Project Grant 2541050

This National Science Foundation (NSF) Computer and Information Science and Engineering (CISE) CFDA 47.070 Project Grant, awarded to The Leland Stanford Junior University on September 1, 2026, provides $424,237 in federal funding through August 31, 2031 to develop a retention-aware computing infrastructure that optimizes memory systems for large-scale artificial intelligence (AI) applications. The research addresses a critical inefficiency in current memory architectures by designing heterogeneous memory systems that treat data persistence as a central design consideration, matching short-lived and long-lived data to differentiated memory technologies optimized for their respective retention windows. This approach aims to reduce energy consumption, improve density scaling, and accelerate AI systems in both datacenter and edge computing environments.

The research is organized around four integrated technical activities: establishing a profiling framework to characterize data lifetime utility and map data to appropriate memory tiers; developing algorithmic techniques to restructure computation and data movement for better retention constraint satisfaction; and advancing the computing stack implementation across hardware and software layers. Additionally, the project supports education and workforce development through new course materials, industry engagement initiatives, and open-source resources to expand participation in next-generation AI hardware accelerator design and computer engineering, contributing to the NSF CISE program's broader objective of advancing research cyberinfrastructure and supporting innovation across computing and information science domains.