Hussein Osman, Consumer Segment Manager at Lattice Semiconductor, presents the “Accelerate Adoption of AI at the Edge with Easy to Use, Low-power Programmable Solutions” tutorial at the May 2019 Embedded Vision Summit.
In this talk, Osman shows why Lattice’s low-power FPGA devices, coupled with the sensAI software stack, are a compelling solution for implementation of sophisticated AI capabilities in edge devices. The latest release of the sensAI stack provides a performance increase of more than 10X compared with the previous release. This performance increase is driven by updates to the CNN IP and the neural network compiler tool, including a number of new features, such as support for 8-bit activation quantization and smart merging of layers.
For a seamless user experience, the new release expands the list of neural network topologies and machine learning frameworks supported, and automates the quantization and fraction settings processes. In addition, Lattice Semiconductor provides a comprehensive set of reference designs which include training datasets and scripts for popular machine learning frameworks, enabling easy customization. And, to speed time to market for popular use cases, the company provides full turnkey solutions for human counting, human presence detection and key phrase detection.