Date of Completion

8-25-2017

Embargo Period

8-25-2017

Major Advisor

Marten van Dijk

Associate Advisor

Omer Khan

Associate Advisor

John Chandy

Field of Study

Electrical Engineering

Degree

Doctor of Philosophy

Open Access

Open Access

Abstract

Computation outsourcing to a cloud cluster has become a common practice in today’s world. Unlike traditional computing models, the cloud computing model has raised serious challenges in terms of privacy and integrity of the users’ sensitive data. Depending on their respective adversarial models, modern secure processor architectures offer a promising solution to many of these challenges. However, almost all secure processor proposals dealing with computation outsourcing face two crucial security vulnerabilities that are still treated as open problems; the threat of hardware Trojans embedded inside the secure processor chip, and, the threat of privacy leakage via access patterns of the secure processor to the untrusted memory. In order to deal with the above mentioned vulnerabilities, I propose various architectural primitives that address each of these challenges individually in an efficient manner. For hardware Trojans detection, a first rigorous algorithm HaTCh is proposed that not only detects a small set of publicly known Trojans, but also detects an exponentially large class of deterministic hardware Trojans. Oblivious RAM (ORAM) is an established technique that hides memory access patterns through redundant accesses, thereby preventing privacy leakage. However such redundancy incurs a large performance overhead. Therefore, a dynamic prefetching technique tailored to ORAM is proposed that detects and exploits data locality without leaking any information on the access pattern, and results in significant performance gain. Recent research has shown that information can still be leaked even if only the memory write-access pattern (not reads) is visible to the adversary. For such weaker adversaries, a fully functional ORAM causes unnecessary overheads. With this intuition in mind, an efficient write-only ORAM scheme has been proposed which substantially outperforms the closest existing write-only ORAM from the literature.

COinS