Jane Edwards
A team of researchers at IBM Research’s laboratory in Almaden, California, has developed a brain-inspired chip, called NorthPole, that could pave the way for the development of more energy-efficient and faster artificial intelligence hardware systems.
Dharmendra Modha and his colleagues from IBM Research used a 12-nanometer node processing technology to develop the chip prototype, which has 22 billion transistors and 256 cores with the capability to carry out 2,048 operations for each core per cycle at 8-bit precision, according to an IBM Research blog post published Friday.
“NorthPole is a faint representation of the brain in the mirror of a silicon wafer,” Modha said.
The research team has tested the potential use of the chip in image segmentation, detection, video classification and natural language processing, among other applications.
According to a study published in the journal Science, NorthPole showed lower latency and higher space and energy efficiency when tested on YOLOv4 object detection and ResNet-50 image recognition models.
“Architecturally, NorthPole blurs the boundary between compute and memory,” Modha said.
“At the level of individual cores, NorthPole appears as memory-near-compute and from outside the chip, at the level of input-output, it appears as an active memory,” he added.
