Biological Neural Network Models TRIZ Cases
- Codebook-Based Data Quantization for Lower Transmission Overhead
Codebook selection based on data distribution cuts transmission overhead while preserving quantization performance between devices.
- Machine-Learning ADC With Parallel Comparators for Nonlinear Signals
Parallel comparators plus machine learning convert analog signals despite nonlinear behavior, manufacturing variation, and rigid interface limits.
- Machine-Learning ADC Calibration for Nonlinear Analog Signals
A parallel-array ADC uses machine-learning calibration to map nonlinear analog behavior into accurate digital output without idealized circuitry.
- Delta-Sigma Modulation With Neural Compensation for Spectrum Leakage
A neural network predicts transmission distortion in delta-sigma modulation to suppress spectrum leakage and preserve signal quality.
- CTU Block Partitioning in Neural Network Model Compression
CTU block partitioning signals tensor partitioning in compressed neural networks, cutting storage and compute for constrained devices.
- SSD Two-Level Parity Protection for Extreme Temperature Reliability
Two-level parity protection strengthens SSD data retention under extreme temperatures by adding stronger ECC to error-prone data regions.
- Machine-Learning ADC Calibration for Non-Ideal Analog Circuits
Calibrating a machine-learning ADC against a physical parallel array converter improves conversion accuracy when ideal analog circuitry is impractical.
- SSD Two-Level Parity for High-Temperature Data Reliability
Two-level parity protection strengthens SSD error correction for data stored under elevated temperatures and retention-related error risk.
- Edge AI Training Compression Using Cluster-Based Surrogate Data
Lossy cluster encoding cuts IoT-to-cloud data transfer while surrogate sampling and adversarial filtering improve AI training robustness.
- 3D Block Partitioning for Neural Network Model Compression
3D pyramid block partitioning compresses neural network tensors to cut storage and decoding load on resource-constrained devices.
- AI Signal Analysis Using Bicoherence for Jitter Classification
Bicoherence and neural networks separate jitter from other serial signal distortions, improving waveform measurement and clock recovery.
- Single-MOSFET Charge Multiplier for Low-Transistor Neural Computing
An analog MOSFET-capacitor multiplier cuts transistor count for neural computing while preserving multiplication accuracy through charge-based operation.