Two papers accepted to MICRO 2026

Two papers were accepted to the 59th IEEE/ACM International Symposimum on Microarchitecture (MICRO) 2026. One paper proposes a fault entry merging mechanism for efficient batching in unified virtual memory (UVM) of GPU. Our proposed merging scheme enables direct fault entry merging on GPU without intervention from the host. Another paper presents Complex Tensor Core, an efficient Tensor Core architecture for accelerating complex-valued neural networks (CVNNs) on GPU. Complex Tensor Core combines real and imaginary part products into a single operation on Tensor Core.