Kronos : verifying leak-free reset for a system-on-chip with multiple clock domains

Abstract

Notary [3] uses formal verification to prove a hardware-level security property called deterministic start for a simple system-on-chip (SoC). Deterministic start requires that an SoC's state is fully reset by boot code to ensure that secrets cannot leak across reset boundaries. However, Notary's approach has several limitations. Its security property requires that all of the SoC's microarchitectural state can be reset to known values through software, and the property and proof technique apply only to SoCs with a single clock domain. These limitations prevent Notary's approach from being applied to more complex systems. This thesis addresses these limitations through Kronos, a system consisting of a verified SoC that satisfies a new security property called output determinism. Output determinism provides the same security guarantees as Notary without requiring that all of an SoC's state be reset by software. The SoC used in Kronos, called MicroTitan, is based on the open-source OpenTitan [16] and includes multiple clock domains. This thesis evaluates Kronos and demonstrates that existing open-source hardware can be modified to satisfy output determinism with minimal changes, and that the process of proving output determinism reveals hardware issues that violate desired security guarantees.