Matrix Processing Engine architecture (MPE)
Deep Learning; Computational Infrastructure
noun phrase
Definition: A hardware accelerator architecture based on a multidimensional array of multiply–accumulate (MAC) units designed to execute matrix operations efficiently, especially for convolutional neural networks and related deep-learning workloads [Sun et al. 2018].
Example in context: “To maximize the benefits of sparsification, we design the unified Matrix Processing Engine (MPE) to handle all operations related to matrix computation…” [Zeng et al. 2024]
Synonyms: MPE; matrix processing engine
Related terms: MAC array; CNN accelerator architecture; matrix accelerator; processing-in-memory accelerator