The second-generation studio chip uses the Intel four pre-production procedure and reaches 1 million neurons. Intel adds an open software framework to drive developer innovation and the path to commercialization.
SANTA CLARA, Calif. , Sept. 30, 2021 — (BUSINESS WIRE) — What’s new today, Intel brought Loihi 2, its second-generation neuromorphic studies chip, and Lava, an open-source framework for new application technologies inspired by neuroinstructions. Its advent indicates Intel’s continued progress in advancing neuromorphic technology.
This press has multimedia. Read the full press here: https://www. businesswire. com/news/home/202109330005258/en/
One photo shows Intel’s Loihi 2 neuromorphic chip at the tip of a finger. Loihi 2 is Intel’s second-generation neuromorphic studies chip. It supports new categories of algorithms and neuro-inspired applications, while providing faster processing, higher resource density, and advanced energy efficiency. It was brought through Intel in September 2021. (Credit: Walden Kirsch / Intel Corporation)
Loihi 2 and Lava are collecting data from several years of collaborative Loihi studies. Our second-generation chip dramatically improves the speed, programmability and capability of neuromorphic processing, expanding its uses in intelligent computer programs with constraints of strength and latency. Lava source to address the need for software convergence, benchmarking and cross-platform collaboration in the field, and to drive our progress towards advertising viability. Mike Davies, Director of Intel’s Neuromorphic Computing Laboratory
Why it matters: Neuromorphic computing, which relies on neuroscience to create chips that look more like the biological brain, aspires to bring innovations of several orders of magnitude in power, computational speed, and learning power in a variety of cutting-edge applications: vision, voice popularity, and gestures to search for recovery problems, robotics and optimization under restrictions.
Applications that Intel and its partners have demonstrated to date include robotic arms, neuromorphic skin, and olfactory detection.
About Loihi 2: The study chip classes learned over 3 years of use with the first-generation studio chip and leverage advances in Intel’s generation procedures and asynchronous design methods.
Advances in Loihi 2 allow the architecture to help new categories of neuro-inspired algorithms and applications, while offering procedures up to 10 times faster1, up to 15 times more resource density2 with up to 1 million chip-compatible neurons and improved energy efficiency. From a close collaboration with Intel’s generation progression group, Loihi 2 was manufactured with a pre-production edition of the Intel four procedure, which highlights the aptitude and progress of Intel Four. The use of excessive ultraviolet lithography (EUV) in Intel four has simplified design regulations compared to previous procedure technologies, making possible the temporary expansion of Loihi 2.
As an open, modular, and extensible framework, Lava will allow scholars and application developers to take advantage of the progress of others and converge towards a non-unusual set of tools. strategies and libraries. Lava works seamlessly in heterogeneous architectures in traditional and neuromorphic processors, enabling cross-platform execution and interoperability with a variety of ai, neuromorphic, and robotics frameworks. Developers can start building neuromorphic programs without access to specialized neuromorphic hardware and can give a contribution to Lava’s codebase, adding clothing to run on other platforms.
“Researchers at Los Alamos National Laboratory used the Loihi neuromorphic platform to examine the trade-offs between quantum and neuromorphic computing, as well as to set in motion on-chip learning processes,” said Dr. Gerd J. Kunde, staff scientist at Los Alamos National Laboratory. “This study has shown exciting equivalences between complex neural networks and quantum annealing approaches to solve complicated optimization problems. We have also shown that the backpropagation algorithm, a basic detail for the formation of neural networks and in the past it was believed that it cannot be applied in neuromorphic architectures, can be done well in Loihi. Our team is excited to continue these studies with the second-generation Loihi 2 chip. “
About Key Breakthroughs: Loihi 2 and Lava provide equipment for researchers to expand and characterize new neuroinstruct-inspired programs for real-time treatment, challenge solving, adaptation, and learning.
Faster and overall optimization: Loihi 2’s increased programming capability will bring broader elegance to challenging optimization issues, adding real-time optimization, planning, and decision-making from edge systems to knowledge centers.
New approaches to associative and seamless learning: Loihi 2 improves complex learning methods, adding variations of retropropagation, the running set of rules of deep learning. -power form points operating on in-line parameters.
New neural networks that can be trained through deep learning: Fully programmable neural models and generalized spike messages in Loihi 2 open the door to a wide diversity of new neural network models that can be trained through deep learning. times fewer inference operations in Loihi 2 to popular deep networks operating in the original Loihi without loss of accuracy3.
Seamless integration with real-world robot systems, traditional processors, and new sensors: Loihi 2 solves a practical limitation of Loihi by incorporating faster, more flexible, and popular input/output interfaces. Loihi 2 chips will have Ethernet interfaces, glue-free integration with a diversity of event-based vision sensors, and wider mesh networks of Loihi 2 chips.
More main points can be discovered in the Loihi 2/Lava product knowledge sheet.
About the Intel Neuromorphic Research Community: The Intel Neuromorphic Research Community (INRC) has nearly 150 members, with several new additions this year, adding Ford, Georgia Institute of Technology, Southwest Research Institute (SwRI) and Teledyne-FLIR. Enroll in a strong network of academic, government, and industry partners working with Intel to promote real-world advertising uses of neuromorphic computing (read what our partners are saying about Loihi technology).
“Advances like the new Loihi 2 chip and the Lava API are vital advances in neuromorphic computing,” said Edy Liongosari, lead researcher and managing director of Accenture Labs. “The next-generation neuromorphic architecture will be for Accenture Labs’ studies on the brain PC vision algorithms inspired by intelligent edge computing that can force long-term long-term long-term real headsets or smart cell robots. provides developers with a simpler and more optimized interface for building neuromorphic systems. “
On the path to commercialization: Advancing neuromorphic computing from laboratory studies to commercially viable generation is a three-front effort. This requires an uninterrupted iterative lie of neuromorphic hardware in reaction to algorithmic and application-seeking results; Development of a non-unusual cross-platform software framework so that developers can compare, integrate and the most productive algorithmic concepts from other groups; and deep collaborations between industry, academia, and government to create a rich and productive neuromorphic ecosystem to explore instances of commercial use with short-term commercial value.
Intel’s announcements cover all of those areas, putting new equipment in the hands of a developing ecosystem of neuromorphism researchers committed to rethinking computing from its foundations to achieving breakthroughs in intelligent data processing.
Next step: Intel is lately providing two Loihi 2-based neuromorphic formulas through the neuromorphic research cloud for committed INRC members: Oheo Gulch, a single-chip formula for early evaluation, and Kapoho Point, an eight-chip formula that will soon be the Lava software framework is available for loose download on GitHub. A presentation and tutorials on Loihi 2 and Lava will be presented on the next Intel Innovation occasion in October.
More context: Neuromorphic computing: Next-generation AI (Intel. com) | Neuromorphic Computing on Intel (Press Kit)
About Intel
Intel (Nasdaq: INTC) is an industry leader creating a world-changing generation that enables global progress and enriches lives. Inspired through Moore’s Law, we continually work to advance semiconductor design and production to help address our customers’ biggest challenges. cloud intelligence, netpaintings, edge and any type of computing device, unlock the perspective of knowledge for companies and society for the better. To learn more about Intel’s innovations, visit newsroom. intel. com and intel. com.
1 Based on Lava simulations in September 2021 of a nine-layer variant of the DNN PilotNet inference workload implemented as a sigma-delta neural network in Loihi 2 compared to the same network implemented with SNN rate encoding in Loihi . The Lava functionality style for any of the chips is based on the characterization of the silicon edition 1. 0. 0 of the Nx SDK with an Intel Xeon E5-2699 v3 2. 30 GHz processor, 32 GB RAM, as a host running the Ubuntu edition. 20. 04. 2. Loihi’s effects use the Nahuku-32 ncl-ghrd-04 system. Loihi 2’s effects use the Oheo Gulch ncl-og-04 system. Results would possibly vary. 2 Based on Loihi 2 core length of 0. 21 mm2 supporting up to 8192 neurons compared to Loihi core length of 0. 41 mm2 supporting up to 1,024 Array3 neurons Based on PilotNet inference workload measurements Nine-layer DNN referenced above, with a sigma-delta neural network implementation in Loihi 2 achieving a mean square error (MSE) of 0. 035 with 323815 synaptic operations compared to a rate-encoded SNN in Loihi 1 achieving an MSE of 0. 0412 with 20,250,023 synaptic operations.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and trademarks will likely be claimed as assets of others.
See the businesswire. com edition: https://www. businesswire. com/news/home/20210930005258/en/
Contacts
Supriya Venkat503-320-8024supriya. venkat@intel. com