Syracuse University

Eager: Radiant: Resilient Architecture for Dram-based In-memory-processing Acceleration Using Non-hardened Technology -computing Systems in Space Enable Essential Technologies Like GPS, Satellite Communications, and Agricultural Monitoring. However, These Systems Face Harsh Challenges, Particularly Radiation, Which Can Severely Degrade or Destroy Electronic Components. Traditional Radiation-hardened Techniques That Address This Challenge Are Both Costly and Based on Outdated Technology, Limiting Performance and Flexibility. at the Same Time, Renewed Interest in Lunar and Martian Exploration Is Driving Demand for Far More Capable Space-based Computing. Fortunately, a Promising Approach Called In-memory Processing Is Being Explored Where Memory Directly Performs Computation. Using This Method, Memory Can Function Like an Accelerator Suitable for Enabling State-of-the-art Image and Signal Processing and Artificial Intelligence (AI) Approaches That Would Be Otherwise Impractical. Memory-based Acceleration Reduces the Burden on and Complements Central Processors for Space Computing Systems. the Radiant Project Investigates Whether Modern Commercial Memory Devices, Not Originally Designed for Space, Can Function Reliably and Provide In-memory Processing Capabilities in Radiation-rich Environments Through Appropriately-designed Error Correction Techniques. the Research Supports National Priorities by Advancing Space Computing Capabilities, While Also Offering Interdisciplinary Education Opportunities That Span Computer Science, Engineering, and Physics. Radiant Has Two Main Technical Goals. First, It Characterizes the Behavior of Commercial Dynamic Random-access Memory (dram), the Dominant Technology for Main Memory in Modern Computers, Under Space-like Conditions. the Behavior of Commercial Dram in Environments With Elevated Radiation and Temperature Shifts Still Remains Relatively Poorly Understood. the Study Examines Dram During Both Conventional Use and In-memory Processing to Identify Common Fault Modes and Their Dependence on Memory Architecture and Access Patterns. Second, the Project Develops Fault Tolerance Techniques to Protect Dram for In-memory Computing, Which Introduces Unique Challenges Not Addressed by Traditional Error Correction Methods. These Include Developing Novel Error Correction Codes That Work for In-memory Processing Operations, as Well as Memory Mapping Strategies That Account for Weak or Failure-prone Regions. Together, These Efforts Aim to Make Advanced, Low-cost Memory Technologies Both Viable and Substantially More Capable. Radiant Can Provide the Fundamental Capabilities to Allow Supporting Tensor-based Algorithms, Including the Latest Ai Approaches Such as Large Language and Foundation Models, in Future Space Missions. This Award Reflects NSF'S Statutory Mission and Has Been Deemed Worthy of Support Through Evaluation Using the Foundation's Intellectual Merit and Broader Impacts Review Criteria.- Subawards Are Not Planned for This Award.