Trusted Memory Access Monitoring (TMAM): Detecting Fine-Grained DDR4 Access Patterns in FPGA Clouds

Vineet Chadalavada1, Nahush Tambe1, Naseeruddin Lodge1, Dhurva Aklekar1, fareena saqib2
1UNC Charlotte, 2University of North Carolina at Charlotte


Abstract

Modern multi-tenant FPGA cloud platforms enable hardware acceleration and high performance, but also introduce new vulnerabilities, particularly through shared memory resources as side-channel leakage. In particular, DDR4-based Prime+Probe attacks exploit row buffer contention to infer sensitive victim activity, across isolated hardware regions. This work presents a shell-level Trusted Memory Access Monitor (TMAM) that detects interference patterns indicative of such attacks. TMAM passively observes memory access sequences in real time, identifies irregular burst behavior, and flags anomalies without requiring modifications to tenant kernels or application logic. Integrated into the FPGA shell, TMAM operates non-intrusively, offering a practical and deployable defense against cross-tenant memory-based covert channels in reconfigurable cloud infrastructures.