How Tiny Voltage Changes Can Help Identify CPU Instructions

What is it about?

Every time a computer processor runs a program, it executes a sequence of small operations called instructions. These operations affect how much electrical current the processor draws, causing tiny and rapid changes in its supply voltage. Although these changes are not intended to reveal program behavior, they can contain clues about what the processor is doing. In this study, we investigate whether these subtle voltage variations can be used to identify executed instructions on a commercial x86 processor running at 3.1 GHz. We collected more than 50,000 voltage waveform samples covering 645 different instruction types, including advanced vector instructions. Based on this dataset, we developed a machine learning framework that analyzes voltage patterns over time and predicts sequences of processor instructions. In our experiments, the proposed approach achieved 34.70% accuracy on the instruction-level sequential prediction metric, outperforming the baseline methods we evaluated. This demonstrates that instruction-related information can remain observable in voltage fluctuations, even in processors with complex execution mechanisms. Our findings provide a new perspective on understanding unintended information leakage in modern processors and open up opportunities for further research into hardware security.

Featured Image

Photo by GAMERCOMP.RU on Unsplash

Photo by GAMERCOMP.RU on Unsplash

Why is it important?

Most previous instruction-level power side-channel studies have focused on low-frequency embedded processors with relatively simple architectures. However, modern commercial CPUs are much more complex, featuring high operating frequencies, out-of-order execution, deep pipelines, and multi-level caches, all of which make instruction-level analysis considerably more challenging. This work presents the first systematic demonstration of instruction-level side-channel disassembly on a commercial high-frequency x86 processor. By analyzing 645 instruction types and introducing a sequence-aware learning framework, we demonstrate that meaningful instruction-related information can still be extracted from voltage waveforms despite complex processor behaviors. This finding highlights an often-overlooked aspect of hardware security: the electrical behavior of a processor can unintentionally reveal information about its internal execution. Understanding this leakage is important for evaluating the security of modern computing systems and developing more effective hardware and software countermeasures. Beyond demonstrating this possibility, our work provides a foundation for further research into processor-level side channels, cross-platform generalization, and practical leakage mitigation. While the current study is an initial demonstration on one processor platform using direct voltage measurements, it opens an important direction for understanding and addressing unintended information exposure in complex CPUs.

Perspectives