학술논문

CurIAs: Current-Based IC Authentication by Exploiting Supply Current Variations
Document Type
Periodical
Source
IEEE Transactions on Computers IEEE Trans. Comput. Computers, IEEE Transactions on. 72(2):466-479 Feb, 2023
Subject
Computing and Processing
Integrated circuit modeling
Counterfeiting
Cloning
Inverters
Hardware
Switches
Silicon
PUF
hamming distance
FPGA
LFSR
NIST
HSPICE
Monte-Carlo simulation
counterfeit
intrinsic
CRP
authentication
randomness
modeling attack
HaHa board
Language
ISSN
0018-9340
1557-9956
2326-3814
Abstract
Physical unclonable functions (PUFs) have emerged as one of the most notable hardware primitives to mitigate the ever-growing global issue of counterfeiting and cloning of integrated circuits (ICs) in recent times. PUFs exploit the intrinsic manufacturing process-induced parametric variations for generating unique chip identifiers. However, most of the existing PUF implementations require complex structures or the inclusion of additional components, which incur performance and area overheads. In this work, we introduce CurIAs , a supply current-based novel PUF implementation to authenticate ICs and to protect them from counterfeiting attacks. It exploits the dynamic current stemming from temporal switching activities in existing on-chip structures as an entropy source to generate high-quality IC-specific digital signatures. First, we investigate the source of the entropy of this PUF, i.e. , the dynamic current variations in different circuit structures, with transistor-level Monte-Carlo simulations in HSPICE. Next, to evaluate its effectiveness in Silicon, we apply this approach to map LFSR (Linear Feedback Shift Register) designs into 20 FPGA chips (fabricated in TSMC 55nm process node), perform practical measurements, and generate digital signatures. These signatures show high uniqueness, robustness, uniformity, and randomness features, and the overall implementation requires modest hardware overhead ($<1%). We assess and substantiate the robustness of this approach at eight different operating points by varying supply voltage and temperature. Furthermore, we upscale the design to more extended LFSR sizes, and it exhibits a constant trend of performance improvement over the operating points. Through a judicious selection of challenge vectors, CurIAs demonstrates a high resilience against model learning attacks, with an average prediction accuracy of 50%. These intrinsic variations in supply current across ICs for varying workloads entail unique chip-specific signatures, which are extremely difficult to clone, and can be deployed effectively against IC counterfeiting issues.