Neuromorphic AI Chip Boosts Spiking Neural Networks

An innovative inference accelerator chip tailored for processing spiking neural networks (SNNs) has been created by the Fraunhofer Institute for Integrated Circuits IIS. Named SENNA, this chip is designed to facilitate the rapid processing of low-dimensional time series data in artificial intelligence (AI) applications. Drawing inspiration from the human brain, SENNA is constructed with artificial neurons and synapses, allowing it to directly handle electrical impulses known as spikes. Its exceptional speed, energy efficiency, and compact form factor make it ideal for deploying SNNs at the edge, where data is generated. The development of SENNA is a part of the Fraunhofer SEC-Learn project.

At present, SENNA houses 1,024 artificial neurons within a chip area of less than 11 mm². Boasting a swift response time of as low as 20 nanoseconds, the chip ensures precise timing, particularly in time-critical applications at the edge. Its capabilities shine in real-time assessment of event-based sensor data and closed-loop control systems, such as in the management of small electric motors with AI. SENNA can also facilitate AI-optimized data transmission in communication systems, enabling the AI processor to analyze signal streams and adjust transmission and reception procedures to enhance efficiency and performance.

One of the key reasons behind the energy efficiency of SNNs is that neurons activate sparingly and in response to specific events. Leveraging spiking neurons, SENNA maximizes this energy-saving attribute. Its fully parallel processing architecture enables the artificial neurons to accurately capture the temporal behavior of SNNs. Additionally, SENNA can directly interact with spike-based input and output signals through its integrated spike interfaces, seamlessly integrating into an event-based data stream.

Michael Rothe, Group Manager Embedded AI at Fraunhofer IIS, highlights, “With its innovative architecture, SENNA effectively addresses the trade-off between energy efficiency, processing speed, and versatility, making it an ideal edge AI processor for resource-constrained applications requiring ultra-fast response times in the nanosecond range.” The current reference design of SENNA is tailored for 22 nm manufacturing processes, allowing the SNN processor to be integrated into various applications cost-effectively. Its scalable design can be customized to meet specific application needs, performance criteria, and target hardware features before chip production.

Even post-manufacturing, SENNA offers maximum flexibility as it is fully programmable. The SNN model utilized can be modified and retransferred to SENNA as needed. To simplify the implementation of AI models for developers, Fraunhofer IIS provides a comprehensive software development kit for SENNA, ensuring a user-friendly experience for leveraging the chip’s capabilities.