Brain-inspired neuromorphic AI chip accelerates spiking neural networks
An inference accelerator chip designed for processing spiking neural networks (SNNs), called SENNA, has been developed by the Fraunhofer Institute for Integrated Circuits IIS to enable swift processing of low-dimensional time series data in AI applications.
Inspired by the functioning of the brain, SENNA is composed of artificial neurons and synapses and can process electrical impulses (spikes) directly. Its speed, energy efficiency, and compact design enable SNNs to be utilised directly where data is generated: in edge devices. Fraunhofer IIS developed the neuromorphic SNN accelerator, SENNA, as part of the Fraunhofer SEC-Learn project.
In its current iteration, SENNA comprises 1,024 artificial neurons within a chip area of less than 11 mm². With a rapid response time down to 20 nanoseconds, the chip ensures precise timing, particularly in time-sensitive applications at the edge. As a result, its strengths come to the fore in real-time evaluation of event-based sensor data and closed-loop control systems, such as in the control of small electric motors with AI. SENNA can also be used to implement AI-optimised data transmission in communication systems. In this context, the AI processor can analyse signal streams and adapt transmission and reception procedures as required to improve the efficiency and performance of the transmission.
One reason SNNs operate so energy-efficiently is that neurons activate only sparingly and in response to specific events. With its spiking neurons, SENNA fully utilises this energy-saving advantage. Thanks to its fully parallel processing architecture, the artificial neurons accurately map the temporal behaviour of SNNs. SENNA can also interface directly with spike-based input and output signals through its integrated spike interfaces. In this manner, it integrates seamlessly into an event-based data stream.
“With its novel architecture, SENNA resolves the trade-off between energy efficiency, processing speed, and versatility like no other edge AI processor. This makes it perfect for resource-limited applications that require extremely fast response times in the nanosecond range,” explains Michael Rothe, Group Manager Embedded AI at Fraunhofer IIS.
The current SENNA reference design is designed for 22 nm manufacturing processes, enabling the SNN processor to be used as a chip in various applications and implemented cost-effectively. Its design is scalable and can be adapted to specific applications, performance requirements, and special features of the target hardware before chip production. However, even after the chip has been manufactured, SENNA retains maximum flexibility because it is fully programmable. The SNN model used can be repeatedly changed and retransferred to SENNA. To make it as easy as possible for developers to implement their AI models, Fraunhofer IIS also provides a comprehensive software development kit for SENNA.
Image: SENNA is a neuromorphic SNN chip for ultra-fast and energy-efficient low-dimensional time series data processing. Copyright: Fraunhofer IIS / Paul Pulkert.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
