AI Engine and PL Kernels Data Communication - 2024.2 English

AI Engine Kernel and Graph Programming Guide (UG1079)

Document ID
UG1079
Release Date
2024-11-28
Version
2024.2 English

The AI Engine array interface contains modules to communicate between AI Engines and PL kernels using AXI4-Stream connections. Generally, PL interfaces produce or consume data through stream interfaces. Based on whether buffer or stream data is communicated by the AI Engine kernels, DMA and ping-pong buffers could be involved.

Note that PL kernels run at a lower frequency than AI Engine kernels. Data must cross the clock domains between the AI Engine clock and PL clock. The AMD Vitis™ environment handles the clock domain crossing (CDC) path automatically. It is recommended to run the PL kernel frequency as an integer factor of the AI Engine frequency, if possible. For instance, as ½ or ¼ of the AI Engine clock frequency.