Algorithm-Hardware Co-Design in Computing Systems: From Embedded Systems to the Cloud

In this paper, we propose algorithm-hardware co-design methods for computing systems, from the embedded systems level to the datacenter level, to bridge the affordability compute-demand gap between software and hardware. At the embedded systems level, we propose hardware (HW) / software (SW) co-design methods under increased uncertainty in design parameters. Our work lays the foundation for uncertainty modeling and robust multi-objective optimization for embedded systems. Specifically, it provides computer-aided tools for solving the problem of mapping that incorporate novel design methods to achieve robust high performance embedded systems design. At the datacenter level, we propose algorithm-hardware co-design that leverages deep learning methods to reduce energy consumption at both chip multiprocessor (CMP) / server and datacenter levels. Our work provides advances in computing systems from the embedded systems level to the datacenter level.