Neural Networks

The Competence Microelectronics Intelligence builds intelligent sensors by integrating advanced micro/nano sensors with artificial intelligent processing by providing reliable neural network circuits built in integrated analog hardware for real-time sensory applications.

Memory modules used for high performance electronics are being optimized for use in trainable artificial neurons. Circuit interfaces to these memory modules have been developed to build analog neurons. Large scale networks of these neurons are currently being investigated to build an integrated circuit along with several hidden layers consisting of modules like convolution and deconvolution to produce an array of neural networks with capabilities of learning in different systems. Additionally, we are also working to increase the reliability of manufacturing of large arrays of neuron-type circuits in a single chip by undertaking development in a vertically integrated environment of design, fabrication, and testing at the same site.

In parallel, we are also integrating sensors particularly CMOS image sensors with these neural networks. By integrating sensors with brain type processor-memory systems we aim to eliminate the sensor-processor and sensor-memory bottleneck present in modern high performance yet cheap arrays of sensors. This leads to intelligent and adaptable sensory systems, casting an integral part of the human brain, its sensory systems, into physical electronics. By building in analog hardware and aiming for a stack of few microelectronic chips, we provide a low power, real-time and low-cost solution to a range of sensory challenges for the internet of things and industry 4.0 era.
The team works closely with Siegen University and is funded by the Fraunhofer Attract scheme and ECSEL projects.