Today, Intel launched the second-generation neuromorphic research chip Loihi 2 and the open source software framework Lava for developing neuro-inspired applications. Their launch marks Intel’s continued progress in advancing neuromorphic technology. Neuromorphic computing draws insights from neuroscience to create a chip that functions more like a biological brain, hoping to achieve orders of magnitude improvement in energy efficiency, computing speed, and learning efficiency, including a series of edge applications such as vision, speech, and gesture recognition, search and retrieval , Robotics and constraint optimization problems.
"Loihi 2 and Lava have gained insights from years of collaborative research using Loihi. Our second-generation chip greatly improves the speed, programmability, and capacity of neuromorphic processing , Expanding its use in power and latency-constrained intelligent computing applications. We are open source Lava to solve the field of software integration, benchmarking and cross-platform collaboration needs, and accelerate our commercial feasibility Progress."-Mike Davis, Intel Director
So far, Intel and its partners have demonstrated applications including robotic arms, neuromorphic skin, and olfactory sensing. About Loihi 2: This research chip combines the three-year experience of using the first-generation research chip and takes advantage of the advancement of Intel process technology and asynchronous design methods.
Loihi 2's advancements enable the architecture to support new categories of neural-inspired algorithms and applications, while providing up to 10 times the processing speed and up to 15 times the resource density (up to 1 million nerves per chip) Yuan) and higher energy efficiency.Thanks to the close cooperation with Intel's Technology Development Department, Loihi 2 uses a pre-production version of Intel 4 process, highlighting the health and progress of Intel 4. Intel 4 uses extreme ultraviolet (EUV) lithography technology compared with past process technology, which simplifies layout design rules. This makes it possible to develop Loihi 2 quickly.
Lava software framework meets the needs of the neuromorphic research community for general software frameworks. As an open, modular and extensible framework, Lava will allow researchers and application developers to build and integrate a common set of tools, methods and libraries based on each other's progress. Lava runs seamlessly on heterogeneous architectures across traditional and neuromorphic processors, supports cross-platform execution and interoperability with various artificial intelligence, neuromorphic and robotic frameworks. Developers do not need to access specialized neuromorphic hardware to start building neuromorphic applications, and can contribute to the Lava code base, including porting it to run on other platforms.
" Los Alamos National Laboratory researchers have been using the Loihi neuromorphic platform to study the trade-offs between quantum computing and neuromorphic computing on a chip, The learning process," said Dr. Gerd J. Kunde, a scientist at Los Alamos National Laboratory. "This research shows some exciting equivalences between impulsive neural networks and quantum annealing methods for solving difficult optimization problems. We also proved that the backpropagation algorithm is the basic building block for training neural networks. It was previously thought that it could not be implemented on the neuromorphic architecture and could be effectively implemented on Loihi. Our team is very happy to continue this research with the second-generation Loihi 2 chip."
About key breakthroughs: Loihi 2 and Lava provides tools for researchers to develop and characterize new neuro-inspired applications,Used for real-time processing, problem solving, adaptation and learning. Notable highlights include:
- Faster and more versatile optimization: Loihi 2’s greater programmability will allow support for a wider range of difficult optimization problems, including real-time optimization from the edge to the data center system, Planning and decision-making.
- New method of continuous and associative learning: Loihi 2 has improved support for advanced learning methods, including backpropagation variants, the main algorithm of deep learning. This expands the range of adaptive and data-efficient learning algorithms that can be supported by low-power form factors running in online settings.
- A new type of neural network that can be trained through deep learning: the fully programmable neuron model and generalized spike message passing in Loihi 2 open the door to a variety of new neural network models that can be trained in deep learning. Early evaluations show that, compared to the standard deep network running on the original Loihi, each inference operation on Loihi 2 is reduced by more than 60 times without loss of accuracy.
- seamlessly integrates with real-world robotic systems, traditional processors and new sensors: Loihi 2 solves Loihi's practical limitations by integrating faster, more flexible and more standard input/output interfaces. The Loihi 2 chip will support Ethernet interfaces, seamless integration with the wider event-based vision sensor , and a larger Loihi 2 chip mesh network.
