CAREER: Communication Network for Ultra-Large-Scale Chiplet-Based Systems

As the demand for additional compute power and memory continues to increase, the semiconductor industry is shifting towards ultra-large-scale systems capable of providing orders of magnitude greater compute-memory capacity. Applications such as high-performance compute, neural networks, and large language models, stand to benefit significantly from such systems. Some homogeneous ultra-large-scale systems exist today, however, due to yield challenges and lack of heterogeneity, the industry is clearly shifting towards chiplet-based heterogeneous integration platforms. The technology for chiplet-based systems is becoming rapidly available. However, the design aspects of such systems are yet to be addressed. One of the key challenges of ultra-large-scale systems is the communication among chiplets. To address the communication challenge, a network is required that considers the specificity of the technology, chiplets, and scale of the systems. This project stands to significantly impact the way we design computational systems, stepping away from classic architectures and enabling heterogeneous plug-and-play chiplet-based design. The following are the main thrusts of this project: 1) A communication network architecture for chiplet-based ultra-large-scale systems that includes compatible network topologies and allocation of high-bandwidth domains. 2) Routing and built-in self-test algorithms that will consider the limitations of advanced integration technologies. 3) A unified memory architecture where not only memory is shifted, as is typical in modern architectures, but rather both compute and memory can be relocated to enable efficient computation and enhanced performance. 4) Standards interfacing enabling efficient communication among heterogeneous components such as high-bandwidth memory stacks and compute chiplets. 5) Validation of the proposed topologies, methodologies, and algorithms through circuit design and simulation. 6) A comprehensive education, training, and mentoring plan integrated with the proposed research. 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. NSF Award ID: 2543560 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Boris Vaisband | Institution: University of California-Irvine, IRVINE, CA | Award Amount: $344,624 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2543560 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2543560.html