I-Corps: Hardware Accelerators for Real-Time Decision Making at the Edge

California Polytechnic State University Foundation, San Luis Obispo CA

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is the development of a computing platform for solving nonlinear optimization workloads at the edge. Nonlinear optimization is a foundational aspect of many critical and emerging technologies such as routing autonomous vehicles through a busy intersection, optimally pricing intermittent renewable energy on the grid, or predicting premature failure of manufacturing equipment. However, as technologies become distributed and decentralized, there is a growing need for these optimizations to be performed at the "edge" — that is, co-located with the physical device that is generating data and needs to be controlled or optimized, and typically on low-power embedded computing hardware, as opposed to a powerful cloud data center. In addition, existing edge devices typically do not have the capabilities to perform these intensive computing workloads, often comprised of complex nonlinear optimization-based processes. The proposed technology is designed to increase computational speeds and energy efficiency by an order-of-magnitude for solving complex nonlinear optimization problems and high-order partial differential equations. This may benefit multiple industries, including transportation, manufacturing, consumer electronics, and energy, and could enable reductions in both capital and operating expenses by more than fifty percent. Moreover, reductions in infrastructure and energy costs may be possible by minimizing the need to move data from the edge, where the data is generated, to centralized data centers, where workloads are typically processed today. This I-Corps project is based on the development of hardware accelerators that increase computational speeds and energy efficiency by an order-of-magnitude for solving complex nonlinear optimization problems and high-order partial differential equations. The hardware accelerator uses mixed-signal computing techniques for control and optimization and is referred to as Analog Neural Computing (ANC), which is a hybrid computing platform that leverages electronic analog computing techniques to solve nonlinear optimization and partial-differential equation workloads substantially faster and more efficiently than existing embedded computing platforms. The proposed technology has been demonstrated in handling certain nonlinear optimization workloads faster and more efficiently than existing state-of-the-art embedded computing platforms. Software and hardware techniques have been developed to maximize the accuracy, speed, and usability of the proposed computing approach. Nonlinear optimization is a crucial technology for efficiently controlling and monitoring a variety of important industrial processes. Recent progress has demonstrated the feasibility of obtaining robust and accurate solutions using these mixed-signal computing techniques, which are naturally subject to several undesired variations and phenomena, such as noise, operating point dependencies, and manufacturing variations. 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.