Integrated AI for Sensor Systems

Today, intelligent sensor systems perform many different tasks, including speech recognition, intelligent heating control, or autonomous driving functions. By integrating artificial intelligence into sensors, sensor data can be interpreted to enable specific responses or to relieve the human operator of tasks.

Cloud-based technologies offer high accuracy of evaluation and many different functions, which is why these systems are currently in most widespread use. However, it worth remembering that cloud solutions do have disadvantages, especially with regard to energy consumption, data security, and speed of response.

Local AI data processing on embedded platforms, on the other hand, offers an alternative that does not have these drawbacks. Whether or not artificial neural networks or conventional methods of machine learning deliver better data processing results very much depends on the task and the measurement data available.

Get an insight into integrated AI for sensor systems

Description of the technology

The technologies of Fraunhofer IIS go one step further. The functions for signal acquisition and AI-based signal evaluation are combined in an application-specific integrated circuit (ASIC).

With this approach, the circuit components can be optimally harmonized, resulting in a substantial reduction in the power consumption of the entire system.

Technology benefits

Applications

Applications for integrated AI are found wherever battery life and data security are important and signals are available in the Hz or kHz range. In the smart home, that may be non-cloud voice control or a smart window sensor that signals not only the open position but also whether it is raining or the windowpane is being broken. For Industry 4.0, old production plants that still function can be monitored with wireless retrofit sensors in order to optimize maintenance and avoid production outages.

Our offering

At the Fraunhofer IIS Competence Center, we support our customers along the entire development chain from the idea to the machine learning ASIC.

Our experts from Analytics, Data, and Applications provide support with problem analysis, data acquisition and feasibility studies.

Implementation of the solution as embedded electronics is the focus of our embedded ML activities. On this basis, we offer implementation on the platform that is ideal for the customer.

To minimize power consumption and installation space, development of an ASIC is the alternative to embedded electronics. Already completed previous work is incorporated and shortens the development process. Depending on the customer's problem, we partition the system into analog and digital circuitry and software, and optimize the energy consumption, costs, installation space, performance, and security depending on the requirements.