サイエンス
脳の「直感と熟考」をAIチップに実装。速度4倍・省電力5倍を達成した新技術
脳の構造を模倣したスパイキングニューラルネットワークは、長期記憶の保持に多大な計算コストを要する課題があった。研究チームは「高速な反射」と「低速な思考」を分離した新設計により、低消費電力と高度な文脈理解の両立に成功した。
Tech Product
Loihi 2
Overview
最終更新: 2026年7月20日Loihi 2 is Intel's second-generation neuromorphic research chip. It features up to 1 million neurons per chip and is built using the Intel 4 process. It is designed to mimic the human brain's neural structure to provide high-efficiency processing for spiking neural networks and other brain-inspired algorithms.
Mentioned Articles
2 件
External Mentions
5 件- arXivConvolutional Sparse Coding via the Locally Competitive Algorithm on Loihi 2
- arXivCLANE: Continual Learning of Actions on Neuromorphic Hardware from Event Cameras
- arXivBenchmarking Fairness in Spiking Neural Networks: Data Bias, Spurious Features, and Hardware Effects
- arXivReal-Time Frame- and Event-based Object Detection with Spiking Neural Networks on Edge Neuromorphic Hardware: Design, Deployment and Benchmark
- arXivEdgeSpike: Spiking Neural Networks for Low-Power Autonomous Sensing in Edge IoT Architectures