Fully non-linear neuromorphic computing with linear wave scattering
Abstrak
The increasing size of neural networks for deep learning applications and their energy consumption create a need for alternative neuromorphic approaches, e.g., using optics. Current proposals and implementations rely on physical non-linearities or opto-electronic conversion to realise the required non-linear activation function. However, there are significant challenges with these approaches related to power levels, control, energy-efficiency, and delays. Here, we review our scheme [Nat. Phys. 20, 1434–1440 (2024)] for a neuromorphic system that relies on linear wave scattering and yet achieves non-linear processing with a high expressivity. The key idea is to encode the input in physical parameters that affect the scattering processes. Moreover, gradients needed for training can be directly measured in scattering experiments. We propose an integrated-photonics implementation based on racetrack resonators that achieves high connectivity with a minimal number of waveguide crossings. Our work opens the door to a new, easily implementable paradigm of neuromorphic computing that can be widely applied in existing state-of-the-art, scalable platforms, such as optics, microwave and electrical circuits.
Topik & Kata Kunci
Penulis (2)
C. C. Wanjura
F. Marquardt
Akses Cepat
- Tahun Terbit
- 2023
- Bahasa
- en
- Total Sitasi
- 76×
- Sumber Database
- Semantic Scholar
- DOI
- 10.1038/s41567-024-02534-9
- Akses
- Open Access ✓