| Test Process Choices
Influence Cost, Throughput and ATE Design |
|
"Batch
handling [of CSPs] will drive the manufacturer toward parallel
test even more strongly than in the past . . ."
The introduction of chip-scale packaging
in high volume requires a new mind-set when addressing the
test process steps.
Because of the characteristics of the
CSP manufacturing process, classic IC test requires some revision.
Paying careful attention to what testing is done at which
step of the process will pay large dividends in the cost of
the process.
|
 |
|
By
Paul M. Sakamoto
Contributing Editor
|
First, let's look at an extremely simplified IC test flow.
After a wafer is processed through the fab, it is typically tested
to a relatively loose set of parameters at wafer sort test.
This testing will usually relax most of the
AC parameters, such as frequency and propagation delay, while attempting
to test most DC characteristics thoroughly.
The next test step is post-assembly package
test. This test is usually the one that stresses AC parameters,
as well as any functionality that was difficult to test at wafer
sort.
This overly simple test flow has already lost
favor in the commodity memory market, due to the fact that it doesn't
maximize the package test yield.
Why is this important in the DRAM manufacturing
segment, where the package cost is fairly minimal?
There are two reasons: First, DRAM makers are
concerned about those few pennies wrapped-up in the failed packaged
units, due to the extreme level of price/cost competition in the
memory marketplace.
The second reason is that memories are tested
in parallel at all levels of testing. Parallel testing requires
that the yield be as high as possible, and that the speed or other
performance binning be as consistent as possible to ensure efficiency.
That issue may alter the way a CSP test process
is designed, especially if the devices are non-memory ICs.
CSP assembly is particularly suited to handling
the devices in a strip, matrix or panel of devices.
This panel is a ready-made carrier to move multiple
devices through a parallel test system-handler combination.
Batch handling will drive the manufacturer toward
parallel test even more strongly than in the past, since it may
be possible to move 50-100 devices in parallel.
As a result of this efficiency, the average
CSP device will now have the same yield and efficiency issues that
have been driving DRAM testing for some time.
Correlation
Achieving a high level of correlation from wafer sort to
packaged test may require some manufacturers to add at-temperature
test to their wafer sort equipment.
At-temperature testing requires wafer probers
which are equipped with special "hot and cold" chucks or wafer stages.
Additionally, much greater characterization
work may be needed to ensure that the relationship between fab process
parameters and the final performance of the devices is well understood
and guard-banded.
Probe
Cards
Special high-performance probe cards may be needed to achieve
full performance testing at speed and with full specs for analog
signals at wafer test.
This strategy implies that the high-performance
test equipment will be used at wafer sort, and that a lower cost,
lower performance ATE choice may be made for final test.
In fact, the first drive to ATE that fulfills
this niche has already begun.
I expect that ATE capable of 3,000 or more channels
with the capability to efficiently test >64 devices per station
at $10K cost/site is going to arrive within the next year.
These new machines will offer full DC and minimal
AC specifications. It will be interesting to see how this affects
the overall ATE marketplace.
|
Mr. Sakamoto is vice president of the
Memory Products Division at Credence Systems Corp., Fremont,
Calif. Contact him at paul_sakamoto@credence.com.
|
|