Guide to Deploying Edge AI Environments Using DEEPX DX-M1 NPU and AWS IoT

Combining DEEPX's DX-M1 NPU, which delivers high-performance, low-power inference in edge environments with limited power and heat dissipation, with AWS IoT services can significantly reduce the operational burden of deploying AI models across multiple field sites.

Model Preparation: Converting ONNX to DXNN

In the cloud, uploading the ONNX model and compilation settings to Amazon S3 automatically triggers an event-based workflow. AWS Lambda and Step Functions temporarily launch EC2 instances based on the DEEPX Compiler AMI to convert the model into DXNN, the NPU execution format. After compilation is complete, the EC2 instance terminates, and the results are stored in S3.

Edge Deployment: Zero-Touch Provisioning

Devices are grouped into Thing Groups via AWS IoT Core and Greengrass, and runtimes are deployed using the Zero-Touch Provisioning (ZTP) method. This approach automatically installs the NPU driver, dx_rt, firmware, and dx_stream over-the-air (OTA), allowing consistent management of the execution environment at the group level without direct access to individual devices.

Subscribing to the solution on AWS Marketplace allows you to configure both the cloud compilation infrastructure and the edge deployment environment with a single CloudFormation stack.