Scalable and efficient implementation of correlation power analysis using graphics processing units (GPUs)

Abstract

Correlation Power Analysis (CPA) is a commonly used side-channel attack (SCA) on cryptographic devices, which analyzes power consumption to extract secret information like cryptographic keys. In this work, we have developed an open-source side-channel evaluation platform to evaluate the resilience of a range of devices to SCAs. Our platform includes an experimental setup for power trace collection and a trace analysis library. The time and effort to extract key values can greatly hamper our ability to analyze a single device. In this paper, we describe our work to leverage a Graphics Processing Unit (GPU) to accelerate key extraction. We develop a parallel framework in the Open Computing Language (OpenCL). OpenCL allows our framework to remain portable across a range of processing devices including CPUs, GPUs, and FPGAs. We describe the capabilities of our side-channel evaluation platform, and demonstrate how we leverage parallel processing to provide for more efficient and scalable side-channel analysis.