Major Research Topics
SAIL is devoted to creative computer system design and problem solving. Our mission is to build powerful micro data centers that support emerging intelligent digital services for the next 10 years and beyond.
Some of our on-going data center system research projects are described as follows:
 |
Chip Architecture for Multimodal AI The advancement of AI and IoT technologies has boosted the adoption of multimodal computing, which refers to the complex computing process of federating multiple modalities (e.g., image, audio, etc). It is important to build efficient AI chips and systems for handling this type of new workload. Representative Papers: ISCA'23, RTSS'23, Euro-Par'23, IWQoS'24, RTSS'24, TACO'24 |
 |
Acceleration for Complex Big Data As a ubiquitous data structure, graph holds the information of entities and the relationship between them. Graph computing and sampling are important procedures for understanding graphs. It is crucial for parallel computing systems to better support graph big data. Representative Papers: ICCD'17, TPDS'19, PACT'21 TACO'21, TC'23, VLDB'24 |
 |
Hardware Tracing and Optimization Emerging cloud/AI infrastructures allow users to abstract away the low-level hardware complexity from users. However, to fully unleash system performance, the next-generation data center must support efficient hardware tracing and fine-grained resource optimization. Representative Papers: ICPP'19, TC'21, SoCC'23, IPDPS'23, Science China'24 |
 |
Memory and Data Storage Management The ever-growing memory demand of applications pose important challenges to today's AI data centers. The continuous advancement of new storage and memory devices motivates the development and optimization of more efficient server memory management strategies. Representative Papers: HPCA'18, TC'19, IPDPS'22, ICCD'22, SC'24 |
 |
Energy-Efficient Hardware Architecture Green consciousness is quickly making its way into computer design. We have strong incentives to limit system power to improve sustainability. Datacenters need appropriate power management strategies and highly efficient chip hardware to reduce energy while ensuring quality of service. Representative Papers: ISCA'16, ICCD'18, ICPP'19, SC'20, TPDS'24, ISCA'24 |
Acknowledgement
The members of SAIL would like to thank our research sponsors and collaborators:
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |