12/13/2023 Eleanor Wyllie
ECE assistant professor and Y. T. Lo Faculty Fellow Jian Huang has won the inaugural ACM SIGMICRO Early Career Award for pioneering contributions to persistent memory, storage architectures and systems. This is a new annual award recognizing a young researcher who has made outstanding research contributions to the field of microarchitecture in the early part of their career.
Written by Eleanor Wyllie
ECE assistant professor and Y. T. Lo Faculty Fellow Jian Huang has won the inaugural ACM SIGMICRO Early Career Award for pioneering contributions to persistent memory, storage architectures and systems. This is a new annual award recognizing a young researcher who has made outstanding research contributions to the field of microarchitecture in the early part of their career.
We use memory and storage every day – whether it’s saving files onto a laptop or uploading photos to the cloud from your phone. Huang’s work has tackled access to storage, securing data against ransomware attacks, and using machine learning to optimize memory and storage performance. Now, he’s excited about advancing memory and storage technologies to support AI infrastructures.
Three points summarize Huang’s groundbreaking work over the past five years: data access interface, architecture specialization, and development approaches.
The traditional method of accessing storage devices uses a generic “coarse-grained” block interface. This created a hard boundary between memory and storage, which has become a fundamental bottleneck for scaling the memory hierarchy today. Huang challenged this by using a more “fine-grained” byte interface and unifying the memory and storage. This helped expand memory capacity of data centers and is extremely important given the global memory shortage.
Huang’s team build specialized storage for improved storage efficiency and security. These specializations enable new storage properties, and Huang has made significant progress in generalizing them into essential parts of modern storage architectures. The ransomware-aware storage developed in his group can recover data and enable trustworthy forensics after ransomware attacks. Huang also developed the first trusted execution environment for computational storage, and the first-generation software-defined rack-scale storage system for data centers with network-storage co-design.
As today’s storage systems and architecture grow increasingly complex, it is challenging to meet the application requirements with current human-driven approaches. Huang’s team is developing a machine learning-based memory/storage ecosystem. The team have already developed a learning-based storage controller that automatically learns address mappings, a learning-based framework to drive the design of software-defined storage devices, and a learning-based storage harvesting system for data centers. His research on learning-based storage received an NSF CAREER Award.
The next step is AI infrastructure. Although the industry has made advances in building the chips and accelerators powerful enough to fuel this new trend, Huang says there is still a longstanding gap between compute and storage. AI needs data to advance, and AI chips must be able to access that data at high speeds. Huang is excited to combine memory and storage technologies with new computing hardware to build the next-generation AI infrastructure that can accelerate knowledge discovery and retrieval. We look forward to seeing the results.