Seamless Design Of Smart Edge Processors

点击翻译按钮获取中文摘要

With the rise of deep learning (DL), our world braces for Artificial Intelligence (AI) in every edge device, creating an urgent need for edge AI processing hardware. Unlike existing solutions, this hardware needs to support high throughput, reliable, and secure AI processing at ultra-low power (ULP), with a very short time to market. With its strong legacy in edge solutions and open processing platforms, the EU is ideally positioned to become the leader in this edge-AI market. However, certain roadblocks keep the EU from assuming this leadership role: Edge processors need to become 100x more energy efficient; Their complexity demands automated design with 10x design time reduction; They must be secure and reliable to get accepted; Finally, they should be flexible and powerful to support the DL domain. CONVOLVE addresses these roadblocks and thereby enables EU leadership in Edge-AI. To that end, it will take a holistic approach with innovations at all design stack levels, including: 1.ULP memristive circuits for computation-in-memory 2.Fast compositional design of System-on-Chips (SoC) 3.Transparent compilers supporting automated code optimizations and domain-specific languages 4.Rethinking DL models through dynamic neural networks, event-based execution, and sparsity 5.On-edge continuous learning for improved accuracy, self-healing, and reliable adaptation to non-stationary environments 6.Holistic integration in SoCs supporting secure execution with real time guarantees The CONVOLVE consortium includes some of Europe's strongest research groups and industries, covering the whole design stack and value chain. In a community effort, we will demonstrate Edge-AI computing in real-life vision and audio domains. By combining these innovative ULP and fast design solutions, CONVOLVE will, for the first time, enable reliable, smart, and energy efficient edge-AI devices at a rapid time-to-market and low cost, and as such open the road for EU leadership in edge-processing.