Collaborative Research: CSR: Medium: MemDrive: Memory-Driven Full-Stack Collaboration for Autonomous Embedded Systems

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

Embedded systems, relying on embedded hardware with stringent size, weight, and power constraints, play a crucial role in the modern world. The resource-constrained nature of such systems has the well-known memory wall problem, impacting both capacity and latency. To address this challenge, this project aims to establish a comprehensive memory management ecosystem, called MemDrive, to optimize memory usage throughout the application and system stack and explore opportunities for cross-stack optimization. MemDrive encompasses three key innovations: (i) At the application stack, cross-task optimization techniques will be designed to share intermediate results across different tasks without compromising performance. (ii) At the system stack, memory and resource management schemes will be implemented that could embrace the hardware trend of integrated physical memory and further reduce memory redundancy across multiple processes. (iii) Cross-stack collaboration will be explored to further improve memory usage efficiency and latency by leveraging application-stack properties, such as shared intermediate results among tasks and predictable memory usage patterns. Extensive experiments for running real-world applications on top of a set of robotic testbeds will be conducted. This research has the potential to open new possibilities for the development of novel applications that were previously limited by memory constraints. As embedded systems play an increasingly significant role in various domains, such as robotics and the Internet of Things, the project's outcomes will lead to improved user experiences and have a large social impact by making low-cost embedded systems products more accessible. Moreover, this project aims to cultivate a pipeline of skilled engineers and computer scientists with interdisciplinary expertise in embedded systems. Graduate and undergraduate course modules will be developed, benefiting students in all engineering disciplines in Massachusetts and California. Efforts will be made to recruit underrepresented students through diversity programs and outreach initiatives to K-12 students. This inclusiveness will contribute to a diverse and dynamic community of researchers and innovators in the field of embedded systems. 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.