Original component manufacturers and original equipment manufacturers should be on the same page when it comes to quality. In fact, according to Michael Schuldenfrei, CTO at Optimal+, the exponential growth of semiconductor components in mission-critical electronic devices is introducing a need for new levels of quality and reliability.
Unfortunately, improving quality can be expensive and time-consuming because companies across the supply chain are often reluctant to share data in the interest of protecting IP, and that needs to change. “Testing in silos across the electronics value chain is no longer a viable option,” said Schuldenfrei in a recent phone interview. The good news, he added, is that Optimal+ has analyzed more than enough customer data to conclude that there are proof points that clearly indicate that data sharing creates win-win situations for both suppliers and customers.
Data sharing, he said, can lower RMA costs through board-to-chip correlations that can speed root-cause analysis, can reduce NTF rates, and can facilitate proactive targeted recalls. It can also improve time to quality by reducing the time needed to reach board-level DPPM goals through escape prevention and outlier detection, and it can support enhanced functional safety (in accordance with ISO 26262 in the automotive industry, for example).
Further, data sharing can improve system performance, he said, by supporting smart pairing—selecting the right chip for the right board. Finally, data sharing can enable adaptive test—testing “suspect” parts more and “perfect” parts less.
The data-sharing challenges, Schuldenfrei said, center on data-flow, data-management, and integration concerns as well as IP protection and the possible perception of a “customer wins, supplier loses” situation. In addition, each member of the supply chain may lack the cross-domain expertise necessary to analyze the shared data.
The Optimal+ solution is what the company is now calling a “Shared Quality Network,” or SQN, based on a trusted third-party SQN hub, which handles implementation and application details and maintains traceability. With the SQN approach, the hub provides the analytics expertise necessary to identify issues, and only data relevant to resolving identified issues is shared. Then, each supply chain member leverages its own expertise to resolve issues (Figure 1).
Courtesy of Optimal+
Even shared data can be obfuscated, Schuldenfrei emphasized. The SQN hub might identify an issue with an OCM’s IDDQ test, for example, but the shared data might simply refer to an issue with regard to “Test A.” SQN data sources, he said, can range from full test data to partial genealogy information at the IC level and include parametric test data and MES information at the board level. Finally, data from SMT/pick-and-place machines can establish a linkage between component and board.
Schuldenfrei outlined how SQN might work. Given a specific failure mode at the board level, apply machine-learning techniques to look for correlations in test data from the relevant ICs. Significant correlation would suggest that additional ICs with similar test characteristics should be eliminated from the board-assembly process. Lack of correlation would suggest that the failures are not related to the ICs (or that the IC test is insufficient) and may lie in the board assembly and test process.
Schuldenfrei cited the automotive industry as an obvious use case where SQN can add value. He noted that semiconductors in cars are increasing exponentially, not just in autonomous vehicles but in sensor-laden connected cars as well.
The consequences for quality are clear. According to the J.D. Power 2017 Vehicle Dependability Study, “Continuing increases in technology-related problems have contributed to dependability worsening in the industry for a second consecutive year…. The audio/communication/entertainment/navigation (ACEN) category continues to be the most problematic area, accounting for 22% of all problems reported—up from 20% last year.”
Schuldenfrei cited Audi estimates that if a premium car has 7,000 semiconductor devices with 1-ppm failure rates, the company would experience seven failures per 1,000 cars. For a production rate of 4,000 cars per day, that would translate to one failure per hour. In contrast, the industry is pursuing quality targets of zero failures at 0 km and field failures of less than 10 ppm/year.
Automakers can’t solve the quality-related problems themselves. “There’s no magic bullet,” he said. “Customers need multiple methods to reduce quality issues, and we recommend sharing data across the supply chain.”
He added that SQN can support data sharing not only between semiconductor makers and their customers but among tier 2 and tier 1 suppliers and the automakers, and so on.
Optimal+ highlighted the benefits of SQN at SEMICON West and the co-located Test Vision 2020 in San Francisco in July and will make its case to the automotive industry at the Automotive Testing Expo October 24-26 in Novi, MI.
