Testing of semiconductor chips is an expensive and time-consuming operation. Each chip must go through several rounds of complex test suites comprising of thousands of tests. Significant efficiencies can be gained if one can differentiate between good and anomalous chips earlier in the testing cycle


The team analyzed a dataset comprising of almost a million chips and over thirteen-thousand independent tests. They built several independent ML models, each designed to differentiate between anomalous vs. normal test patterns. These models were then combined to select chips that could be excluded from further testing.


  • Electronics

Tools Used

  • Data Science Experience (DSX)

Data Science Techniques

  • Supervised and unsupervised anomaly detection


  • Savings in time and cost of semiconductor testing
  • Accelerated market launch time for new products

Need help with a similar case?

Contact us to find out how we can help!