Distributed On-Chip Power Supply for Security Enhancement in Multicore NoC
Xingye Liu, Paul Ampadu · 2021
We propose a distributed DC/DC converter against correlation power analysis and input voltage glitch injection attacks in emerging multicore applications. Moving towards Internet-of-Things, heterogeneous integration and Network-on-Chip architectures, protecting data privacy and critical operation information for every device become crucial. The proposed DC/DC converter is designed to mitigate correlation power analysis by weakening the input-output relationships for various types of workloads. For steady-state workloads, the correlation factors between input and output currents are reduced to only 0.05. For digital workloads, the correlation factors can vary by 10 times when there are only a few nanoseconds delay of the load current, which greatly prevents attackers from deriving the circuit real operations. Meanwhile, the converter is able to resist 20% input voltage glitches without generating additional spikes or droops while controlling the voltage ripples within 7.5% of the output. Implemented and simulated in 32nm CMOS technology, single channel peak efficiency reaches 85% based on post-layout models. The converter is also able to provide up to 19.3V/µs reference tracking capability for dynamic voltage scaling requests. The security enhancement induces about 20% area overhead and an average 3% efficiency loss. Overall, the proposed converter is a promising countermeasure solution for side-channel attacks in multicore systems without sacrificing too much performance.