Heterogeneous System on Chip

Heterogeneous System-on-Chips (SoCs) integrate diverse processing units to efficiently execute complex tasks, particularly deep neural networks (DNNs), at the edge. Current research focuses on optimizing DNN mapping and scheduling across these heterogeneous architectures, employing techniques like graph neural network architecture search and hardware-aware training to minimize latency and energy consumption while maintaining accuracy. This work is crucial for advancing AI and computer vision applications in resource-constrained environments, such as mobile devices, autonomous systems, and space missions, by enabling efficient and powerful on-device processing. Improved thermal modeling is also a key area of investigation to ensure reliable operation.