Peter Grünberg Institute (PGI)

Integrated Computing Architectures (PGI-4)

Neuromorphic Edge Computing Systems

As the core research topic of the Neuromorphic Computing group in ICA, we research and develop CMOS-based systems that employ bio-inspired computing paradigms. A focus is laid on the solution's scalability that leverages the maximum potential of emerging devices and novel algorithms, especially for edge computing applications where power consumption, compactness, and costs are typical constraints. Hence, integrated circuits are at the core of these developments.

RECENT TAPEOUTS

Technology Demonstrator for Memristor-CMOS cointegration

DO

Dolphin-DAD

Dolphin-SNN

6x6 mm 28nm CMOS Chip
Integrated CMOS structures for 1T1R, 2T1R memristor units
Integrated Content Addressable Memory units and arrays
Integrated Compute in Memory unit and arrays
VCM and PCM mode memristor units

1.4 x 1 mm 28nm CMOS Chip in QFN 40
Integrated Memristor Control circuit HRS, LRS and read for 2 x 2 arrays
Current mode SAR ADC with adaptive Dynamic range
Integrated RISC-V and memory for error correction and control

1.4 x 1 mm 28nm CMOS Chip in QFN 40
7 coupled PLLs for Oscillatory Neural Network
Analog and multilevel phase coupling
Scalable ONN with all to all and multi cluster coupling

Contacts

Dr. Andre ZambaniniScientific CoordinatorBuilding 02.5 / Room 235+49 2461/61-96916

Christian GrewingScientific CoordinatorBuilding 02.5 / Room R 132+49 2461/61-96430

Sabitha KusumaScientific CoordinatorBuilding 02.5 / Room E1+49 2461/61-96750

Dr. -Ing Arun AshokScientific CoordinatorBuilding 02.5 / Room 202+49 2461/61-6542

Neuromorphic Systems with Novel Computing Paradigms and Emerging Devices

Comparing the biological efficiency of brains with the calculation efficiency of traditional, von Neumann computing architectures yields a vast difference, especially for certain computational tasks important for feature recognition and AI applications. Inspired by evolved biological principles, the research includes investigation on new computing approaches from physical and mathematical principles (e.g. ANN, SNN, ONN). However, mapping these approaches on conventional hardware loses some of the benefits these paradigms provide. Emerging devices, like memristors, are extremely promising but need to be combined with the well-established benefits of modern CMOS-based technologies to keep the potential of all components of system.

Hence, our group focuses on integrated circuit systems that leverage the maximum potential of bio-inspired computing paradigms and elements on a system level. With this research, we provide a crucial contribution to lift fundamental research into a higher readiness level. It allows us to benchmark the systems in a laboratory environment and bring it closer to commercialization.

At its core, we are using commercial CMOS technologies to build the foundation of integrated circuits. This is then enhanced by emerging devices, either from internal, yet experimental post-processing steps or from other available prototype lines. The resulting complexity of the systems requires a contribution from several interdisciplinary experts. Therefore, projects in this field will most likely be part of bigger endeavors, spanning over several groups with a significant contribution to integrated circuit design from our team.

Main activity

NEUROTEC II

The BMBF-funded NEUROTEC project is at the forefront of advancing high-tech capabilities and fostering structural transformation in the digitalization sector.

MORE