The e-AI is solution to enable the Artificial Intelligence(hereafter AI) technology to embedded systems.
AI consists of “Training” and “Inference”.
Renesas‘ e-AI solution enables the use of AI by running only “inference” on MCUs and MPUs.
Also refer the “What is Renesas’ e-AI Solution?

Image
learning vs interference

Advantage of e-AI

The biggest advantage of e-AI is “real-time processing”
e-AI can judge or response without delay of network, therefore it can get the result of inference faster than cloud case.
e-AI can be used where continuous input data is judged one after another with AI by using this real-time performance.

Renesas will contribute to the realization of an environmentally friendly smart society that supports safer and healthier lifestyles through innovations from endpoint intelligence.

Image
Cloud Endpoint

An example technologies can be supported by AI

  • Anomaly detection
  • Object detection
  • Voice recognition
  • Classification
  • Signal Processing
  • Pose estimation

Refer to the actual application example by videos or partner solutions.

e-AI development environment

Renesas provides the e-AI development environment to realize AI implementation easier for MCU or AI accelerator.
When user inputs the trained AI model into this tool, then the tool converts to a program that runs on the MCU or AI accelerator without any additional operations.

Image
e-AI Tools

e-AI development environment for microcomputer

This tool supports many kinds of Renesas’ microcomputers. The tool can convert from trained model of “PyTorch”, “Keras”, “Tensorflow”, or 8-bit quantized model of “TensorFlow Lite” and import it easily to e2 studio that is Renesas' integrated development environment. These kind of Renesas products are suitable to be used for relatively small-scale AI for endpoints.

e-AI development environment for AI accelerator

This tool supports RZ/V series equipped with the AI accelerator "DRP-AI".
This tool can convert from ONNX format that trained by framework such as “PyTorch” to DRP-AI object code.
These Renesas products are suitable to be used for relatively mid-scale AI for endpoints or edge.

e-AI supported products

Each development environments are supported following products

Supported Products by e-AI Translator

Supported Products by e-AI Translator
RA Family ecosystem partner
RZ/A Series ecosystem partner
RE Family ecosystem partner
RL78 Family ecosystem partner (under preparation)
RX Family ecosystem partner (under preparation)
Renesas Synergy™ Platform ecosystem partner

Supported Products by DRP-AI Translator

Supported Products by DRP-AI Translator
RZ/V series ecosystem partner

Our experiment by Renesas Naka Fab

A demonstration experiment on equipment abnormality detection and predictive maintenance by e-AI was conducted (2015) at the semiconductor front-end factory -"Renesas Naka factory (Renesas Semiconductor Manufacturing Co., Ltd.)"- in Hitachinaka City, Ibaraki Prefecture. New added value, such as abnormality detection and predictive maintenance, can be realized without making major remodeling to existing facilities, while utilizing the existing facilities. In addition, e-AI has greatly improved detection precision with regard to facility abnormalities that previously could only be properly judged by a skilled technician or operator. We have received inquiries from more than 40 customers from Japan and overseas, and started discussions with more than 10 AI-related partners for concretization of business. This example convinced customers that e-AI could contribute to their business and could solve social problems. At the Naka factory, preparations are now under-way for practical use of e-AI for abnormality detection and predictive maintenance.

Plug-ins for e² studio for connecting AI and embedded systems are available.

Image
Renesas NAKA Fab e-AI PoC

descriptionDocumentation

Title language Type Format File Size Date
Application Notes & White Papers
Embedded AI-Accelerator DRP-AI 日本語 White Paper PDF 642 KB
Other
Enhancing Endpoint Intelligence Flyer 日本語, 简体中文 Flyer PDF 1.25 MB

printNews & Additional Resources