Page 34 - ChipScale_May-June_2020-digital

features. It is no longer possible to co-
optimize all the different parts at <10nm
geometries, so the parts are splitting.
In the case of multiple components,
errors add—or more precisely, success
rates multiply—as in 10 components
10
with 99% success have a 0.99 or 90.4%
success rate).
Keeping the above discussion in mind,
there is an additional wrinkle: the increase Table 1: Cost of failures for a phone component.
in chip count also increases the demands
on the bandwidth between the different
components. With more bandwidth,
there is a greater need for connection
density. So even with a slowdown in the
typical Moore’s Law implementation,
package components will add connection
density, and will therefore continue to
reduce pitch, likely at an accelerated
rate. An extended roadmap is shown in
Figure 2. Pitch continues to decrease, and
the supported bandwidth continues to double Table 2: Cost of failures for an automotive controller.
roughly every two years. The International
Technology Roadmap for Semiconductors consider in light of the above: first, that simple phone discussed above. In this
(ITRS) data is fit to the roadmap, but the cost to the end device manufacturer case, we have a more expensive unit, but
linearized. This works out to an ~8% per is much, much higher than a .01% lower volumes. There are about 70M cars
year pitch decrease for the small scale, failure of a $5 part would indicate. And sold worldwide, and if we assume 1/7th
and ~4% for the larger bumps. In addition, that is a core lesson of reliability: the use these electronics, then we can make
the number of integrated components will cost is borne by the final manufacturer the calculations shown in Table 2.
increase significantly. or ultimately, the consumer. Second Similar to the phone “case,” the cost to
is that at high volumes, extensive the IC manufacturer of low yield is fairly
Case studies inspection even for fairly low cost items trivial (keep in mind that the manufacturer
To put the above discussion in practical pays for itself. will make about $1B in revenue). The cost
terms, let’s look at a pair of simple cases. These costs also ignore the cost of of repair in this case may well be higher at
A phone. Consider a simple device failure analysis on the producer side, as the assembly level—an ECU replacement
on a phone, with an ~100M per year rate well as goodwill and reputation. These is perhaps a $1000 fix, and maybe more.
(~20% market share) and a $5 cost. So we may well outweigh the direct costs. On That means that for a 500ppm failure rate,
can assume the following components: the production side, the critical benefit the replacement cost is $5M annually.
a sound chip, an amplifier, some modem of inspection is typically in product If the car fails, this is a $20-100M cost.
components, and some Bluetooth device. qualification and test. For example, Across the industry, a 100ppm defect rate
The board for this is approximately $200, months could be saved from qualification on critical components still gives rise to
from TechInsights and IHS Markit reports. time, saving (in this “toy” model) ~$10M an ~$10-$150M in costs. Considering the
Table 1 shows the cost of fails, where we per week. vast increase in electronic components
look at failure rates in parts per million An automotive electronic control projected in the next decade because of
(ppm). The failure rate is the rate for a unit (ECU). An automotive ECU is a new automated driver assistance programs,
single component. more complex device compared to the major improvements are needed.
A rate of 100ppm is a .01% fail rate.
And it costs about $10M per year in
returns at the assembly (phone) level. I
calculate a logic board cost for the phone
of ~$260, and I rounded up the entire
assembly to $1000. For a basic 100ppm
failure rate, that is $2.6M per year in
costs just to replace the board, if the
design allows that. And that is ignoring
return costs and failure analysis (FA)
costs. An improvement from 500 to
100ppm defect rate would save $10M per
year at the integration level.
There are two notable aspects to
Figure 3: Simplified process flow for 100% inspection, optimized etest, and 100% post-attach inspection.

32
32 Chip Scale Review May • June • 2020 [ChipScaleReview.com]