NEUROTEC - Neuro-inspired artificial intelligence technologies for the electronics of the future

While von Neumann computers (classical computers) perform numerical calculations extremely fast and with low energy consumption, the human brain as a biological neural network (BNN) is much more efficient in cognitive tasks such as pattern recognition. The energy consumption of the brain - in real-time pattern recognition - is only 1/10,000 compared to the energy consumption of a classical computer, for the same task. These significant differences in performance on different tasks are related to fundamental differences in the structure of the human brain and a classical von Neumann computer architecture.
The limitation of classical von Neumann computers with respect to fast and energy-efficient processing of large amounts of data, as required for future electronics in the field of artificial intelligence (AI), lies in the concept of hardware-implemented separation of logical processing unit and peripheral memory blocks, as well as the resulting constant data transfer via BUS systems.

This project addresses two "Beyond von Neumann" concepts that are the subject of intensive international research. These are, on the one hand, "computation-in-memory" concepts (CIM: computation-in-memory), in which computational operations are performed directly in memory matrices, thus avoiding a significant portion of the data transfer that occurs in classical systems. Secondly, hardware components for use in artificial neural networks (ANNs) are to be researched. In ANNs, the biological model of neurons and synapses is imitated in an electrical circuit. Here, no classical logic is used for computation, but rather the interconnection of neurons. This design achieves massive parallelization while avoiding energy-intensive and time-consuming data transfer.

The planned project clearly distinguishes itself from both purely software-based AI concepts and CMOS hardware-based AI concepts, since both approaches will be limited in the long term by excessively high power requirements and insufficient data transfer rates of classical computer architectures.
Revolutionary approaches for "Beyond von Neumann" computers require a complete rethinking of standard CMOS semiconductor technology and the classical computer architectures based on it. This revolution includes

the use of new materials and material stacks in new components,
their integration into dense networks with high connectivity that continue to be driven by CMOS circuits.
the switching behavior must be physically understood in order to perform component optimizations and develop compact models for use in circuit simulators.

Chip fabrication of neuro-inspired components and circuits requires solutions to all three categories and also assume that

test systems and test protocols exist to electrically characterize the novel components and circuits, which is not the case to date.
Realizing the potential of neuro-inspired technologies requires the development of novel concepts for neuro-inspired circuit designs.

NEUROTEC is funded by the German Federal Ministry of Education and Research (BMBF).

Project partners:

Electronic Materials (PGI-7), research centre Jülich, Jülich, Germany

JARA-Institute Energy-efficient information technology (Green IT) (PGI-10), research centre Jülich, Jülich, Germany

Chair of Integrated Digital Systems and Circuit Design (IDS), RWTH Aachen University, Aachen, Germany

Chair of Semiconductor Electronics and Institute of Semiconductor Electronics , RWTH Aachen University, Aachen, Germany

Institute for Communication Technologies and Embedded Systems, RWTH Aachen University, Aachen, Germany

Compound Semiconductor Technology, RWTH Aachen University, Aachen, Germany

Lehrstuhl für Werkstoffe der Elektrotechnik II und Institut für Werkstoffe der Elektrotechnik, RWTH Aachen University, Aachen, Germany

Associated partners:

aixACCT Systems GmbH, Aachen, Germany

AIXTRON SE, Herzogenrath, Germany