This CAREER project will investigate future memory systems that are scalable with processor technologies and application demands. It identifies memory architecture as a key issue in developing next-generation memory systems that are scalable in terms of performance, power-efficiency and cost, and support heterogeneous memory devices. A Scalable and Universal Memory Architecture (SUMA) is proposed to replace the overly restrictive and homogeneous memory architecture in current computer systems. The project will further investigate a set of innovative techniques to improve memory scalability, including adaptive memory organization, heterogeneous memory management, proactive memory power and thermal management, and flexible memory error protection schemes.
With a scalable and universal architecture, computer memory may scale cost-effectively and power-efficiently from gigabytes now to hundreds of terabytes in the future. Many memory-bounded computing problems will be solved at a much larger scale, and many previously intractable problems will become solvable. The proposed education component will expose undergraduate students to the complex performance, power, and thermal issues in the multi-core era, with a focus on memory systems.
