ABSTRACT
In this work, we present an automated AI-supported end-to-end technology validation pipeline aiming to increase trust in semiconductor devices by enabling a check of their authenticity. The high revenue associated with the semiconductor industry makes it vulnerable to counterfeiting activities potentially endangering safety, reliability and trust of critical systems such as highly automated cars, cloud, Internet of Things, connectivity, space, defence and supercomputers [7]. The proposed approach combines semiconductor device-intrinsic features extracted by artificial neural networks with domain expert knowledge in a pipeline of two stages: (i) a semantic segmentation stage based on a modular cascaded U-Net architecture to extract spatial and geometric information, and (ii) a parameter extraction stage to identify the technology fingerprint using a clustering approach. An in-depth evaluation and comparison of several artificial neural network architectures has been performed to find the most suitable solution for this task. The final results validate the taken approach, with deviations close to acceptable levels as defined by existing standards within the industry.
