CLASSIFICATION OF MICROCONTROLLER INTEGRATED CIRCUIT ON THE POCKET OF JEDEC TRAY USING CONVOLUTIONAL NEURAL NETWORK IN EMBEDDED MACHINE LEARNING SYSTEM

One serious issue in the microcontroller manufacturing environment is the mixing of microcontroller unit (MCU) parts, leading to the wastage of materials, dissatisfied customers, and the implementation of non-value-adding activities to address it.More adverse effects include negative feedback from customers, loss of confidence, and impact on business growth.One root cause traces back to the final testing of the manufacturing back-end process when reusing unemptied standard JEDEC matrix trays for good and bad units in the test handler.Currently, emptying the JEDEC matrix tray and inspecting it is a manual process prone to human error due to high-volume test runs, small package sizes, and package colors matching the tray. The feasible solution is developing a Machine Vision System (MVS) to automate MCU Integrated Circuit (IC) tray inspection and detect empty trays. This study proposes an MVS classifying MCU IC, JEDEC matrix tray pocket, and background using a Convolutional Neural Network (CNN) deployed inOpenMV H7 Plus, a low-power, memoryconstrained Embedded Machine Learning System (EMLS).The CNN achieves 94.8% training accuracy and 86.67% testing accuracy.Deployed in OpenMV H7 Plus, it classifies 128-pin, 64-pin, and 48-pin TQFP MCU IC packages, tray pockets, and background in real time.Results show an EMLS-based MVS can address MCU IC mixing issues.However, future work should collect more data and explore additional feature extraction and data augmentation techniques to enhance CNN accuracy further.