Physical Realisation TRIZ Cases
- Chiral Perovskite Carbon Nanotube Synapse for Circular Polarization Detection
H-PVK/SWNT heterostructure generates nano-ampere photocurrents for circular-polarization UV light.
- GNN Pre-processing Acceleration via COO to CSR Conversion
A conversion unit transforms coordinate list graphs into compressed sparse row format for graph neural network processing.
- Softmax Processor Accelerator with Offset Indexing
A neural network processor accelerator generates softmax values using offset-based index mapping and look-up tables.
- Broadband Diffractive Optical Neural Networks for Spectral Filtering
A multi-layer diffractive optical network framework processes a continuum of wavelengths using deep learning to design task-specific components.
- Neural Processor Circuit for Convolution Work Unit Segmentation
A neural processor circuit segments input data and kernels across multiple engine circuits to perform parallel convolution operations.
- Distributed Neural Network Training with Gradient Clipping
A distributed training apparatus clips local gradients before exchange to speed up neural network convergence.
- Neuron-Centric Local Learning Rate for Non-Volatile Memory Neural Networks
Local learning rates adapt per neuron to suppress weight dithering and prevent network freeze-out in non-volatile memory neural networks.
- Back End Floating Gate Structure for Non-Volatile Memory
Floating gate structure in back end region stores charge without power, enabling long-term data retention alongside volatile memory cells.
- Bio-Organic Photonic Neural Network for Edge Computing
Bio-organic photonic neural networks replace inorganic materials with conductive polymers, reducing energy consumption while maintaining device stability.
- Federated Learning Server Using Norm-Clipped Averaging
A collaborative training server aggregates parameter updates using norm-clipped averaging to enhance model robustness.