Page 13 - ChipScale_Sep-Oct_2020-digital

Anatomy of field returns (RF) components in specialty fabs (GaAs, quality level of this stage reduced to the bare
The International Automotive Task Force GaN, SiGe, and so forth) with a lower than minimum, which ensures that the outgoing
(IATF) standards essentially require us to average level of automation and detectability. product does not receive the full benefit of
strive for zero defects. If we look at where Second, propulsion is an increasingly large detection at electrical test. This translates
the industry performs today, data suggests group with the advent of electric vehicles (EV) to most customer failures being test-related
that the automotive supply chain resides at and can be anywhere from 180nm to 32nm. issues and increased costs of non-quality
approximately one defective parts per million This bucket includes power components and further results in the motivation to move
(PPM) or above. When reviewing field and engine control units (ECUs). These towards a holistic approach as a matter of
returns that constitute either warranty returns items would be considered mission-critical progress and safety.
or zero-kilometer failures (Figure 2), the field components in a way, requiring a higher level
of reliability. Finally, Framing the issue
infotainment does If we are to consider a strategy to move the
not warrant the same needle closer to a zero-defect concept, then
level of reliability and several things must change from the way
has some flexibility in facilities operate today. Most of these legacy
terms of sampling the facilities are semi-automated or manual.
latest and greatest fab This means that they fundamentally have a
nodes. multitude of point solutions and disciplines
Given the wide to govern their quality standards. Simply put,
range of fab nodes they neither capture all the data needed to
that are sampled in govern their processes effectively, nor analyze
the automotive supply this data in a manner that allows for speedy
chain, we are subject and effective feedback. Based on field returns
to a variety of available and internal failures, we have seen how far this
quality levels. Reducing strategy can take us. It is unlikely that we will
cost per component be able to breach the 1PPM barrier consistently
to maximize profit without rethinking the process entirely.
also means reducing To paraphrase the problem, we have an
Figure 2: Fab and packaging contributions to field returns. the acceptable quality inability to scale our detection, coupled
level to an acceptable with the inability to assure our test coverage
estimates suggest that roughly 25% of the bare minimum. Every compounded by an inability to assign root
failures come from the front end fab. Within detection step is a non-value-added step, cause with total certainty in too many of
this data set, 50% of failures that leave the and hence the cost is passed off further our cases. All these problems result from a
facility essentially have a parametric test, downstream. This means that even with patchwork approach to quality and a siloed
but somehow elude our ability to detect the the best fabs in the business, PPM-level perspective on quality data management.
problem. Another 30% have no test coverage detection is neither offered, nor discussed A holistic approach would streamline
and therefore, no detection. Another 15% for parts that are in volume production. This information sharing and facilitate first-time-
are undefined, meaning we can’t assign situation deteriorates further as older nodes right decision-making (Figure 3). So why
the failure to any specific cause. This third or specialty technologies are used. haven’t systems evolved to be holistic?
category is essentially the unknown. In these The best chance for detecting defects on
cases, no real corrective action is deployed a part is with electrical testing, which entails Adopting the needed changes
and the gap in detection persists. And finally, wafer probe or final automated test equipment The fundamental flaw with point solutions
approximately 5% of the failures reflect (ATE) in packaged form. Conceptually, if is that they don’t address the challenging
disagreements within the supply chain as to the part has been characterized thoroughly issues observed at the fab level, with
the origin of the failure. and built on a known technology with a low packaging and surface mount lines. The
defect rate, then any remaining PPM-level solution to this problem originates from a
What eludes the supply chain failures can be captured at the electrical test holistic approach. Holistic in this case refers to
The constituents of the current supply step. Characterization depends on: 1) the the ability to look at the entire manufacturing
chain – fab, packaging, and electrical test – design failure mode and effects analysis line. Beyond the single facility, holistic means
(DFMEA) being able to simulate all failure
can be evaluated as detection gates, with an the entire supply chain as a single entity.
increasing granularity of detection, but also modes; 2) product engineers being able to Moreover, we must adopt new principles of
in the order of increasing cost. test the parts to cover all customer mission signal detection to successfully manage chart
From a wafer fab perspective, automotive profiles; and 3) with increasing software scaling. Most systems today are predicated on
components can be broken out into the components in the parts, ensuring that data the idea that if a measurement chart is set up
following buckets: safety/advanced driver fidelity is maintained throughout. properly for a known parameter, it will detect
assistance systems (ADAS), propulsion, and Most parts do not have wafer-level anomalies accordingly. While in principle
infotainment. First, the safety/ADAS bucket traceability, which further degrades the this is true, it becomes a daunting problem
ability to tie back failures to fab processing
includes sensors (microelectromechanical to manage 100,000 charts in a single facility
systems [MEMS], optical, temperature) that as warranted. In the face of meeting ultra- with a skeleton crew. Each measurement value
are made on larger nodes, or radio-frequency aggressive customer timelines and cost irrespective of control and specification limit
pressures, we again see the acceptable

Chip Scale Review September • October • 2020 [ChipScaleReview.com] 11 11