Chinese scientists have developed a reconfigurable integrated photonic chip capable of running diverse neural networks, including those for image and speech processing, with high efficiency.
Key Details
- 1The chip integrates microring resonators and Mach-Zehnder interferometers, powered by a soliton microcomb light source.
- 2Supports fully-connected, convolutional, and recurrent neural networks within a single hardware architecture.
- 3Area efficiency reaches up to 2.45 TOPS/mm² at 10 GHz frequency.
- 4Demonstrated on tasks: image classification (MNIST 92.93% accuracy, CIFAR-10 56.57%), sentiment analysis (IMDB 80.81%), and speech recognition.
- 5Device enables dual-path computation per resonator, doubling throughput versus traditional schemes.
Why It Matters
This advance shows that photonic computing chips can support the flexible and scalable deployment of multi-model neural networks for imaging and multimodal AI tasks on efficient, high-throughput hardware. Such technology may lead to breakthroughs in hardware acceleration for radiology AI and similar domains.
Source
EurekAlert
Related News
•EurekAlert
Major Study Reveals Barriers to Implementing AI Chest Diagnostics in NHS Hospitals
A UCL-led study identifies significant challenges in deploying AI tools for chest diagnostics across NHS hospitals in England.
•EurekAlert
AI Model Enhances Prediction of Infection Risks from Oral Mucositis in Stem Cell Transplant Patients
Researchers developed an explainable AI tool that accurately predicts infection risks related to oral mucositis in hematopoietic stem cell transplant patients.
•EurekAlert
AI-Enabled Hydrogel Patch Provides Long-Term High-Fidelity EEG and Attention Monitoring
Researchers unveil a reusable hydrogel patch with machine learning capabilities for high-fidelity EEG recording and attention assessment.