Artificial intelligence, and machine learning specifically, is transforming industries from Consumer to Industrial. To date, many applications host AI/ML inferencing on conventional computers in the cloud or locally. Meanwhile, edge computing is enabling other computing workloads to move from conventional information technology (IT) to lower-cost systems close to where data is generated. Although many AI/ML workloads run fine on edge systems’ CPUs, others are more intense: either multiple AI/ML functions must run simultaneously or performance requirements (e.g., frame rates) are too great. The solution to gaining the combined benefits of AI/ML and edge computing is acceleration.
At the 2020 Consumer Electronics Show, NXP demonstrated the LS1046A-FRWY platform simultaneously running two or more high-intensity AI/ML functions. These include face recognition, object detection (both general and safety gear), posture recognition, and gaze detection. The scenario demonstrated is factory safety. An operator within a safety zone is monitored for attentiveness, personal protective equipment, and access control. Helping to make this possible is external acceleration based on the Google Edge TPU. Interfacing to the Layerscape LS1046A processor via its copious PCI Express ports, two M.2 TPU cards slotted in the FRWY system offload AI/ML inferencing. Based on the Layerscape LS1046A processor with four powerful Arm Cortex-A72 CPU cores, the compact, cost-effective LS1046A-FRWY platform gives developers a leg up on implementing high-performance AI/ML applications at the edge.
|LS1046A Freeway Board | NXP|