According to Cointelegraph, IBM has recently introduced a prototype artificial intelligence (AI) chip, named NorthPole, which is claimed to be faster and more energy-efficient than any existing chip. Research published in Science Magazine on October 19 revealed that NorthPole achieves a 25 times higher energy metric and a 22 times lower time metric of latency on a relevant benchmark. This could potentially lead to post-GPU performance with significantly reduced energy requirements.
Damien Querlioz, a nanoelectronics researcher at the University of Paris-Saclay, described the energy efficiency of NorthPole as 'mind-blowing' in an article published on Nature. The IBM Research team's paper stated that NorthPole outperforms all prevalent architectures, even those using more advanced technology processes. One of the main challenges in improving AI processing is the 'von Neumann bottleneck', which causes latency when information is sent between the processing unit and random access memory. IBM Research claims that the new prototype chip, built in the company's Alamaden, California laboratory, bypasses this bottleneck by integrating the memory component onto the processing chip itself.
The benchmark used to demonstrate the chip's effectiveness, ResNet50, is a 50-layer neural network primarily used to test computer vision tasks such as image classification. The reported results of the NorthPole hardware on this benchmark suggest that it could perform exceptionally well in tasks such as autonomous surgery, operation of self-driving cars and other vehicles, and various robotics-related endeavors. IBM Research is already working on the next chip using the NorthPole architecture, with the company blog stating that 'this is just the start of the work for Modha on NorthPole.'
