Fraunhofer Institute for Electronic Microsystems and Solid State Technologies EMFTFinding out whether a neuromorphic processor suits your application usually means building everything at once: the spike encoding, the network model and the system interfaces. SENNA Connect removes that hurdle. The project develops a developer kit around the SENNA processor architecture so that researchers, technology developers and industrial users can run spiking neural networks on real hardware instead of in simulation only.
SENNA Connect - a developer kit for spiking neural networks
Test spiking neural networks on real neuromorphic hardware
Spiking Neural Networks (SNN) for a neuromorphic processor
What the developer kit includes
The first SENNA proof of concept demonstrated the architecture. SENNA Connect turns it into a platform you can work with.
Redesigned SENNA processor
Built for laboratory use, with configurable inputs, outputs and timing parameters.
Hardware platform with FPGA
Connects to FPGA development boards via an FMC connector for signal processing that is not yet in an ASIC.
Configurable spike input
Generate spike streams or convert analogue and digital data into spike representations.
Analogue and digital interfaces
Connect converters, sensors and measurement equipment to the processor.
SENNA software suite
Map, compile and run spiking neural networks, with the hardware behaviour visible during development.
Documentation and workflows
Validation workflows so that results are reproducible outside our own lab.
Prototype the signal path before you commit to silicon
The step that usually decides a neuromorphic project is the one between the sensor and the processor. With the FPGA platform you can emulate that step instead of taping out a front-end for every experiment. This approach allows users to explore spike encodings and system architectures before committing to application-specific integrated front-ends.
Once an encoding proves itself, it can move into an application-specific front-end. How those circuits are designed is described on our expertise about neuromorphic hardware.
The FPGA platform can support:
- generation of spike streams for SENNA,
- conversion of analogue or digital input data into spike representations,
- emulation of application-specific Spike Encoding Front-Ends,
- connection to analogue-to-digital converters and sensor interfaces,
- rapid prototyping of sensor-to-SENNA signal paths,
- real-time interaction between conventional digital processing and neuromorphic processing.
What Fraunhofer EMFT contributes
SENNA Connect is a joint project with Fraunhofer IIS. We contribute to the hardware development and test, from the processor redesign to the measurements that tell us whether the kit works in someone else's hands.
Processor redesign
Adapting the proof of concept for developer-kit operation, flexible inputs and outputs and adjustable neuron behaviour.
Hardware development
Board design and integration of signal sources, measurement equipment and digital control functions.
Software integration
Bringing the kit into the SENNA software suite so that real circuit behaviour reaches training and validation.
Test and user feedback
Characterisation with internal and external users, feeding requirements into the next SENNA generation.
More application areas for spike-based processing you find in our research area for neuromorphic computing.
The project is funded by the German Federal Ministry of Research, Technology and Space under funding reference number 03VP13121.