GGrantIndex
← Search

CIRC: New: MemSysExplorer: Community Tools and Interfaces for Exploring and Evaluating Next-Generation Memory Systems

$1,772,466FY2024CSENSF

Tufts University, Medford MA

Investigators

Abstract

Computer systems store information in memories; the specific physical mechanisms of storing and moving these ones and zeroes are currently under intense study by the research community. The successful design of a memory system provides a balance between cost and performance by applying a hierarchical combination of memory technologies, circuits, and architectures. Historically, the design of such systems has consistently relied on specific technologies to serve a specific function in the hierarchy — static random access memory (SRAM) for on-chip caches and scratchpad memories, dynamic random access memory (DRAM) for main memory, and magnetic disks or flash drives for storage. More recently, many ideas for different memory designs have emerged at varying stages of development and deployment, each with its costs and benefits. It is hard for a hardware designer or programmer to predict how a specific new memory will affect the speed and energy consumption of a computing system. MemSysExplorer will provide the research community with the tools required to explore this new complex design space and identify principled design solutions that satisfy the requirements of different applications and systems. The recent growth in data-intensive applications, heterogeneous architecture design, and new memory technologies has drastically changed both design requirements and the landscape of design options available to meet them. This project develops a cross-community design space exploration and evaluation framework offering researchers the capability of providing design inputs and simulating the resulting memory system solutions at different levels of the design stack, which are broadly defined as 1) application design space, 2) system design space, and 3) technology design space. Users can configure each level independently and evaluate the holistic impact of specific design optimizations. The framework's flexibility in generating a large variety of design solutions is supplemented by an integrated web-based data visualization tool to simplify the result navigation and filter to identify optimal design points. The MemSysExplorer framework introduces the opportunity to conduct multidisciplinary research leveraging a common simulation and evaluation platform. The MemSysExplorer project brings together researchers from industry and academia with a range of specialties from memory devices, circuits, computer architecture, computer systems, programming languages, and compilers. Through a steering committee, public tutorials, and online resources, MemSysExplorer will enable researchers to collaborate across disciplines and contribute to communal tools and datasets. The openly accessible and easy-to-use web-based interface will introduce students and early practitioners to new computing research problems in which memory plays a leading role. MemSysExplorer is primarily developed and supported by a team of researchers at Tufts University, Harvard University, and Amherst College. The project website will be updated regularly with example code, detailed models, tutorials as well as a calendar of public tutorials and outreach events. Resources and general information about the MemSysExplorer framework are available at https://memsysexplorer.eecs.tufts.edu 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.

View original record on NSF Award Search →