‘Mind-blowing’ IBM chip speeds up AI

24/10/2023

Researchers at IBM in San Jose, California, have developed a groundbreaking brain-inspired computer chip called NorthPole, designed to significantly enhance artificial intelligence (AI) capabilities. Unlike traditional chips that rely on external memory (RAM), NorthPole integrates computing and memory on a large scale, eliminating the need for frequent access to external memory. This innovation resolves the Von Neumann bottleneck, a phenomenon that hampers computing speed and energy efficiency in AI applications.

NorthPole consists of 256 cores, each with its own memory, wired together in a network inspired by the connections in the human cerebral cortex. It outperforms existing AI chips in image recognition benchmarks and uses only one-fifth of the energy, even without utilizing the latest manufacturing processes. Although NorthPole has limitations, such as a fixed neural network and limited memory for large language models, its architecture holds promise for applications like self-driving cars.

The NorthPole chip, developed by IBM researchers, represents a significant leap in AI technology. By integrating computing and memory within the chip itself, NorthPole eliminates the bottleneck caused by external memory access, resulting in faster and more energy-efficient AI computations. This innovative approach, inspired by the human brain’s neural networks, has the potential to revolutionize various fields, including self-driving cars. While challenges remain, such as accommodating larger language models, NorthPole’s design principles pave the way for future advancements in AI hardware, offering a glimpse into the future of high-speed, low-energy computing.