Additionally, the rapid rise of deep learning and artificial intelligence (AI), driven by large language models (LLMs) like ChatGPT, requires unprecedented computing power. While deep learning and AI offer numerous benefits, they also pose new challenges. This insatiable demand for computation translates into equally vast energy consumption, as the data centers powering these models are projected to consume up to 9% of the total electricity generated in the United States by 2030.

A new era of chip design is needed for the Age of AI and LLMs. Existing solutions, such as CPUs and GPUs, will not be sufficient as scalable and affordable solutions. New domain-specific chips and architectures will need to be designed in order to meet the rigorous demands of deep neural networks and the computationally intensive tasks of model training and inference.

Professor Louri and his team investigate novel parallel computer architectures and technologies that deliver high reliability, high performance, and energy-efficient solutions to essential application domains and societal needs for highly parallel systems, ML, and AI applications. The research has far-reaching impacts on the computing industry and society at large. Current research topics include: (1) design of computer architecture and high-performance computing with emphasis on energy efficiency, reliability, performance scalability, and security; (2) design of scalable, power-efficient, reliable and secure Network on Chips (NoCs) for parallel computing systems and AI accelerators; 3) hardware-software co-design of reconfigurable, adaptable and reliable deep neural networks (DNNs) and graph neural networks (GNNs) accelerators for various AI applications; 4) design of emerging interconnect technologies (optical, wireless, RF) for multicores, AI accelerators, and data center architectures; 5) exploration of approximate computing and model reductions techniques for large-scale and energy-efficient LLMs; 6) exploration of hybrid technologies (optical, electrical) for sustainable computing.