DECODE: Detecting Structural Obfuscation in HLS based DSP Designs using Machine Learning Framework

Structural obfuscation has been a widely used hardware security technique that is used to transform the internal structure of a high-level synthesis (HLS) digital signal processing (DSP) intellectual property (IP) design such that it becomes unobvious and obscure. The intent is to thwart deobfuscation/reverse-engineering (RE) attempts. These transformations in structurally obfuscated DSP designs are elusive and non-intuitive, making them arduous to identify and detect from an attacker's perspective. In this paper, we for the first time, present a novel strategy for detecting structural obfuscation in HLS based DSP designs using machine learning (ML) frameworks. This paper presents a novel technique called 'DECODE' : DEtecting StruCtural Obfuscation in HLS based DSP Designs using Machine LEarning Framework . The presented approach introduces a set of unique/distinct features that can successfully classify between structurally obfuscated HLS DSP designs and its non-obfuscated versions. The paper also presents a novel ML database consisting of several instances of structurally obfuscated and un-obfuscated DSP HLS designs that can be fed into our ML framework (logistic regression and random forest classifier). Results of the proposed approach revealed successful detection of structurally obfuscated DSP designs with high accuracy.