Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier

https://orcid.org/0000-0003-3933-2099

AccessType

Open Access Dissertation

Document Type

dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Program

Electrical and Computer Engineering

Year Degree Awarded

2024

Month Degree Awarded

February

First Advisor

Daniel Holcomb

Subject Categories

Electrical and Computer Engineering

Abstract

Field programmable gate arrays (FPGAs) are integrated circuits that consist of programmable logic that a user can configure and deploy for applications such as hardware emulation and accelerating high performance computing. In recent years, the emergence of FPGAs in the cloud has led to research on multi-tenant FPGAs. In a multi-tenant scenario, the same FPGA fabric is shared among multiple users, or among multiple untrusting IP cores. Multi-tenancy has economic benefits, largely due to improvements in resource utilization, but also brings new security concerns since the tenants could behave maliciously. Although the tenants sharing an FPGA are logically isolated from each other, they may still have unintended interactions through side channel attacks and fault attacks. In this dissertation, we aim to evaluate security threats and defenses in the multi-tenant FPGA scenario. Firstly, the work in this dissertation studies a true random number generator (TRNG) on cloud FPGAs that is robust against voltage manipulation from co-tenants. The TRNG design is based on harvesting clock jitter using a tunable time-to-digital converter circuit. In accordance with best practices, a stochastic model is built to evaluate the min-entropy of the design, and further validated by NIST entropy assessment test suite and NIST statistical tests. The basic version of the TRNG is extended with a linkable sampling module to increase min-entropy per sample and throughput at a modest resource cost. Then the dissertation analyzes a type of fault attack that can be conducted by one tenant against another in a multi-tenant setting. Specifically, the fault attack is differential fault intensity analysis (DFIA), which is a biased-fault based attack on Advanced Encryption Standard (AES) circuits. Ring oscillators (ROs) are deployed as effective power wasters to cause a supply voltage drop through the shared power distribution network (PDN) of tenants. The attack is highly relevant to multi-tenant scenarios because the attacking tenant can create the voltage drop without physical access, and can precisely control the shape of the voltage drop by adjusting both the number of activated ROs and their duration as required for the attack. The voltage drop will in turn increase the delay in the logic and eventually cause specific timing faults which are analyzed to successfully recover the AES keys. In the last part, we use on-chip voltage sensors to detect the location of a target circuits. The sensing scheme leverages time-to-digital converters (TDCs) as voltage sensors, and a novel differential analysis is applied to the sensor data. In a multi-tenant setting, this method can be used either as part of a defensive scheme to monitor against attacks, or it can be used to probe a system and determine how to effectively target an attack to a particular co-tenant victim.

DOI

https://doi.org/10.7275/36470318

Download

Share

COinS