Jon Talbot, Field Applications Engineer in the Solutions Group at Synopsys, demonstrates the company’s latest embedded vision technologies and products at the February 2018 Embedded World conference. Specifically, Talbot demonstrates a YOLO (You Only Look Once) v2 CNN graph used for object detection and identification.
Deep learning inference is running on the combination of a DesignWare EV61 embedded vision processor core and a CNN880 neural network coprocessor core, both implemented on a Xilinx FPGA in Synopsys' HAPS-80 ASIC prototyping system. In the FPGA prototype, the Synopsys IP runs at 30 MHz, translating to ~1 fps (frame per second) inference performance. In a 16 nm-fabricated production ASIC, on the other hand, the cores are capable of running at between 800 MHz and 1.2 GHz.
