Intel unveils second-gen neuromorphic chip
Loihi 2 is designed to mimic the architecture of the human brain
Intel has announced Loihi 2, the second generation of its neuromorphic chip designed to model the workings of the human brain.
The company claims that Loihi 2 is nearly 10 times faster than its predecessor, and can be programmed to help researchers handle more computing tasks.
A neuromorphic chip is a device that uses physical artificial neurons made from silicon to perform computations. Such chips make use of very-large-scale integration (VLSI) systems containing electronic analogue circuits that mimic the functions of the nervous system.
Neuromorphic computing aspires to deliver orders of magnitude improvements in computational speed, energy efficiency and learning across a range of applications: from voice, vision and gesture recognition to robotics, search retrieval and constrained optimisation problems. Intel and its associates have demonstrated applications including olfactory sensing, neuromorphic skins and robotic arms.
The Loihi 2 chip is 30mm² and features up to 1 million computational neurons per chip - more than seven times those in the first-gen Loihi 1 device.
Intel says the chip has 15 times greater resource density, as well as better energy efficiency.
Early tests showed that Loihi 2 required 60 times fewer operations per inference when running standard AI models compared to Loihi 1, without a loss in accuracy.
"Our second-generation chip greatly improves the speed, programmability, and capacity of neuromorphic processing, broadening its usages in power- and latency-constrained intelligent computing applications," said Intel neuromorphic computing director Mike Davies.
The chip giant now offers two Loihi 2-based neuromorphic systems: Kapoho Point and Oheo Gulch, which will be available to members of the Intel Neuromorphic Research Community (INRC) and Lava via GitHub.
Intel also launched Lava at the same time as it revealed Loihi 2: a new underlying software framework for the neuromorphic research community to develop 'neuro-inspired' applications.
The open source framework is intended to enable researchers to collaborate on a common set of tools, libraries and methods, with software that runs on both conventional and neuromorphic processors.
Lava framework is designed to be modular, extensible and hierarchical. It will be available for free use under LGPL-2.1 and BSD-3 at GitHub.