Intel 64-chip neuromorphic system is launched, with a surge in speed and efficiency

Intel announced that the 8 million neuron neuromorphic system, codenamed "Pohoiki Beach", is now available to researchers. It contains 64 Loihi research chips. Through Pohoiki Beach, researchers can conduct experiments using Intel's Loihi research chip, which is inspired by the brain and applies biological brain principles to computer architecture. In professional applications such as sparse coding, graph search, and constraint satisfaction problems, Loihi allows users to process information at a speed of 1,000 times faster and 10,000 times more efficient than a CPU.

A close-up of Intel's neuromorphic research chip Loihi. Intel's latest neuromorphic system Pohoiki Beach is made up of 64 Loihi chips.

 “Earlier, we expanded Loihi to create a more powerful neuromorphic system, and we are excited about the initial results of this work. Pohoiki Beach is now available to more than 60 ecosystem partners who will use this specialized system to solve complex, compute-intensive problems.”

——Rich Uhlig, President of Intel Labs

Rich Uhlig, director of Intel Labs, holds an Intel Nahuku baseboard, each of which contains 8 to 32 Intel Loihi neuromorphic chips. Intel's latest neuromorphic system, Pohoiki Beach, consists of multiple Nahuku baseboards with 64 Loihi chips.

Why it matters: With the introduction of Pohoiki Beach, researchers can efficiently scale new neural-inspired algorithms — such as sparse coding, simultaneous localization and mapping (SLAM), and path planning — that learn and adapt based on the data they take in. Pohoiki Beach is a major milestone in Intel’s neuromorphic research, laying the foundation for Intel Labs’ plan to scale the architecture to 100 million neurons later this year.

Why it’s unique: To continue to reduce power consumption and increase performance in line with Moore’s Law, more than just continued process node scaling is required. As new and complex computing workloads become the norm, there is a growing need for application-specific architectures.

From the Pohoiki Beach neuromorphic system, we can see that specialized architectures can bring many benefits to emerging applications, including some difficult computing problems that are difficult to support by the Internet of Things (IoT) and autonomous devices. Using this specialized system, which is different from general-purpose computing technology, it is expected to achieve orders of magnitude speed and efficiency improvements in many real-world application areas, such as self-driving cars, smart homes, network security, etc.

A close-up of Intel's Nahuku baseboards, each containing 8 to 32 Intel Loihi neuromorphic chips. Intel's latest neuromorphic system, Pohoiki Beach, is made up of multiple Nahuku baseboards containing 64 Loihi chips.

Research Partner Feedback: With the introduction of Pohoiki Beach, Intel will support global ecosystem partners to continue pioneering the next frontier of neural-inspired algorithm research.

For example, at this week’s Telluride Neuromorphic Cognitive Engineering Symposium, researchers applied the Loihi system to address cutting-edge challenges in neuromorphic engineering. These projects include making the AMPRO prosthetic more adaptive, object tracking with emerging event cameras, automating table football through neuromorphic sensory control, learning to control a linear inverted pendulum, and providing tactile input to the electronic skin of the iCub robot.

In addition to the demonstrations from Telluride, other research partners have already seen huge benefits from Loihi:

“Loihi chips use 109 times less power than GPUs to run real-time deep learning benchmarks, and 5 times less power than dedicated IoT inference hardware,” said Chris Eliasmith, professor at the University of Waterloo and co-CEO of Applied Brain Research. “An even more exciting result is that when we scaled the network 50 times, Loihi was able to maintain real-time performance with only a 30% increase in power consumption, while IoT hardware increased power consumption by 500% and failed to maintain real-time performance.”

 “With Loihi, we built a spiking neural network that mimics the brain’s underlying neural representations and behavior. The SLAM solution emerges as a property of the network structure. We benchmarked the network running on Loihi and found it to be as accurate as a widely used SLAM approach running on CPUs for mobile robots, but at one-hundredth the power,” said Rutgers University Professor Konstantinos Michmizos, describing his lab’s SLAM research results, which he will present at the International Conference on Intelligent Robots and Systems (IROS) in November.

Next steps: In 2017, Intel launched its first neuromorphic research chip, Loihi, taking an important step in the development of neuromorphic hardware. In March 2018, the establishment of the Intel Neuromorphic Research Community (INRC) further promoted the development of neuromorphic algorithms, software, and applications. Through INRC, Intel provides its Loihi cloud system and the Loihi-based USB-shaped system Kapoho Bay to researchers. The Kapoho Bay system has strongly promoted research on the practical application of neuromorphic technology.

Pohoiki Beach will provide Intel's research partners with greater computing scale and greater computing power, which will further accelerate the progress of neuromorphic technology.

An Intel Nahuku baseboard connected to an Arria 10 FPGA development kit is shown here. Each baseboard contains 8 to 32 Intel Loihi neuromorphic chips. Intel's latest neuromorphic system, Pohoiki Beach, consists of multiple Nahuku baseboards with 64 Loihi chips.

Later this year, Intel will launch a larger Loihi system, codenamed “Pohoiki Springs,” built on the Pohoiki Beach architecture and designed to deliver unprecedented performance and efficiency for scaled-out neuromorphic workloads.

Intel engineers say that measurements from these research systems are expected to quantify the gains that neuromorphic computing methods can bring and indicate the application areas that are best suited for this technology. This research paves the way for the eventual commercialization of neuromorphic technology.