
design can accommodate the same contactors in production. This reduces
the need for procuring extra contactors for low-volume production runs.
Load Board Simplification
On a load board set up to do testing on a handler for both Kelvin
and a differential RF signal, the low-frequency lines and Kelvin-ready
sense lines are routed to internal layers separated from the top-layer
RF signals by a ground plane. The RF differential lines run parallel to
each other and are routed directly to two connectors on the edge of the
Adjacent ground traces on either side of the differential pair use
vias to provide a high isolation fence to the ground plane to enhance
signal integrity and shield adjacent lower frequency lines. If
decoupling parts were needed. they would be placed between the force and
sense pads, their bodies protruding into the decoupling pocket shown in
Increased Mean Time Between Assists
Extra switches can be added to the Kelvin circuit to help determine
if cleaning or maintenance is required. When a device fails, the
switches can provide a feedback path that is measured and compared to
some baseline resistance. If the value is above the path resistance
limit, the contacts are dirty, debris is present in the contactor, or
the contactor is broken and causing a false failure. If the loop-back
circuit resistance is lower than the feedback limit, then the part is
determined to be bad. If the loop-back circuit resistance is higher than
the feedback limit, the part could still be bad, but the contactor
definitely needs to be cleaned before the device can be remeasured.
Because many packaged devices are completely tested in a fraction
of a second, the force and sense contact resistances are not tested for
every device. Many times, they are only checked if the device fails to
determine if cleaning is needed. Using this checking method helps reduce
false failures. In many cases, it also eliminates retesting parts and
determines when maintenance is needed on the contactor.
[FIGURE 4 OMITTED]
A new material promises even longer life for the force contacts.
Data taken for both soft matte tin and harder NiPdAu device leads
supports a life expectancy of at least 500,000 test insertions for the
force contact. Because the sense contact has a smaller wipe function,
other test data has shown similar results to beyond 1,000,000
insertions. Actual results depend on the insertion speed of the device
into the contactor. Results have shown life expectancy can be greatly
exceeded if the insertion speed is lowered.
Because NiPdAu is a hard material and does not contain any oxides,
a low-force elastomer can be used to help extend the contact's
life. For NiPdAu testing, the average contact resistance is very flat,
and its standard deviation is less than 5 m [ohm] for the entire test.
Reducing False Failures
Checking the Kelvin circuitry every time a failure occurs
eliminates the need for devices to be rechecked. Since the sense contact
is only measuring a voltage at the device and you know the input
current, you can maintain a stable DC power into the device. With a
Kelvin feedback system, you can tell which parts pass for a given
constant input. This also may support up-binning of devices with better
performance, allowing them to be sold at a higher price.
On a rigid, one-piece contact, the data on one force/sense pair has
much less variability than a similar spring-pin pair, which has three to
five moving parts that may or may not make contact with each other at
the same locations. For that reason, it might be advantageous to use
Kelvin on the voltage lines and calibrate the data for the rest of the
connections to the package.
With the device still in the contactor, the loop-back can determine
if either the force or sense contact is not connecting to the device
pad. If so, an open will be measured by the ohmmeter. A short between
the force and sense can only be measured by running a signal through the
feedback network without a device present. If a short is measured, the
force and sense are shorted together. If either a short or open occurs,
the testing should be stopped and the contactor serviced.
For a device sufficiently sensitive to contact resistance, big
improvements in yield can be attained using a self-cleaning Kelvin-ready
contacting system. In an actual non-Kelvin application involving an
automotive voltage monitoring device, the contactor had to be cleaned
more often, resulting in yields slightly above the 90% limit the
For this device, a yield falling below 90% triggered a cleaning
interval or investigation into the type of failures. Because there were
no false failures when the same parts were subjected to Kelvin testing,
there was no need to retest parts that failed at ambient. This reduced
the number of parts to be tested at cold and consequently increased
The same contactor was used for both ambient and cold testing, the
only difference being the handler was soaked at -40 [degrees] C with the
parts prior to testing. Each ambient or cold lot consisted of 8,000 to
10,000 packaged devices, almost 400,000 test insertions total. The yield
at ambient was 95.98% and 97.46% for cold testing.
Most Kelvin applications have performance requirements for a small
bandwidth and typically don't need to process frequencies above 1
GHz. This allows Kelvin to be installed on non-RF lines in an RF device
to provide feedback to improve on MTB A cycle times and reduce tester
down times. This is important for customers who want to supply their
devices with exactly the same DC power every time so variations in test
data are only due to variations in the devices.
Without the sense contact attached, the -1-dB insertion loss
bandwidth is about 33 GHz, and -3-dB bandwidth is beyond the 40-GHz
upper test limit of the network analyzer used to measure the contact.
Even with the sense contact attached and properly terminated, the -1-dB
bandwidth for Kelvin testing is 3 GHz, more than sufficient for a Kelvin
connection. A ground-signal-ground configuration is the preferred layout
for high-frequency lines and signals that require minimal crosstalk
between adjacent signals.
Contactor reusability is extremely important for reducing lead
times and test costs. The Kelvin-ready solution offers the adaptability
needed for today's continuously changing test environments and
provides the flexibility to accommodate future test needs. Using Kelvin
techniques, customers can easily identify higher performing devices,
increase the time between maintenance and cleaning cycles, and
accomplish a variety of other test objectives.
by Jeff Sherry, P.E., Johnsfech Infernafional
About the Author
Jeff Sherry, P.E., is a senior RF/high speed digital R&D
engineer at Johnstech International with 27 years semiconductor industry
experience. He received M.S.E.E. and M.B.A. degrees and has presented
numerous papers at BiTS, ITC, and SEMI events and conferences. Johnstech
International 612-378-2020, firstname.lastname@example.org