Page 23 - Chip Scale Review_September-October_2023-digital

chip within that package. This time and cost lost in the process
itself are substantial, yet they are often overlooked. These
kinds of challenges result in an increased need for advanced
process control throughout chip manufacturing.

360º inspection for classification and reclamation
For 2.5 and 3D multi-die packages, six-side packaged
chip inspection has become essential to ensuring device
integrity. By optimizing the resolution and illumination
of the inspection system, smaller defects can be identified
and classified with increased accuracy. The flexibility of
tuning illumination (angle and color and combinations
thereof) can enable increased contrast for higher detection
and more accurate defect classification. Advanced image
processing algorithms are also used to further differentiate
defects by types.
Achieving automated inspection at high speed also requires
continuous advancements in data processing and high-
performance cameras with large fields of view (FOV). This
provides the ability to inspect more packages simultaneously
or inspect larger packages in one single FOV for greater
efficiency. Advanced inspection capabilities like these
can help avoid escapes, which is the primary concern for
the automotive OEM and Tier 1 supplier where safety and
reliability are concerned. Moreover, enhanced inspection
adds the ability to correctly classify the specific nature of a
defect—i.e., scratch, crack, copper exposure, etc.
Reliable classification ensures that chip manufacturers
can, in multiple cases, rework packages that were first
identified as defective and get them qualified again—another
important yield boost. An inspection system’s precision
ability to distinguish, for example, a fiber from a more
troublesome crack across a wide range of advanced package
types makes this reclamation capability possible. The broader
implications for automotive device integrity and safety speak
for themselves. For the automotive OEM and Tier 1, yield
optimizations like these are likely to be directly reflected
in updated pricing models set by outsourced semiconductor
assembly and test (OSAT) services. Yield improvements
aided by advanced inspection and metrology will likely
make a significant impact on the economics of these pricing
arrangements going forward.

Deep learning enhancements
AI implemented as a deep learning solution for component
inspection can deliver additional gains in sensitivity,
productivity and classification accuracy. Deep learning
technology has already proven instrumental in increasing
packaging yield, and going forward it will continue to
be essential.
Deep learning satisfies a robustness to process variation,
instilling greater stability by enabling the inspection tool to
automatically cope with variations in the devices’ surface
features caused, for example, by changes in raw materials or
surface roughness. Deep-learning models can automatically
adjust for minute process and material changes to keep
inspection tools running without the need for manual tuning
by engineering. This advanced capability contributes to
increased mean time between assists (MTBA) and reduced
tool downtime.

21
Chip Scale Review September • October • 2023 [ChipScaleReview.com] 21