Michael Schuldenfrei

The quest for zero defective parts per million (DPPM) in the automotive industry starts with the ability to move beyond looking at process-related data and understand what is going on with the individual product itself. Companies must look at the combination of both process and product data to correctly identify potentially faulty parts. Data can be collected across the supply chain for populated PC boards and electronic systems and can include parametric test data, chip and board genealogy, repair/rework, MES process data and information about RMAs. Collectively, this data is the “voice of the product” and provides a blueprint to identify unique characteristics in devices during the manufacturing test process. When combined with prescriptive and predictive analytics engines, this information enables brand owners to make rapid, decisive actions that can dramatically improve overall automotive electronics quality.

Why Process and MES Data is Not Enough

A Manufacturing Execution System (MES) is a control system for managing and monitoring work-in-process on a factory floor. It receives real-time data from the factory floor, keeping track of all manufacturing information. The goal of an MES system is to improve productivity and reduce manufacturing cycle-time. It’s not designed to manage quality beyond what has been “designed into” the product, and MES data alone is not enough to determine if a product will fail prematurely.

To accurately identify defective or suspect parts, manufacturers must be able to analyze product data that has been collected at every test step from wafer sort to final assembly in order to determine if a part is truly “good”. Semiconductor companies produce massive amounts of product data, generating tens of thousands of parametric test data each day on various chips and devices. This deep parametric data is a key ingredient in advanced analytics when striving for zero DPPM in automotive electronics.

The Analytics Continuum

Gartner categorized the analytics continuum into four categories: descriptive, diagnostic, predictive and prescriptive. Per Gartner, descriptive and diagnostic analytics determine what and why something happened, respectively. Descriptive and diagnostic information can be gathered directly from many existing data sources including MES and process data.

However, the ability to deliver predictive and prescriptive analytics requires the combination of complete manufacturing data and powerful data analytics to enable the decision support systems and automated actions that are necessary to deliver predictive and prescriptive results. For example, within an automotive electronics supply chain, a semiconductor vendor can create a rule that automatically alerts manufacturing operations about a “tester freeze” by identifying when “n” number of consecutive chips produce the same parametric test measurement (highly unlikely in real-world manufacturing). Every one of those consecutive chips that returned the same parametric test value should be a concern from a quality perspective because their results are not accurate. The use of big data analytics can automatically flag these devices and prevent them from being shipped to a Tier 1 automotive supplier until a proper disposition has been made, ensuring greater quality for the downstream electronic devices they go into.

The Voice of the Product

Semiconductor and electronics companies all want to take full advantage of the efficiencies that the Industrial Internet of Things (IIoT) promises to manufacturing organizations.  Therefore, companies that can supply intelligence on machine data and improve overall efficiencies are gaining significant corporate attention. However, many advocates of the IIoT for today’s smart manufacturing are overlooking an important component of data analytics. The “voice of the product”, whether it be chips, devices or boards, is a powerful benefit for brand owners. Achieving the goals of the IIoT, and delivering a common language across the supply chain, will require companies to collect and analyze data not just from one product line or factory floor, but for all product lines and factory floors regardless of where they are located. By applying real-time product analytics to the manufacturing data from an entire supply chain, leveraging deep parametric product data and providing visibility into a product’s genealogy, a powerful new level of manufacturing intelligence is now available to semiconductor companies to drive the next level of product quality.

For example, manufacturing operations are now able to better understand why there is a performance drift or prevent a likely RMA device from being shipped into the supply chain. The ability to do a big data deep dive also contributes to reduced costs, increased profits and overall revenue growth. Other key benefits include improved product consistency, stronger brand protection, faster time-to-market and enhanced product reliability.


The quest for delivering autonomous vehicles is driving new needs and requirements for semiconductor companies supplying to the automotive industry. Automotive companies demand consistent performance and every chip needs to work at its full potential over its expected lifetime. Devices that fail earlier than expected or have too much variation in device performance will simply not be acceptable for inclusion in mission-critical applications such as the autonomous car. It’s more important than ever for automotive companies to protect their brand and they need greater visibility and transparency from all their suppliers on the performance and quality of every sensor, chip and board manufactured for their vehicles. Semiconductor and electronics companies need to analyze more than just process and MES data, they must also analyze product data, because with autonomous cars on the horizon, any defective chip can be deadly.

Michael Schuldenfrei is CTO at Optimal+.