Nanotechnology Now – Press Release: Light meets deep learning: computing fast enough for the next generation of AI
Home > Press > Light meets deep learning: computing fast enough for the next generation of AI
| A team of Greek academic researchers and Californian entrepreneurs compared their Silicon Photonic (SiPho) neural network technology with processing units currently on the market and a six-year-old technology with projections. CREDIT Publication author |
Abstract:
Artificial intelligence (AI) models are essential for sophisticated image classification, the most important part of digital analysis. Researchers who recently published “Universal Linear Optics Revisited: New Perspectives for Neuromorphic Computing with Silicon Photonics” have moved the needle for image classification. The speed they achieved on a new chip platform (silicon photonics) using the computing power of neural networks is impressive.
Light meets deep learning: computing fast enough for the next generation of AI
Piscataway, N.J. | Posted on March 24, 2023
Nonetheless, note here the modal auxiliary verb “could.” Just because something can be done, questions remain. Will it be fast enough? Will it have sufficient accuracy? How energy efficient is it? Are the chips big and heavy? This study tackles them all.
One of the attributes of AI is that you can use it at the edge of the physical network; on camera for example. The camera on a drone is a better example. To enable drones with AI, you want the on-board AI chip to be powerful, but energy efficient, small and light, and capable of doing lots of complex math at lightning speed. That way, drones can warn humans when something undesirable is detected (cancer, saboteurs, railroad damage).
Meanwhile, in Greece, researchers have built a 50 GHz computational neuromorphic photonic processor capable of classifying images with ~95% accuracy. Let’s break this down, starting with the photonic part.
After Silicon Electronics? Silicon Photonics.
AI processor chips often start life as a graphics processing unit (GPU) for high-end video games or a tensor processing unit (TPU) specifically designed for neural networks, meaning that computation mimics the human brain. (Unless they like linear algebra!) But conventional processors use silicon electronics as a physical platform, which reaches quantum limits.
Switching from electrons to photons increases computational power because the speed of light is much faster than the speed of electrons. More energy efficient too. The “wire” does not heat up. Light physics can be used for matrix-vector multiplication operations, the computational backbone of neural networks.
After Conventional Mathematics? Neuromorphic Computing with Trillion Operations per Second
Now the neuromorphic part. The Greek research team, together with Celestial AI, developed a new design for the chip using a crossbar layout. The layouts outperform their advanced photonic counterparts in terms of scalability, technical flexibility, ease of programming and fault tolerance. In other words, by combining the benefits of the crossbar layout architecture with the SiGe electro-absorption modulator used in their first prototype, the researchers project that a purely optical implementation can perform trillions of matrix-vector multiplications per second, without sacrificing processing accuracy, while consuming very low power. .
Compared to six years ago, silicon photonics are in a much better position to pull neural morphic processors from their currently low computational efficiency and physical size (footprint) to lighter ones. Notice in Figure 1 the placement of the IBM TrueNorth chip, the Intel Loihi chip, the HICANN (High Input Count Analog Neural Network) chip from the University of Heidelberg Germany and the Stanford U neurogrid kit. Compare that to the lattice layout chips discussed here, which fall squarely along the roadmap photonic silicon in terms of computation and size efficiency. Powerful photonic synergy with a new cross architecture can enable the next generation of neuromorphic computing engines. Let’s change that modal auxiliary verb to “will”.
####
For more information, please click Here
Contact:
Media Contact
Mahan Christian
Institute of Electrical and Electronics Engineering
Expert Contact
Mahan Christian
IEEE
Office: 17322723320
Copyright © Institute of Electrical and Electronic Engineers
If you have any comments, please Contact us.
Publisher of the news release, not 7th Wave, Inc. or Nanotechnology Now, is solely responsible for the accuracy of the content.
Bookmark:
News and information
Optical switching at record speed opens the door to light-based ultrafast electronics and computers: March 24, 2023
Robotic caterpillars demonstrate a new approach to locomotion for soft robotics March 24, 2023
Semiconductor lattice marrying electrons and magnetic moments March 24, 2023
Solid polymer electrolyte reinforced bilayer PET/PVDF substrate enhances solid state lithium metal battery performance March 24, 2023
Possible Futures
New experiment translating quantum information between technologies in important step for quantum internet March 24, 2023
Graphene is growing – and we can see it March 24, 2023
HKUMed creates new two-dimensional (2D) ultrasonography-responsive antibacterial nanosheets to effectively treat bone tissue infections March 24, 2023
Universal HCl assistant powder-to-powder strategy for preparing lead-free perovskites March 24, 2023
Chip Technology
Graphene is growing – and we can see it March 24, 2023
Optical switching at record speed opens the door to light-based ultrafast electronics and computers: March 24, 2023
Semiconductor lattice marrying electrons and magnetic moments March 24, 2023
Researchers develop innovative tools to measure electron dynamics in semiconductors: Insights can lead to more energy-efficient chips and electronic devices March 3, 2023
Optical computing/Photonic computing
Optical switching at record speed opens the door to light-based ultrafast electronics and computers: March 24, 2023
Photonic Materials: Recent Advances and Emerging Applications February 10, 2023
New study opens the door to ultrafast 2D devices that use nonequilibrium exciton superdiffusion February 10th, 2023
Manufacturing advances bring the material back into vogue January 20, 2023
Invention
New experiment translating quantum information between technologies in important step for quantum internet March 24, 2023
Graphene is growing – and we can see it March 24, 2023
HKUMed creates new two-dimensional (2D) ultrasonography-responsive antibacterial nanosheets to effectively treat bone tissue infections March 24, 2023
Universal HCl assistant powder-to-powder strategy for preparing lead-free perovskites March 24, 2023
Announcement
Robotic caterpillars demonstrate a new approach to locomotion for soft robotics March 24, 2023
Semiconductor lattice marrying electrons and magnetic moments March 24, 2023
Solid polymer electrolyte reinforced bilayer PET/PVDF substrate enhances solid state lithium metal battery performance March 24, 2023
Understanding the mechanism of non-uniform diamond film formation on tools: Paving the way to a dry process with less environmental impact 24 March 2023
Interviews/Book Reviews/Essays/Reports/Podcasts/Journals/White Papers/Posters
HKUMed creates new two-dimensional (2D) ultrasonography-responsive antibacterial nanosheets to effectively treat bone tissue infections March 24, 2023
Universal HCl assistant powder-to-powder strategy for preparing lead-free perovskites March 24, 2023
Optical switching at record speed opens the door to light-based ultrafast electronics and computers: March 24, 2023
Robotic caterpillars demonstrate a new approach to locomotion for soft robotics March 24, 2023
Artificial intelligence
Stanford researchers develop new way to identify bacteria in fluids: Innovative adaptation of technology in old inkjet printers plus AI-assisted imaging yields a faster, cheaper way to find bacteria in blood, wastewater and more March 3, 2023
3D print decoder, AI-enabled image compression could enable higher-resolution displays December 9, 2022
New chip improves AI computing efficiency August 19, 2022
Artificial Intelligence-Centered Cancer Nanomedicine: Diagnostics, Therapies, and Bioethics 3 June 2022
Photonics/Optics/Lasers
Optical switching at record speed opens the door to light-based ultrafast electronics and computers: March 24, 2023
Stanford researchers develop new way to identify bacteria in fluids: Innovative adaptation of technology in old inkjet printers plus AI-assisted imaging yields a faster, cheaper way to find bacteria in blood, wastewater and more March 3, 2023
Photonic Materials: Recent Advances and Emerging Applications February 10, 2023
New study opens the door to ultrafast 2D devices that use nonequilibrium exciton superdiffusion February 10th, 2023
