Efficient Acceleration of Stencil Applications through In-Memory Computing.

摘要: The traditional computer architectures severely suffer from the bottleneck between processing elements and memory that is biggest barrier in front of their scalability. Nevertheless, amount data applications need to process increasing rapidly, especially after era big artificial intelligence. This fact forces new constraints architecture design towards more data-centric principles. Therefore, paradigms such as in-memory near-memory processors have begun emerge counteract by bringing closer computation or integrating them. Associative are a promising candidate for computation, which combines processor same location alleviate bottleneck. One iterative huge stencil codes. Considering this feature, associative can provide paramount advantage For demonstration, two 2D codes proposed, implemented both emerging memristor SRAM technologies. proposed achieves efficiency variety thus proves its applicability scientific computing.