Research

Robust and Secure Deep Learning Computing

Given the ever-increasing integration density and design complexity of computing systems, security and robustness become one of the first-priority citizens that need to be considered while designing a system. Though various solutions have been proposed and implemented in real systems for the known attacks, different systems and workloads have unique security holes and need different defenses. We would like to explore unique vulnerabilities of various systems and workloads and research effective countermeasures. [NSF 2114514]

Energy-efficient & Scalable Throughput Processors

Most of emerging workloads need massive data- and thread-level parallelism to handle big data inputs. Various throughput processors (e.g., GPU) are equpped with hundreds of cores and several mega-bytes of on-chip memories. These abundant resources are a double-edged sword that are inevitable for increasing execution concurrency but at the same time require high power consumption. In this project, we would like to optimize or revisit throughput processors to achieve higher throughput per energy consumption. 

Next-generation Accelerator Design

In the post-Moore era, we can't simply rely on more transistors per unit area to produce better throughput. Domain-specific accelerators may be able to satisfy ever-increasing computing power demands of emerging workloads with given transistor and power budget. However, here comes another challenging questions,  'what types of workloads should we support with dedicated accelerators?' and 'how can the existing computing systems smoothly support the accelerators?'. In this project, we would like to answer these questions such that the new computing era can be smoothly and cost effectively phased into.

Emerging Workload Acceleration

For an agile support for quickly evolving computing fields, it is essential to have a thorough undestanding of new workloads. To enable fair evaluations of various accelerators, it is important to have balanced and good-quality benchmark suites of emerging workloads. In this project, we aim to design various cutting-edge applications and explore optimization potentials. The workloads will be compiled into benchmark suites so that the community can use the suites for their research. 

Publications 

* People in MoCA Lab

Yuan Feng* and Hyeran Jeon
ACM Workshop on General Purpose GPUs (GPGPU), Montreal, Canada, Feb 2023


Nigel Bernard*, Hoa Nguyen*, Aman Chandan*, Savyasachi Jagdeeshan*, Namdev Prabhugaonkar*, Rutuja Shah*, and Hyeran Jeon
arXiv Preprint, July 2022


Mujahid Al Rafi*, Yuan Feng,* and Hyeran Jeon
arXiv Preprint, July 2022




Mujahid Al Rafi*, Yuan Feng*, and Hyeran Jeon
Workshop on Negative results, Opportunities, Perspectives, and Experiences In conjunction with ASPLOS-27 (NOPE), Feb 2022









































Patents

US20150199150 A1


Hyeran Jeon, Woohyong Lee, Mingyu Lee, Woongee Kim, Jiseong Oh, Jagun Kwon, and Taekgyun Ko

US8423723 B2