DPU Memory Model - 1.2 English

For edge DPU, Vitis™ AI compiler and runtime N2Cube work together to support two different DPU memory models: unique memory model and split IO model, which are described below. The unique memory model is the default when network model is compiled into DPU kernel. To enable a split IO model, you specify the options --split-io-mem to the compiler while compiling the network model.