How to Identify Dark Yield

Back in 2018, Bloomberg published a story alleging that Chinese spies planted clandestine chips onto circuit boards destined for the services of major U.S. companies. Apple and Amazon categorically denied the allegations. The plausibility of such an attack vector against any company led me to contribute a story to Forbes on how implanting chips on circuit boards was both easy to do and incredibly difficult to detect. The vulnerability that was highlighted is another flavor of a problem that has plagued many brands and their manufacturing partners for years: dark yield.

Dark yield refers to products that escape factory quality assurance processes with defects or unwanted alterations and are shipped to customers. Dark yield is incredibly prevalent, but by definition, impossible to measure. Once you start improving your ability to capture dark yield defects (usually by improving manufacturing traceability), those captured units are now yield fallout, which can be analyzed to determine a permanent corrective action. Whether it’s clandestine chips, counterfeit parts, or simple quality escapes, dark yield has a very real bottom line impact. It results in unhappy customers, bad Amazon reviews that can affect future sales, expensive warranty returns, and even people getting hurt or killed.

At Instrumental, we’re experts in shining light on dark yield. Instead of looking for secondary metrics to measure (Amazon review scores, total warranty returns, customer care calls), what if we just find those defects in the first place? Two core innovations enable Instrumental technology to discover and identify these defects.

Uncovering Hidden Defects in Production

The first was the somewhat obvious realization that you cannot find something you aren’t looking for – so our very first offering was the ability to collect a 100% digital traceability record of everything that is built. We chose to implement our track and trace manufacturing solution with images, knowing that many dark yield issues have physical evidence, but that some do not. This traceability record enabled our customers to get unprecedented views of their product and effectively, their assembly process, which enabled them to identify more issues, faster. Our average customer finds 20 issues in development that they claim they would have never found without Instrumental, including things like dented speaker drivers, scratched sealing surfaces, and unmated connectors. Any issue not found in development automatically becomes a dark yield issue in production.

Identifying Dark Yield with AI

The second innovation we needed was to design a way to find anomalies in the data record, when we don’t know in advance what we need to find. Most conventional vision systems require specific rules and training with a bunch of golden and defect samples before they start to be useful. But for dark yield issues, you don’t know what they are, so you don’t have golden or defect samples. Instrumental uses machine learning in manufacturing to learn what is normal variation from typical units, and when high levels of variation are found, we can highlight those units as anomalous. As it is today, the technology is incredibly powerful – it can look through hundreds of units in seconds to pull out those that are different. If those differences are useful, our customers can setup a test to catch differences like those, whether it’s a specific defect type (like a missing screw) or a range of variation (screws at various levels of proudness). One top electronics brand used Instrumental anomaly detection to discover that their PCBs were fraying, while another found that copper vias were malformed – both disqualified upstream suppliers who were shipping parts that did not meet their specifications. FLIR used our anomaly detection tools to find that some of their units had screws that weren’t fully seated, and stopped those units before they became dark yield. FLIR engineers were then able to monitor for that issue over time, to make sure it doesn’t come back. Today, we can do that monitoring for our customers. Once a test has been validated to be passing and failing units as intended, our customers can run them live on the production line.

The very same tools that enable you to identify and to eliminate dark yield can also be used to improve conventional yield and time to market. Today’s functional testing environments are quite sophisticated, enabling teams to catch the majority of defects before they become dark yield. Once found, however, the process to dissect those units, discover the root cause, and validate a corrective action can be laborious and long. 100% traceability can be leveraged to get quick answers about how affected units were assembled, oftentimes leading to clues that either eliminate paths of inquiry or directly indicate the root cause. A leading electronics brand wondered if their speaker sound issues were coming from damaged drivers, and seconds later were able to see inside the specific units with the issue and validate that the drivers looked good. This eliminated a wild goose chase and let the team remain focused on other potential root causes. A top cell phone brand noticed unusual camera performance in their functional test stations, and when the engineer reviewed the data record, he saw abnormalities in the camera lenses. Using Instrumental algorithms, in a few clicks he found more and was able to trace it back to a specific camera vendor. The whole process took minutes, instead of days or weeks.

Reducing dark yield, improving yield, and enabling faster time to market by leveraging a 100% traceability record is our core business. Greater oversight and auditability of manufacturing processes reduces the likelihood of unintentional defects or alterations from leaving the factory – and increases a company’s ability to respond quickly if a concern is discovered.