Google Research

Google Hardware Accelerator Exploration Data


A dataset which contains the latency of running a variety of neural network models (e.g. image classification, object detection, image segmentation, and speech recognition) on a Google template-based hardware accelerator across different architecture configurations.

Each data element of this dataset contains ten accelerator parameters defining the configuration of the hardware accelerator, the latency of running the target neural network model on the accelerator, the estimated chip area usage of the accelerator, and the feasibility of the design point.