CSR: Small: Cross-Layer Solutions Enabling Instant Computing for Edge Intelligence Devices

点击翻译按钮获取中文摘要

Cloud-based computation and communication encounter several severe challenges, including high latency, questionable scalability, quality of service, privacy, and security. These issues can be addressed by transitioning computing architecture from a traditional cloud-centric mindset to a thing-centric (data-centric) perspective. However, nearly 90% of the data generated by the Internet of Things (IoT) is not analyzed using the conventional mechanism, due to limited computing ability and constraints in power and area. Hence, resource-limited IoTs should be developed optimally to enhance overall performance and extend lifetime. Our objective in this project is to transition from the conventional paradigm to a Sense-Decide-Action mechanism, which can analyze data locally, autonomously, and sustainably instead of sending it to the cloud, thereby reducing the amount of communication via a new cross-layer co-design approach. The broader impact and significance of our project lie in systematically paving the way for innovative foundational computing schemes. These strategies enable efficient instant computing and the necessary design approaches for real-time processing and decision-making systems. This progress brings us closer to the reality of accommodating over a trillion interconnected devices and improving the data privacy of resource-limited IoTs for critical applications, including healthcare monitoring, automotive applications, and industrial sensing. The technical approach of this research revolves around the development of low-overhead strategies to accelerate edge intelligence sensory nodes' autonomous operation within an environment. This is achieved by (1) designing and analyzing non-von-Neumann architectures, co-integrating conversion and processing capabilities in conjunction with an unconventional number system to alleviate the existing data movement issue between off/on-chip and processor; and (2) implementing processing units for both generic computations and domain-specific and emerging applications. This automated process of architecture engineering creates search space, defines design strategy, and identifies the optimal architecture to improve metrics such as lifetime energy reduction and overall performance. We evaluate the functionalities and performance of the proposed Thrusts through extensive modeling and simulations, beginning from circuit-level design and progressing upwards. This research effectively establishes novel lightweight heterogeneous edge intelligence capable of autonomously processing various data and compute-intensive tasks in an energy-scarce environment, marking a significant step towards the efficient future of IoT applications.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.