Physical Realisation TRIZ Cases
- MAFT-ONN Optical Neural Network Architecture for Low Latency Inference
Multiplicative analog frequency transform optical neural network encodes data in the frequency domain to perform matrix-vector products.
- Configurable Accelerator Framework for Mobile DCNN Processing
A configurable accelerator framework routes data streams between convolution accelerators and DSP clusters to perform efficient neural network operations.
- Aperture Function IC for Neural Network Row Processing
An integrated circuit applies an aperture function to input data streams ordered row by row.
- Neuromorphic Neural Network Noise Leveraging via Low-Rank Approximation
A noise leveraging method generates a low-rank matrix from neuromorphic device covariance to adjust neural network weights.
- Runtime Predictors for ANN Computation Reduction
A runtime predictor generates simplified outputs to identify salient neural network computations.
- Neural Memory Network for IoT Edge Data Management
Neural memory networks segment and nest data layers to reduce redundant storage while maintaining high capacity for IoT edge devices.