“Revolutionizing Computing: Hyper-multiplexed Integrated Tensor Optical Processor (HITOP)”
A technical paper titled “Hyper-multiplexed Integrated Tensor Optical Processor” has been released by a collaborative effort from researchers at the University of Southern California, Massachusetts Institute of Technology (MIT), City University of Hong Kong, and NTT Research.
READ: OAS to Step in as Mediator in Guatemala’s Power Transition Crisis
Hyper-multiplexed Integrated Tensor
The abstract of the paper outlines the pressing need for energy-efficient and scalable computing hardware due to the escalating data volume and complexity driven by the rapid expansion of artificial intelligence (AI), internet of things (IoT), and 5G/6G mobile networks. In response to this demand, the researchers introduce the hyper-multiplexed integrated tensor optical processor (HITOP), a groundbreaking technology capable of performing trillions of operations per second (TeraOPS) with an impressively low energy cost of 25 femtojoules per operation (25 fJ/OP).
HITOP operates on the principles of space-time-wavelength three-dimensional (3D) data streaming, utilizing arrays of wafer-fabricated III/V-based micron-scale lasers spanning approximately 1 terahertz (THz). These lasers incorporate thin-film Lithium-Niobate electro-optic (EO) photonics, providing the necessary foundation for achieving energy-efficient and high-performance computing.
The multidimensional parallelism inherent in HITOP allows for matrix-matrix multiplications (N3 operations) using O(N) devices, enabling scalable on-chip integration. Each device is capable of activating 10 billion parameters per second, demonstrating HITOP’s scalability in machine learning models with 405,000 parameters—an impressive 25,000 times more than previous integrated optical systems.
The paper emphasizes the combination of high clock rates (10 GS/s), parallel processing, and real-time reprogrammability as key features that unlock the full potential of light. This makes HITOP an ideal candidate for next-generation AI accelerators, applicable in diverse areas such as training with trillions of parameters, real-time decision-making in autonomous vehicles and robotics, dynamic optimization in smart manufacturing, and complex simulations for climate modeling and drug discovery.
In essence, the Hyper-multiplexed Integrated Tensor Optical Processor represents a significant leap in computing capabilities, offering a scalable, energy-efficient solution to meet the demands of modern data-intensive applications across various fields.
