Synth∀ Instruction-Centric Microarchitectural Side Channel Mitigations
Speaker: Alexandra Michael
Alexandra Michael is a fourth-year PhD student and NSF GRFP recipient at the University of Washington, advised by Dan Grossman and David Kohlbrenner. Alexandra's research focuses on using formal methods to precisely specify and build verifiable solutions to challenges in hardware security, particularly around hardware side channels. She is interested broadly in using formal methods to provide sound security guarantees, including in emerging spaces like agentic security.
Abstract
As hardware manufacturers increasingly build novel microarchitectural optimizations into their chips, these optimizations often create exploitable side channel vulnerabilities. Yet the availability of defenses has not kept pace. Both hardware- and software-level mitigations take time and expertise to develop—time in which a vulnerability remains open, and expertise that may not always be readily available. We address these challenges by (1) defining a lightweight formalism for specifying instruction-centric leakage semantics; and (2) ingesting such specifications into a custom program synthesis toolchain that automatically generates transforms for a dynamic set of instructions and leakage semantics. We present Synth∀, a framework for taking user-defined leakage semantics and using them to generate non-leaky transformations for target x86 instructions. Synth∀ combines traditional program synthesis and superoptimization tools to provide provably safe and correct program transformations. We evaluate Synth∀ on an abstract x86_64 CPU implementing a broad set of computation simplification optimizations.