3D Network-On-Chip Data Acquisition System Mapping Based on Reinforcement Learning and Improved Attention Mechanism

Abstract The 3D Network-on-Chip (NoC) data acquisition system utilizes NoC technology to establish a time-interleaved data acquisition system The mapping scheme determines the location of each Intellectual Property (IP) node in the NoC topology. The optimization of the mapping algorithm is one of the important means to reduce the communication delay of the acquisition system. The abundance of functional IP nodes in the 3D NoC data acquisition system creates a mapping challenge. To address this, we propose a mapping algorithm called Reinforcement Learning and an improved Attention Mechanism Mapping algorithm (RA-Map). The RA-Map mapping algorithm employs node function encoding and node position encoding to express the properties of an IP node in the task graph preprocessing. The local attention mechanism is used in the mapping network encoder, and the fusion of dynamic key node information is proposed in the decoder. The mapping result evaluation network achieves unsupervised training of the mapping network. These targeted improvements ultimately lead to an enhancement in mapping quality. Experimental results demonstrate that when compared to the discrete particle swarm algorithm and simulated annealing algorithm, the RA-Map mapping algorithm reduces the average communication cost by 6.5% and 8.5%, respectively. Furthermore, while ensuring mapping quality, it also shortens the mapping time.