Devices need to do Vmin test to find its minimum working voltage. For IP tests, we usually use the search method to find the Vmin points. It will take lots of test time. In order to reduce test time, designers implement machine learning in the test program. We collect PD(Process Detector) and Vmin data of thousands of devices first. Machine learning could find the relationship between these data, then give the predicted Vmin value of tested devices. We only need to test pass/fail under the predicted voltage, thus test time will be greatly reduced. The machine learning model is provided by the product designer in Python code. What the test engineer need to do is send the required PD data to python function, then get the predicted voltage, do the corresponding IP test. This paper will introduce how to implement this method in the test program. We use PyGXL to send data between the VBT program and the python program.

• How to do Vmin tests with prediction method
• Test flow introduction
• PyGXL usage in this case
• Trouble shooting