English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Silicon photonic neural networks (PNNs) utilizing wavelength-multiplexingapproaches, such as those based on micro-ring resonators (MRRs), offer superiorperformance due to their weight assignment, exceptional energy efficiency, andhigh computing density achieved through small device footprints. However, MRRsare highly sensitive to ambient disturbance, with even slight resonance driftsof tens of picometers potentially compromising MRRs-based PNNs. Currentsolutions often rely on complex control methods, resulting in high hardwarecomplexity impractical for large-scale PNNs. Here, we propose a novelhardware-aware training and pruning approach. The core idea is to train theparameters of a physical neural network towards its noise-robust andenergy-efficient region. This innovation enables control-free andenergy-efficient photonic computing. Experimentally, this method improvescomputing precision by 4 bits without requiring intricate MRR control orpower-consuming temperature stabilization circuits. It notably increases theaccuracy of experimental handwritten digit classification from 67.0%, initiallyimpacted by the thermal variances, to 95.0%, a figure comparable to thetheoretical value and achieved without a thermoelectric controller.Additionally, this approach reduces the energy required to tune the MRRsten-fold. Furthermore, our numerical verification indicates that this method isbroadly applicable across various NN sizes and machine-learning tasks. Itseffectiveness is particularly marked in larger NNs, where it boosts accuracyfrom 10.9% to 98.0% and slashes power consumption by 160 times. Thisadvancement represents a significant step towards the practical,energy-efficient, and noise-resilient implementation of large-scale integratedPNNs.

Control-free and efficient silicon photonic neural networks via hardware-aware training and pruning