Fraunhofer ITWM at the »Embedded World«

What To Do with All the Data?

Press Release / April 04, 2024

The amount of data produced worldwide continues to increase. This also increases the need for efficient data processing in order to reduce energy and infrastructure costs. The Fraunhofer Institute for Industrial Mathematics ITWM is working on a hardware-aware NAS (Neural Architecture Search) that brings together several promising technologies. At the »Embedded World« trade fair in Nuremberg, researchers from Fraunhofer ITWM will present their work at the joint Fraunhofer booth (Hall 4, Booth 422) from April 9 to 11, 2024.

At the leading international trade fair for embedded systems, the ITWM team will be demonstrating how it combines the trending topics of edge computing i.e. the decentralized processing of data »at the edge of the network« and machine learning. Here, the researchers are focusing primarily on deep neural networks (DNNs), which are very flexible and process a wide range of inputs, e.g. time series data and images.

Attempts to transfer off-the-shelf DNNs to an embedded system often fail because the models are too large if they have not been specially developed for the borderline case. This is where the Fraunhofer ITWM's solution comes in: the Neural Architecture Search Engine (NASE) searches for a DNN structure so that the DNN fits the hardware and fulfills the application goals.

Scalable Method for Fast Solutions

»With NASE, we at Fraunhofer ITWM are developing a hardware-aware NAS,« explains project manager Dominik Loroch. »This type of search also takes into account the limitations of the hardware platform and we find solutions that run optimally. Those who want to use NASE only provide the data and need little to no knowledge of DNNs. Since both the search and the training are automatic, it's a very scalable way to quickly get solutions for the hardware.« NASE acts as a productivity booster that identifies complex DNNs very quickly and thus significantly reduces development costs.

Award for AI Hardware

The basis is an energy-efficient AI chip for which researchers at Fraunhofer ITWM received the Pilot Innovation Award from the Federal Ministry of Education and Research in 2021. »Back then, a new standardized methodology was developed to increase the energy efficiency of AI systems. At the same time, our methodology reduced the development time for optimized neural networks and corresponding FPG implementations,« says Loroch. »We have continuously developed the expertise we acquired back then and can now transfer it to different industrial scenarios.«

Part of the Joint Fraunhofer Booth

The Fraunhofer ITWM can be found in hall 4, booth 422 at the Fraunhofer joint booth. There, other Fraunhofer Institutes will present current research work and solutions for embedded systems.