SPRY344A January   2022  – March 2023 TDA4VM , TDA4VM-Q1

 

  1.   At a glance
  2.   Authors
  3.   Introduction
  4.   Defining AI at the edge
  5.   What is an efficient edge AI system?
    1.     Selecting an SoC architecture
    2.     Programmable core types and accelerators
  6.   Designing edge AI systems with TI vision processors
    1.     Deep learning accelerator
    2.     Imaging and computer vision hardware accelerators
    3.     Smart internal bus and memory architecture
    4.     Optimized system BOM
    5.     Easy-to-use software development environment
  7.   Conclusion

What is an efficient edge AI system?

In an efficient edge AI system, DNNs cannot operate by themselves. An efficient AI system requires a complex vision pipeline, often including single or multi camera image processing, traditional computer vision and maybe even multiple DNNs. Some applications may also need video encoders and decoders. To process all of these inputs, a system needs high-performance computing. In addition, a system may require enhanced security and functional safety, increasing system complexity and cost.

An efficient edge AI system should be optimized for:

  • Performance. The embedded processor must be able to deliver the speed, latency and accuracy that the system requires while also functioning reliably, even in harsh environments.
  • Design constraints. The embedded processor must operate in designs with power and thermal constraints, including designs that are fanless, have passive cooling or need to operate for longer hours on battery power. The processor must also meet size and weight specifications to comply with physical constraints.
  • Cost. Enabling processing that is high-performance and cost-effective will yield the lowest possible bill-of-materials (BOM) cost.

To build an efficient edge AI system, designers should consider which architecture and cores will best complete the tasks required of the system.