IBM Corp has developed a brain-inspired computer chip called NorthPole that can perform AI-powered image recognition operations 22 times faster and with 25 times the energy efficiency compared to other commercial chips. The chip combines computation and memory, similar to how the human brain works, allowing it to bypass the von Neumann bottleneck and improve overall performance. NorthPole is designed as a 2D array of intertwined computing cores and memory blocks, enabling each computing core to easily access both neighboring and distant memory blocks. The chip outperforms even the latest chips from NVIDIA and delivers better performance per transistor. Although NorthPole is specialized in running AI processes and cannot perform other tasks or handle larger AI models, IBM plans to demonstrate how multiple NorthPole chips could support large language models in the future. The chip’s digital architecture redesigns will be crucial for efficient AI operations in self-driving vehicles and aircraft.
Data
Topic: AI-Optimized Chip