Over the last five years, Intel has been refining its neuromorphic chips and systems to create devices that simulate how the human brain works using computational “neurons.” When Loihi 2 was first introduced, it was only available through Intel’s neuromorphic research cloud for select members of the Intel Neuromorphic Research Community (INRC). Now, Intel is broadening access to Loihi 2 with the launch of Kapoho Point, a small development board for neuromorphic computing featuring eight Loihi 2 chips.
Mike Davies, senior principal engineer and director of Intel’s Neuromorphic Computing Lab, shared that the company started with the Loihi chip five years ago. Since then, they have developed a variety of systems, from small USB-type devices to large rack-mounted units with up to 768 Loihi chips.
Loihi 2, which was introduced a year ago, brings significant improvements over the original chip. It’s faster, has higher bandwidth for communication between chips, and offers better scalability for tackling larger problems. The chip also has more capacity, is smaller due to process scaling, and offers more programmability.
Kapoho Point is a new, ultra-compact form factor from Intel. Roughly the size of a credit card, this board features four Loihi 2 chips on the top and another four on the underside, totaling eight chips in a very small space. According to Davies, Kapoho Point can support up to a million neurons and a billion synapses, providing a solid network scale in a compact form. It can handle optimization problems with up to 8 million variables and is said to be 1000 times more energy-efficient than a typical CPU solver.
One unique feature of Kapoho Point is its stackability. Through an expansion port, users can stack multiple boards to increase its capabilities. Intel has successfully stacked up to four boards, allowing for a total of 32 million neurons, which is quite substantial for optimization problems. The company anticipates that Kapoho Point could eventually be stacked up to eight boards before other configurations become more practical.
Intel is eager to see how the research community utilizes Kapoho Point, with potential applications in robotics and drones. The Loihi and Loihi 2 chips have already been used in various fields, including robotics control, gesture recognition, scene understanding, and even smell detection.
Kapoho Point has already been delivered to a select group of partners. The first to receive the board was the Air Force Research Laboratory (AFRL), which is using it for research on spiking neural network-based learning and real-time optimization. Given that AFRL’s missions involve mobile platforms with limited space, weight, and power, they believe neuromorphic computing offers the best solution for running AI algorithms in such environments.
In addition to the Kapoho Point announcement, Intel also introduced recent updates to its Lava software framework, designed to support Loihi 2. The company also revealed the latest round of sponsored projects through the INRC. Intel is funding several academic institutions to push forward research in neuromorphic algorithms and applications. These institutions include Brown University, ETH Zürich, George Mason University, Graz University of Technology, Pennsylvania State University, Queensland University of Technology, University of Göttingen, and University of Waterloo. Research topics range from auditory feature detection to brain-computer interfaces.
