ETAS Embedded AI Coder

Embedded AI Coder is a code generator tool that effortlessly converts trained neural networks into optimized C-code for a variety of microcontroller and microprocessors. Our tool ensures exceptional speed and efficient memory usage, empowering developers to seamlessly integrate AI capabilities into their products.

Your benefits

Any hardware, any tool chain

No need for special libraries and hardware accelerators, Embedded AI Coder generated code can be deployed to any device.

Resource-efficient and cost effective

Minimizing resource consumption on highly constrained embedded systems.

Compliant with relevant safety requirements

Compliant with automotive and functional safety requirements such as ISO26262 and MISRA.

Features

Ease of Use

Supported formats .ONNX, .LiteRT (Tensor Flow Lite), ASCMO export
Command line tool, as simple as using a compiler.
e.g. coder code myaimodel.tflite -c myoptions -o <target folder>
Generated code with additional artefacts to integrate in any toolchains.
Easy integration into any CI/CD pipelines.

Broad neural network and hardware support

The coder works for various target hardware and is compatible with commonly used layers of neural networks: Batchnorm, Convolutions, Depthwise Convolutions, LSTM, Fully Connected, Elementwise Add, Sub, Mul, Softmax, Relu, Leaky Relu, Logistic, Padding, StridedSlice, Tanh, MaxPooling, AveragePooling and TransposeConv. It supports the data types int8 and float32.
Hardware targets range from tiny microcontrollers to powerful microprocessors with various architectures (ARM Cortex-M, ARM Cortex-A, vendor-specific architectures).

Wide range of possible applications

Our coder is ideally suited for smart sensor and smart actuator systems, safety-critical systems in the automotive, medical technology, and aerospace sectors, and AI on small hardware systems.
You can use it for cost-effective µC targets and complex, safety-critical applications on embedded devices.

High performance and resource efficiency

Bosch submitted results at the MLPerf™ Inference Tiny benchmark
Among state-of-the-art results in all categories
Best-in-class memory efficiency

See Benchmark portal with comparable tools and background information for comparison.

Safety compliance

The Embedded AI Coder complies with all automotive and functional safety requirements, such as ISO26262 and MISRA.
It generates secure, numerically correct code free of runtime errors.

Making embedded AI a reality now

Graphic shows workflow of embedded AI coder.

Embedded AI Coder builds a bridge between the AI and embedded world, making it possible to generate embedded C-code without the help of embedded software experts.

Models from well known machine learning framework like pytorch , tensor flow or ETAS ASCMO can be used as input to our tool.

In contrast to handwritten code, it saves companies high development costs and resources. This is particularly advantageous in view of today's typical rapid development cycles and frequently changing code requirements.

The generated code can be deployed on production hardware or tested early on ETAS Rapid Prototyping toolchain using INTECRIO, EHOOKS or virtual ECUs using VECU-BUILDER.

Client use cases with our Embedded AI Coder

In the automotive industry:

Cost savings through virtual sensors (i.e., AI algorithms that replace physical sensors) in braking systems, steering systems, engine management, etc.
Tire pressure monitoring
Early damage detection
Driving assistance systems, for example ultrasonic parking sensors
Driver monitoring systems (mandatory by law from July 2024)

Our software tool, validated for the automotive industry, can also be adapted for use in other industrial domains upon request. We are committed to conducting the necessary regulatory checks to ensure compliance and successful implementation in your specific sector.