Research
Efficient Emulation on RISC-V Hardware
The modern processor landscape is dominated by two proprietary Instruction Set Architectures (ISA): x86 and ARM.
Developing CPUs based on these architectures requires expensive licensing agreements from Intel or ARM, creating
a financial barrier that prevents small and medium sized companies from entering the market. To address this issue, the research community has proposed RISC-V: an extensible, open-source ISA. While RISC-V has made significant advancements with high-performance open-source CPUs and a robust extension ecosystem, it faces a significant
obstacle for widespread consumer adoption: software compatibility. In this project we are investigating efficient ways of emulating popular ISAs on RISC-V hardware to lift the software barrier. (Student: Victor Jimenez)
High Performance and Efficient Fault Tolerance
As transistors shrink and workload runtimes lengthen, transient faults causing Silent Data Corruptions (SDCs) have emerged as a critical concern in High-Performance Computing (HPC). Hardware-based protection strategies offer highly efficient instruction replication and early recovery, but their inherent lack of application-level visibility forces them to indiscriminately protect all operations, incurring severe performance penalties. Prior studies have shown
that exhaustive protection is unnecessary since many instructions have a small probability of inducing SDCs. While software-based mechanisms can selectively target these vulnerable instructions, they incur significant recovery overhead and inefficient instruction replication. In this project we are investigating potential approaches to efficiently and effectively prevent SDCs from ocurring in long running applications. (Student: Mateo Cantagallo)
Efficiently Evaluating Performance and Security of Microarchitecture
Comparing the security of microarchitectural defenses is a difficult task. Defenses are often evaluated in performance-oriented simulators, which lack the ability to also measure security. Recent security quantification methods are typically implemented in custom frameworks that do not report performance. Furthermore, prior work either lacks in generalizability or adaptability, making it difficult for computer architects to evaluate the robustness of the design or for non-security experts to reason about the security challenges. In this project we are working to develop generalizable, accessible and intuitive metrics and frameworks to elevate the evaluation of microarchitectural security to the same level of performance and energy. (Student: Phaedra Curlin)
Spaceborne Low-Energy AI Computing
In this project we are developing and analyzing secure architectures for using an AI accelerator. e are investigating the RISC-Vdesignwiththe Keystone trusted execution environment (TEE). The TEE offers a third execution mode (the M mode) that has exclusive control over how interrupts are delivered to the core and controls the entries inside the physical memory protection module. This statically isolates memory regions to different execution domains. (Student: Ian Barnaby).
Software-Hardware Co-Design for High Performance Acceleration
Modern domain-specific accelerators expose substantially different bottlenecks in memory bandwidth, limited on-chip
storage capacity, numerical format support, and data-movement overhead. Performance gains from algorithmic transformations—such as reduced precision, iteration-persistent dataflows, and partitioned on-chip storage—are increasingly dependent on algorithm–hardware co-design rather than raw peak throughput alone. In this project we leverage the heterogeneity of commercial accelerators to understand the advantages and disadvantages of each. We are particularly interested in accelerating non-linear integral equation solvers as these are commonly found in a variety of applications
Secure Computer Architectures
With the emergence of public cloud computing and the internet of things, modern computer architectures are inadequate to address the security and privacy concerns of users. In this work, we look for innovative solutions to secure the computer from the hardware up. By designing the computer architecture with security as a number one priority, along with performance, we can provide additional guarantees about the security and privacy of data.
Secure Non Volatile Memory
Non-volatile storage technologies have advanced so much in the last several years that system architects are now considering to use them as main memory. Recently Intel came out with their new non-volatile technology, , that integrates a non-volatile storage device to be used as main memory. A lot of work has gone into making sure this new technology performs just as fast as its volatile counterparts. However, not many pieces of work have looked at how to secure the data in non-volatile memory against physical attacks. In this work we are looking at ways of redesigning the secure memory constructs to be applicable to non-volatile memory devices. We are looking to integrate both hardware and software techniques to protect a system that uses non-volatile main memory from physical attacks, privilege escalation attacks and attacks that look to scramble values in memory to steal secrets (e.g. rowhammer attacks). We also want to take advantage of the durability of the non-volatile devices to design ways of recovering from attacks.
Re-EnforcingCritical Thinking Skills to Defeat Fake News
Fake news have proliferatedsocial media platforms for a while now. It is clear that foreign and other actors have used fake news to increase polarizationin our society and impact democracy in the United Stated.In this work we are looking to understand what affects the ability of a person to identify fake news. Withthese observations we willdevelop tools that provide resilience against malicious actors that use fake news to affect our democracy.
Evaluating The Impact of The Russian Agency in Twitter
Social media platforms have been playing an important role in democracy. The ability to reach millions of users with a single Tweet, or Facebook post, is a simple way to get a meesage across. This is why foreign actors have used social media platforms to try to influence elections. In the United States, the Russian Internet Research Agency (IRA)tried to influence the 2016 presidential election on Twitter. While Twitter deleted their accounts once they were identified, there have been lots of work in both the public and private sector to understand their impact. A lot of these pieces of work concluded that the IRA behavior on Twitter hadno impact on users. Unlike prior work, we analyzed the before and after behavior of the users contacted by the agency, and actually did find a difference in their behavior.
GHOST: 5G Hidden Operations through Securing Traffic
5G is emerging and is expected to soon become near-ubiquitous around the world. Hence, U.S. government organizations such as the military and State Department, as well as nongovernmental humanitarian aid organizations, and private sector enterprises should take advantage of indigenous 5G networks to eliminate the costs of installing and maintaining an alternate communications infrastructure. However, in many areas of the world, 5G networks are deployed and operated by organizations that are untrusted and potentially hostile to the U.S. In these environments, new security technologies must enable secure operations over untrusted networks. The GHOST project protects end-user devices and non-indigenous networking equipment from potential compromise through the use of Trusted Execution Environments. The GHOST project prohibits traffic analysis through two mechanisms: the use of Software Defined Credentials; and the use of anonymization techniques to obfuscate communications connections. Finally, the GHOST project obscures changes in traffic volume by maintaining a minimum level of “GHOST” traffic, and provides for pre-scripted traffic models to confuse and mislead traffic analysis.