Issue link: https://iconnect007.uberflip.com/i/1539960
OCTOBER 2025 I SMT007 MAGAZINE 65 variables, including proximity to the limits. The AI model has been implemented for multiple manufac- turers worldwide and is found to have 92% accu- racy on the real production line data. The practical benefits can be understood through a case study. The model was implemented for a manufacturer where the test engineers did not have visibility into the offshore supply chain. They faced issues with high board retest rates and low first- pass yield, coupled with many failures in ICT pins. Once the AI model was implemented, test engi- neers were able to quickly gain visibility into the test names with high failure rates. The specific test name can be further broken down to equipment and fixture to identify which spe- cific equipment is creating the problem. The anom- aly detection model highlighted that the issue was observed only on one fixture (anomalies are high- lighted as pink dots on the scatterplot). Upon inves- tigating the problematic fixture, it was identified that the solder paste used in that fixture had been changed from a foreign brand to a local brand. The local brand solder paste had 4% higher flux content, causing a build-up on probes. Rectifying this issue resulted in a 4% increase in first pass yield and reduced retest time by 13 hours. Manually debugging this issue would have taken multiple days, leading to production loss and con- suming crucial manpower. Use Case#2: Degradation in equipment performance One of the most common issues faced in manufac- turing is the performance degradation of equipment, which impacts measurement accuracy and leads to potential failures. Predictive AI models can solve this issue by understanding the degrading behavior and providing early alerts before failures occur. During the ICT testing phase, one common equip- ment degradation issue is related to the probes. They get damaged or undergo wear-and-tear over time, impacting accuracy and increasing false fail- ures, thereby raising retest rates. Manually identi- M e a s u re m e nt va l u e s d e g ra d i n g f ro m t h e n o r m a l b e h av i o r (u pwa rd s i n t h i s c a s e) a n d c l o s u re to t h e fa i l u re zo n e. ▼ D e g ra d at i o n a n o m a l i e s d ete cte d by t h e A I m o d e l a n d g u i d e d to t h e p ro b e o n t h e b o a rd . ▼