This research targets the interactions between systems and machine learning.
As machine learning applications become more prominent, efficient systems in all scales (ranging from servers all the way to edge devices) must be designed efficiently to support these workloads, which brings many challenges as well as opportunities. Our research group is currently focusing on connecting the theory behind machine learning robustness requirements with best practices in designing resilience systems to enable resilient, efficient, and flexible systems tailored for machine learning applications. We are also working on leveraging machine learning advances to create new system design techniques.
What is the future of computing? It may be light.
To improve energy efficiency and speed of computing systems, we must fundamentally rethink how information is transported and processed. Utilizing electrical signals on a substrate that is composed of CMOS gates and wires, as how computing elements are made today, is increasingly challenging as technology scales to smaller and smaller feature sizes.
My research group is teaming up with researchers at the University of Illinois Urbana-Champaign and the University of Chicago to investigate a radical computing paradigm – using light to transport and process information. Utilizing a multi-patented three-terminal device, called the transistor laser, which can convert an electrical or optical input signal into an electrical output signal and a beam of light, the focus of our research is to create photonic interconnects to transmit information at very short distances such as between/on semiconductor chips, as well as to invent hybrid photonic/electrical processing elements that can perform computing operations extremely efficiently and quickly.
This project is awarded $2.5 million by the National Science Foundation in conjunction with the Semiconductor Research Corporation for the Energy-Efficient Computing: from Devices to Architectures (E2CDA) program. More details can be found in this news article.
Robust system design ensures that computing systems function as the user expects in the most efficient manner, despite underlying disturbances like hardware failures, design bugs, or malicious attacks.
As the world is heavily dependent on computing systems, system malfunctions can result in severe consequences ranging from productivity losses, financial losses, to even loss of human life. However, two trends pose significant challenges to designing robust systems – (1) massive design complexity for achieving aggressive performance/energy efficiency goals, and (2) increased hardware failures as transistor geometries continue to scale.
In our research group, innovative ideas and fresh perspectives are cultivated to tackle these challenges. Our multifaceted research spans topics such as cross-layer resilience, intelligent and adaptive resilient systems, self-healing architectures, approximate computing, and robust memory systems. Our research is crucial and timely to fuel computing system and technology advances.
Security is a first-order priority in many computing systems. In particular, hardware security is a primary concern and a focus in our research group.
As an example, our fail-secure architecture idea is based on two key observations: (1) the prevalent notion of existing security mechanisms is that the underlying hardware is always correct. However, due to rising hardware failure challenges, this notion may no longer stand. In fact, unreliable hardware can pose big security threats to computing systems; and (2) there are fundamental differences between ensuring functional correctness and guaranteeing security properties. The goal of this research is to investigate low-cost and effective system- and architecture-level techniques to ensure that a system is fail-secure, i.e., no security properties are violated even in the presence of hardware failures.
I am broadly interested in various emerging topics, for example, heterogeneous computing systems, monolithic-3D architectures, and neuroscience, biomedical, data analytics, and high-performance computing applications.