IBM has announced a new microchip design that mimics the human brain in terms of memory and processing. The chip, called NorthPole, could enable smarter, more efficient, and more autonomous devices that do not depend on the cloud or the internet for their intelligence. This could have implications for various fields, such as defense, healthcare, and education.
NorthPole: A neural inference architecture
NorthPole is a microchip that combines memory and processing circuits on the same chip, unlike traditional chips that separate them. This allows the chip to store and process information based on its connections to other circuits, similar to how synapses in the brain work. The chip also uses a novel algorithm that enables it to learn from data without supervision.
According to IBM researchers, who published their work in the journal Science, NorthPole is a “neural inference architecture that blurs this boundary by eliminating off-chip memory, intertwining compute with memory on-chip, and appearing externally as an active memory.”
The chip has 22 billion transistors and 256 cores, which can perform up to 1 trillion operations per second. It can also handle multiple types of data, such as images, text, and speech.
NorthPole: A breakthrough for AI and energy efficiency
One of the main advantages of NorthPole is its energy efficiency. The chip can perform complex tasks such as processing moving images with much less power than conventional chips. For example, against a comparable chip with 12nm silicon technology process node and with a comparable number of transistors, NorthPole delivers 25 times higher frames per joule.
This means that NorthPole can enable devices that can run AI applications without relying on external power sources or network connections. This could be useful for soldiers who operate drones, ground robots, or augmented-reality gear against adversaries who can target electronic emissions. It could also benefit remote areas where internet access is limited or unreliable.
Another advantage of NorthPole is its potential to overcome the limitations of current chip technology. As the demand for AI grows, so does the need for more powerful and faster chips. However, there is a physical limit to how many transistors can be packed on a chip. NorthPole’s unique architecture could pave the way for a different kind of AI, one that does not depend on big cloud and data companies like Amazon or Google.
NorthPole: A step towards cognitive computing
IBM’s NorthPole is not the first attempt to create a brain-inspired chip. In 2014, IBM unveiled TrueNorth, a chip that also combined memory and processing on the same chip. However, TrueNorth was designed for specific tasks and required specialized programming. NorthPole is more flexible and adaptable, and can learn from any type of data.
IBM’s senior fellow Dharmendra S. Modha, who led the research team, said that NorthPole is a step towards cognitive computing, which aims to create machines that can understand, reason, and learn like humans.
“What we really set out to do is optimize every joule of energy, every capital cost of a transistor, and every opportunity for a single clock cycle,” Modha said in an interview. “So it’s been optimized along these three dimensions: energy, space and time.”
IBM plans to make NorthPole available to researchers and developers through its cloud platform in the near future. The company hopes that the chip will inspire new applications and innovations in AI and beyond.
