SENNA – Spiking Neural Network Inference Accelerator

Spiking neural networks (SNNs) are paving the way for the next step in the development of artificial intelligence: even faster, even more energy-efficient, even closer to the way the human brain processes information. Similar to human thinking, information processing in SNNs is based on electrical impulses (spikes) that chase through an artificial network of neurons at top speed.

In simulations, SNNs are able to solve cognitive tasks in nanoseconds. However, anyone who sets out to bring this speed into application is quickly faced with the question: What hardware can keep up with this speed? The answer: only one that is also inspired by the structure of the brain. What is needed are neurons cast in silicon. And that is exactly what our Spiking Neural Network Inference Accelerator SENNA contains.

SENNA is a programmable, neuromorphic chip for ultra-fast and energy-efficient processing of low-dimensional time series in AI applications. Its fully parallel processing architecture – in the current design consisting of an array of 1024 artificial neurons – allows the native integration of spiking neural networks. SENNA works with spike-based input and output signals and thus can be directly embedded in an event-based data stream.

The SENNA reference design is designed for 22 nm semiconductor processes. This means that the SNN accelerator can be used as a chip in a wide range of applications and can be implemented cost-effectively.

SENNA is a programmable chip. The AI model to be executed can be adapted and reimplemented on the ASIC. Even completely new applications can be deployed on an existing chip.

To make this as easy as possible, we provide a software development kit for SENNA. This provides developers with a simple tool to get SNN models up and running on SENNA quickly and easily.