Deloitte Supplement - January 6, 2020 - D18

Industry 4.0 in automotive

Pursuing "Zero Defects" in
automotive electronics
By Oreste Donzella and Janay Camp, KLA Corporation
Innovations in electrification, connectivity, and autonomous driving capabilities are driving a revolution
in automotive electronics-and in the semiconductor components at their core. From a few hundred
semiconductor devices a decade ago, a modern vehicle may now contain up to 8,000 semiconductor ICs
(integrated circuits)1 integrated into complex electronics subsystems that can represent 20-30 percent,
or even more, of the vehicle's cost.2
Because electronics are currently the number one failure item in cars less than three years old,3
improving their reliability can help car manufacturers control warranty costs and avoid expensive
liabilities. Semiconductor ICs are also essential to safety-critical features such as advanced driver
assistance systems (ADAS). Hence, car manufacturers and Tier 1 suppliers are raising the bar for their
electronics' quality and reliability, requiring parts-per-billion (ppb) failure rates and pushing the "Zero
Defects" concept into their electronics supply chains.
Critical to improving electronic systems' reliability is to eliminate latent reliability defects in their
semiconductor components. Latent reliability defects are defects that, because of their size or location
on a chip, do not cause an electrical failure when the chip is tested. As a result, the at-risk chip passes
the electrical test and "escapes" into the supply chain. Once the chip is installed, latent reliability defects
can be "activated" under conditions considered extreme for semiconductor devices, such as intense cold,
high heat, high humidity, and vibration-all of which frequently arise during normal vehicle operation.
Detecting latent reliability defects thus requires new approaches that do not rely solely on traditional
electrical testing methods. Automotive semiconductor manufacturers have traditionally used statistical
analyses of electrical test data to detect potential outlier chips that pass with marginal performance.4
This method is often reinforced with screening methods that inspect every chip being manufactured
at several critical points in the process. A limitation of such comprehensive screening, however, is that
not all defects detected may affect real-world reliability. A conservative approach to removing these
defects can result in an unacceptable level of "overkill"-the removal of chips that would have functioned
normally throughout their specified life.
Machine learning-based methods can help more accurately identify which defects may actually
cause failures. The Inline Defect Part Average Testing (I-PAT™) method, for instance, leverages inline
defect inspection, historical reliability knowledge, and data from electrical wafer sorting, burn-in, final
tests, and field returns to develop a neural network model that can minimize potential "escapes" and
reduce overkill.5 Multiple semiconductor fabs are evaluating the I-PAT™ methodology, with promising
preliminary results. Going forward, this and similar techniques will be pivotal in helping semiconductor
manufacturers achieve the quality levels needed for safe, reliable vehicle operation.

18
Deloitte Supplement - January 6, 2020

Table of Contents for the Digital Edition of Deloitte Supplement - January 6, 2020
