Co-Simulation Framework for Parallel DNN Execution on Chiplet-Based Systems (UW–Madison, Washington State)

Researchers from the University of Wisconsin–Madison and Washington State University have published a technical paper introducing "CHIPSIM," a co-simulation framework for deep learning on chiplet-based systems. Traditional monolithic chips struggle to meet the demands of data-intensive applications like deep neural networks due to manufacturing challenges, leading to the adoption of chiplet-based architectures. CHIPSIM aims to provide a fast and accurate simulation approach for parallel DNN execution on chiplet-based systems, addressing the limitations of existing methods. The framework models computation and communication concurrently, offering improved accuracy, speed, and flexibility, along with power and thermal analysis capabilities.