Authors: Kaiyuan Wang, Yunlong Li, Tiange Wu, Deming Liu, Shuang Zheng, Minming Zhang
Published on: February 21, 2024
Impact Score: 7.8
Arxiv code: Arxiv:2402.13447
Summary
- What is new: The paper presents an innovative photonic computing core that enables simultaneous parallel computing of multiple independent matrix-vector multiplications within the same device, exploiting the orthogonality and conversion characteristic of waveguide modes.
- Why this is important: Existing photonic computing devices for on-chip optical neural networks (ONNs) don’t fully utilize the potential of waveguide modes for parallel computation, limiting performance.
- What the research proposes: The proposed solution is an inverse-designed photonic computing core that utilizes mode conversion for parallel matrix-vector multiplication, optimizing the dielectric distribution within the device.
- Results: The computing core demonstrated the capability of performing two independent matrix-vector multiplications simultaneously, with high computing precision (relative error below 8%) and a compact size at 1550 nm wavelength.
Technical Details
Technological frameworks used: Inverse design methodology
Models used: Photonic computing cores for matrix computation
Data used: Degree of freedoms (DOFs) in photonics including space, wavelength, and mode dimensions
Potential Impact
This innovation could disrupt or benefit companies and markets involved in deep learning accelerators, high-performance computing, integrated circuits, and optical computing technologies.
Want to implement this idea in a business?
We have generated a startup concept here: PhotonIQ.
Leave a Reply