AI chips advance with Intel's Pohoiki Beach

Intel uses 64 Loihi research chips in its neuromorphic system called Pohoiki Beach. (Intel)

Intel advanced its neuromorphic system called Pohoiki Beach to the broader research community Monday, setting the stage for faster processing to be used in AI for IoT and autonomous vehicles and devices. The system will be used by more than 60 partners for solving complex problems requiring heavy numbers crunching.

It runs 64 Loihi research chips and is designed to simulate 8 million neurons. Intel has committed to deliver a 100-million neuron system later in 2019 called Pohoiki Springs.

Loihi, first introduced in 2017, can process information up to 1,000 times faster and 10,000 times more efficiently than CPUs. Creating a 64-Loihi chip system will help researchers scale up sparse coding, simultaneous localization and mapping (SLAM) and path planning to learn and adapt based on new data inputs.

Industry Event

Embedded Technologies Expo & Conference

Limited Passes Available - Register now and save with Early Bird Rates!

Embedded Technologies Expo & Conference is North America’s ONLY embedded event focused on helping design engineers through exhibits showcasing new and emerging products & solutions, education, training, and hands-on learning experiences. Join us in the heart of Silicon Valley - June 22-24, 2020 at the McEnery Convention Center in San Jose, CA. Use promo code FE100 for $100 off Conference Passes.

The benefits of Loihi also include lower power consumption—about 109 times lower power consumption compared to a GPU and a five times lower power consumption compared to some IoT inference hardware, according to Applied Brain Research.  

Loihi was inspired by the way the human brain solves problems-- the basic genesis of all neuromorphic research.  Intel defines neuromorphic computing as computing that emulates the neural structure of the brain, which can apply principles of common sense and context and deal with uncertainty, ambiguity and contradiction. Early versions of AI were more brittle, literal and deterministic.

“Loihi allowed us to realize a spiking neural network that imitates the brain’s underlying neural representations and behavior,” said Prof. Konstantinos Michmizos of Rutgers University, in a statement. With a Loihi-run network, his lab was able to work with mobile robots accurately and with 100 times less energy than a widely-used CPU.

Intel has described uses for AI based on neuromorphic computing to include medical imaging and autonomous vehicles (AV). With AV, solutions could emerge to deal with uncertainties, such as a ball rolling into the street or an aggressive driver in another vehicle on the road. In a medical image, AI can be used to highlight regions where there is more certainty about an illness.

Intel said its research work will eventually lead to commercialization of neuromorphic technology, but didn’t offer a specific roadmap. A major driver for the research is the need to continue the gains in computing power and performance since Moore’s Law and process-node computing can’t continue to expand. Specialized architectures such as the Pohoiki Beach neuromorphic approach are needed for specific emerging applications such as AV, smart homes and cybersecurity, Intel believes.

RELATED: "Machine learning training puts Google and Nvidia on top"


Suggested Articles

Renesas Electronics Corp. has an evaluation board for developers working with the 32-bit RX23E-A microcontroller (MCU) to do IoT development.

“Our moonwalk equivalent,” is how one National Instruments worker describes the colossal effort to provide hundreds of thousands of ventilators.

Even before the current COVID-19 crisis, the global semiconductor materials was flat at best, according to global industry association SEMI.