SENNA: An Ultra-Low Latency, Time-Deterministic Field-Programmable-Spiking-Neuron-Array (FPSNA)
SENNA is a first-of-its-kind mixed-signal circuitswitched field-programmable accelerator for Spiking Neural Networks (SNNs). We demonstrate a 1024-neuron and 25.6k synapse proof-of-concept chip in 22 nm FDSOI that achieves 687 fJ energy per synaptic operation (E/SOP) at 25 MHz and 23 mW. The IC allows a very flexible configuration of parallel processed SNNs. Due to its low power draw, low spike-in-spikeout (SISO) latency of 40 ns and time-deterministic response, this design is particularly promising for real-time edge applications.
Authors
- Sebastian Ruping
- Claudia Schuhmann
- Stefan Köhler (ORCID: https://orcid.org/0000-0003-1365-7506)
- Johannes Leugering (ORCID: https://orcid.org/0000-0003-0956-4139)
- Ferdinand Pscheidl
- Elmar Herzer
- Ralf Brederlow (ORCID: https://orcid.org/0000-0002-8118-757X)
- A. Stoll
- L. Richter
- Matthias Kampa
Institutions
- ZHAW Zurich University of Applied Sciences (CH)
- UC San Diego Health System (US)
- University of California San Diego (US)
- Fraunhofer Institute for Integrated Circuits (DE)
- Fraunhofer Research Institution for Microsystems and Solid State Technologies (DE)
- Technical University of Munich (DE)
Publication Details
- Published
- 2025-09-08
- DOI
- https://doi.org/10.1109/esserc66193.2025.11214098
- Primary Topic
- Advanced Memory and Neural Computing
- Type
- article
- Field-Weighted Citation Impact
- 0.00