Addendum: Systematic Evaluation of Randomized Cache Designs against Cache Occupancy

TL;DR

The work addresses whether randomized caches can offer performance competitive with modern LLCs while mitigating occupancy- and contention-based side-channel leakage. It employs a systematic, benchmarking-centric evaluation across threat models including covert channels, process fingerprinting, and AES-key leakage, and the addendum specifically examines the roles of L1d size and eviction-map seeding (MIRAGE+). The findings show that simply increasing L1d size or randomizing global eviction mappings does not fully prevent AES leakage, implicating LLC contention as the root cause and highlighting the need for isolation-based designs. Domain isolation approaches, such as SassCacheginer-style isolation, offer stronger security at the cost of performance, guiding future hardware cache designs toward leakage-resistant architectures with explicit isolation between processes.

Abstract

In the main text published at USENIX Security 2025, we presented a systematic analysis of the role of cache occupancy in the design considerations for randomized caches (from the perspectives of performance and security). On the performance front, we presented a uniform benchmarking strategy that allows for a fair comparison among different randomized cache designs. Likewise, from the security perspective, we presented three threat assumptions: (1) covert channels; (2) process fingerprinting side-channel; and (3) AES key recovery. The main takeaway of our work is an open problem of designing a randomized cache of comparable efficiency with modern set-associative LLCs, while still resisting both contention-based and occupancy-based attacks. This note is meant as an addendum to the main text in light of the observations made in [2]. To summarize, the authors in [2] argue that (1) L1d cache size plays a role in adversarial success, and that (2) a patched version of MIRAGE with randomized initial seeding of global eviction map prevents leakage of AES key. We discuss the same in this addendum.

Figures (1)

• Figure 1: Kernel Density Estimators for (1) the original L1d cache size considered in chakraborty2025systematic; vs (2) the L1d cache size in cao2025yet. The shape assumed by both distribution is gaussian.