Fraunhofer Institute for Photonic Microsystems
Fraunhofer Institute for Photonic Microsystems

HyDRA - Hybrid AI accelerators for neuromorphic computing

HyDRA - Hybrid AI accelerators for neuromorphic computing

Project duration: 2025

© Fraunhofer IPMS
Overview of the hybrid AI accelerator overall innovation. Here, optimized digital-to-analog (D/A) and analog-to-digital (A/D) converters enable the direct connection of the parallel analog core to the digital domain, outlined by the parallel digital input and output of matrix data.
© Fraunhofer IPMS
Ratio of the output rate of an analog core (10k analog output channel) over the conversion rate of the ADCs as a function of the number of integrated ADCs (e.g. ADC array). The color progression shows the different output rates of the individual analog output channels. The dashed line thus marks the optimum balance between output and conversion, everything above this line is in the ADC bottleneck in which the converters cause a data jam.

HyDRA focuses on the development of hybrid AI accelerators that combine the advantages of analog computing cores with optimized digital-to-analog converters (DACs) and analog-to-digital converters (ADCs). In view of the growing requirements for artificial intelligence (AI) processing in modern technologies, the project aims to create an efficient and scalable architecture that enables seamless integration into digital systems.

Project goal

The main goal of HyDRA is to overcome conversion bottlenecks between analog and digital domains. By developing innovative converters that are specifically tailored to the needs of neuromorphic computing cores, efficient signal processing is to be ensured in order to increase the efficiency and speed of AI applications.

Innovations and benefits

HyDRA brings several key innovations with it:

  • Optimized converter architectures: The development of multiplying DACs (MDACs) enables the simultaneous processing of MAC operations (multiplication and accumulation), which increases efficiency.
  • Adaptation to AI requirements: The new ADC designs are specifically geared towards the high requirements of parallel AI computing cores, which reduces latency and optimizes energy consumption.
  • Scalability: The newly developed technologies are designed to meet the increasing demands of the AI market and support integration into existing digital system landscapes.

Project partners:

A collaboration with TileCore, Fraunhofer IIS and Fraunhofer EMFT as part of the “QNC Space” - the Deep Tech Accelerator for research groups, start-ups and SMEs in the field of quantum and neuromorphic computing.

The QNC Space is part of the “Research Fab Microelectronics Germany - Module Quantum and Neuromorphic Computing” (FMD-QNC), a joint project of the 13 FMD institutes, the four Fraunhofer Institutes ILT, IMWS, IOF and IPM, as well as Forschungszentrum Jülich and AMO GmbH.

Funded by

 

more info
 

QNC Space

Deep Tech Accelerator for research groups, start-ups and SMEs

more info

Related Links: