Xuanhei Technology: Neuromorphic Sensing & Processing 3D Integrated Chip

At the end of 2022, the European Union launched the cutting-edge project Nimble AL, aiming to upgrade neuromorphic visual sensing technology. As Codasip's innovation incubator, Codasip Labs has been exploring the application of new technologies in the future, and recently joined the Nimble AL project to develop a RISC-V customizable core to empower neuromorphic sensing 3D integrated chips.

NimbleAI is a 3-year, €10 million research project that brings together partners from commercial organizations and academia across the EU and UK. The project, funded by the European Union, aims to design a three-dimensional integrated chip for neuromorphic sensing and processing. This dazzling black technology is inspired by the detection of light by the eyes and the processing of visual information by the brain.

Clearly, this is a cutting-edge technology inspired by biology, with the goal of improving the energy efficiency and performance of next-generation neuromorphic chips for event-based visual sensing. If all goes according to plan, a complete neuromorphic architecture will be created to empower end devices to efficiently and accurately run and diversify computer vision algorithms. At present, these application equipment are usually limited in resources and area. Typical application areas include handheld and battery-powered medical imaging devices, smart displays for autonomous vehicles, wearable eye-tracking glasses, and more.

Using biological systems as the "golden benchmark" for electronic systems has been a technology trend in recent years. Inspired by biology, NimbleAI is utilizing the biological eye-brain system as its system architecture. The project is implementing an always-on sensing stage, highly specialized event-driven processing cores and neural networks to perform visual reasoning on selected stimuli using minimal energy.

Codasip is actively exploring the extension of the RISC-V ISA in memory accelerators in this project, which are provided by partner CEA. Customization of RISC-V helps improve the performance of neuromorphic chips and reduce their power consumption. To achieve this goal, we chose Codasip Studio tools and the CodAL architecture description language, which make the customization workflow straightforward and convenient. The RISC-V core developed by Codasip was then mapped to Menta SAS's eFPGA.

Game-changing cool technology

Currently in the NimbleAI project, the Codasip team is developing a next-generation artificial intelligence processor based on RISC-V. At the same time, this is also a perfect case in the series of projects of Codasip Labs. Codasip Labs focuses on key application areas including security, functional safety and AI/ML, among others. To identify and build technologies that expand the limitless possibilities of custom computing. Accelerate time-to-market for differentiated products through customized, domain-specific designs that ultimately win the market. At the end of the 3-year NimbleAI project, we expect to see significant performance improvements compared to current standards for CPU or GPU processing of frame-based video. The project will enable new functionality and practical implementations to unlock more advanced artificial intelligence and computer vision algorithms and applications.