Logarithmic/Exponential Functions TRIZ Cases
- Digital Temperature Compensation in Logarithmic TIAs
Digital sampling and processor-based correction remove analog temperature-compensation errors in logarithmic TIA current measurement.
- Multi-Head Posterior Evaluation of Pre-Trained Embedding Models
Posterior probabilities over clustered entity embeddings provide a faster way to compare pre-trained models without full downstream testing.
- Exponential Approximation Using Bit-Shift Taylor Factorials
Bit-shift factorial approximations cut exponential computation time and hardware size for neural network functions while preserving useful accuracy.
- Exponential Approximation Using Bit-Shift Taylor Factorials
Bit-shift factorial approximation cuts exponential computation time and hardware area for softmax while preserving neural network accuracy.
- Neural Processor Encoding with Fixed Exponents and Variable Mantissas
A mapping table, comparator, and shifter balance numeric range, computation speed, and memory use in neural parameter processing.
- Piecewise Function Evaluation via Lookup Table Segmentation
Segmented lookup tables minimize approximation errors while maintaining single-cycle computation efficiency for complex function evaluation.
- Unified Hardware Pipeline for Transcendental Functions
Shared lookup tables and interpolation logic in a single pipeline reduce gate count while maintaining precision for graphics processor chips.
- Unified Hardware Pipeline for Transcendental Function Computation
A unified hardware pipeline computes transcendental functions using shared data look-up tables and interpolation circuits.
- Risky Vertex Identification via Stacked Medium Networks
Constructing multiple medium networks from black seeds to determine final risk values for high-risk vertices.
- Datapath Circuit Segmentation for DSP Hard Functions
Segmenting the datapath between a MAC and a lookup table calculator resolves power consumption trade-offs while accelerating logarithm execution.