Silicon-Level Trust in the Edge AI Era: A Survey of Hardware Vulnerabilities and Defensive Architectures

Abstract

The modern supply chain of semiconductor manufacturing industry is highly globalized with chip design, thirdparty intellectual property integration, offshore manufacturing, packaging, system integration and deployment being its main components. Even though this distributed paradigm provides rapid innovation and scalability, it provides wide assumptions of trust across the product lifecycle. This lack of transparency and control may cause a hardware system to be prone to vulnerability because of either design flaws or fabrication or deliberate malicious modification by untrusted parties. The current survey questions the origins and spread of such vulnerabilities along the supply chain, highlighting such critical attack entry points as hardware Trojans, malicious code injection, testinfrastructure breach, parametric backdoor, side-channel leakage, and firmware-hardware interface failures. These weaknesses provide the attackers with the opportunity to unleash insidious and relentless attacks, which threaten the system integrity, confidentiality, and availability. We discuss modern countermeasures, which include pre-silicon verification, design-for-trust procedures, post-silicon testing, runtime probing, formal analysis, and machine-learning-based detection. Placing the previous research as part of the progression of the semiconductor supply chain, this survey outlines the gaps in defense and formulates research gaps to address the challenges of implementing complete end-to-end silicon-level trust.