Terahertz Based Machine Learning Approach to Integrated Circuit Assurance

Abstract

Semiconductor production has largely outsourced the design and fabrication of integrated circuits (ICs) to untrusted regions at the expense of hardware trust. Critical electronic equipment found in multiple domains including military, healthcare, and power grid applications rely upon the functionality of ICs to perform various computing, sensing, and actuating operations. Hardware Trojans (HTs) are being increasingly found in defense systems and amongst many critical industries that rely upon components from untrusted regions [1]. Industrial non-destructive techniques (NDT) are unable to ensure total assurance without the use of time consuming endeavors increasing verification price per component. Terahertz (THz) based IC inspection methods have been researched over the past decade [2]. Due to the large amount of data created by THz equipment, there has been widespread adoption of artificial intelligence (AI) based methods for reconstruction, resolution enhancement, and automated detection of counterfeit IC samples. In its raw form, THz radiation suffers from poor spatial resolution, but this can be improved significantly through modelling of the THz source's point spread function PSF. Prior research has shown that object functions (signals from a known object mask) generated from 2D X-ray can be used to simulate the PSF and enhance THz spatial resolution for ICs [3]. In this work, advanced metrology methods along with simulation software across the electromagnetic spectrum are proposed to improve the characterization of authentic versus counterfeit IC samples. This high fidelity model can be used for unsupervised deep learning (DL) utilizing the digital structural and spectral features for comparison. The trained DL model can augment the spatial and spectral enhancement of low-resolution THz data collected rapidly in the far-field. This process framework can provide a method of automated counterfeit detection requiring minimal to no manual labor requirements.