Safe, secure and trustworthy compute-in-memory accelerators

Compute-in-memory (CIM) accelerators based on emerging memory devices are of potential use in edge artificial intelligence and machine learning applications due to their power and performance capabilities. However, the privacy and security of CIM accelerators needs to be ensured before their widespread deployment. Here we explore the development of safe, secure and trustworthy CIM accelerators. We examine vulnerabilities specific to CIM accelerators, along with strategies to mitigate these threats including adversarial and side-channel attacks. We then discuss the security opportunities of CIM systems, leveraging the intrinsic randomness of the memory devices. Finally, we consider the incorporation of security considerations into the design of future CIM accelerators for secure and privacy-preserving edge AI applications. This Perspective examines the development of safe, secure and trustworthy compute-in-memory accelerators, exploring vulnerabilities specific to such accelerators and discussing the security opportunities of the systems.