Quantized Grid Sampling for Tensor Interpolation

Solution Overview

Problem

Existing methods for grid sampling in semiconductor circuits often require floating-point operations, which are slower, more resource-intensive, and costly compared to quantized operations.

Innovation Solution

A method for quantized grid sampling using customized digital hardware, where a multidimensional floating-point input tensor is processed by interpolating grid points, quantizing tensor dimensions, and performing requantization and scaling operations to generate quantized grid samples.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If floating-point operations are used for grid sampling, then measurement precision is maintained, but computational speed decreases and semiconductor real estate increases

Engineering Contradiction:

Improvegrid sampling precisionVSAvoidcomputational speed

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent changes the numerical representation parameter from floating-point to quantized integer format. By representing grid coordinates and tensor values as quantized integers rather than floating-point numbers, the system maintains sufficient precision for grid sampling while enabling faster integer-based computations that are more efficient in terms of speed and hardware resource consumption.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent uses a simplified quantized integer representation instead of complex floating-point operations. The quantized format uses simpler data structures and operations that are computationally cheaper and require less hardware real estate, effectively replacing expensive floating-point units with more efficient integer arithmetic circuits.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

2Measurement precision

If floating-point operations are used for grid sampling, then measurement precision is maintained, but semiconductor real estate increases

Engineering Contradiction:

Improvegrid sampling precisionVSAvoidsemiconductor real estate

Core Design Contradiction:

Measurement precisionVSArea of stationary object

Solution Approach 1:

The patent changes the numerical representation parameter from floating-point to quantized integer format. By representing grid coordinates and tensor values as quantized integers rather than floating-point numbers, the system maintains sufficient precision for grid sampling while enabling faster integer-based computations that are more efficient in terms of speed and hardware resource consumption.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent uses a simplified quantized integer representation instead of complex floating-point operations. The quantized format uses simpler data structures and operations that are computationally cheaper and require less hardware real estate, effectively replacing expensive floating-point units with more efficient integer arithmetic circuits.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

3Measurement precision

If floating-point operations are used for grid sampling, then measurement precision is maintained, but design and development cost increases

Engineering Contradiction:

Improvegrid sampling precisionVSAvoiddesign and development cost

Core Design Contradiction:

Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent uses a simplified quantized integer representation instead of complex floating-point operations. The quantized format uses simpler data structures and operations that are computationally cheaper and require less hardware real estate, effectively replacing expensive floating-point units with more efficient integer arithmetic circuits.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

Solution Approach 2:

The patent changes the numerical representation parameter from floating-point to quantized integer format. By representing grid coordinates and tensor values as quantized integers rather than floating-point numbers, the system maintains sufficient precision for grid sampling while enabling faster integer-based computations that are more efficient in terms of speed and hardware resource consumption.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250131175A1Systems and Methods for Floating Point to Quantized Grid Sampling

Publication Date: 2025.04.24 EXPEDERA INC

AI summary

A method and system for quantized grid sampling of a tensor on customized digital hardware. The method and system receive a floating-point input tensor having one or more of tensor dimensions and a grid point that are performed in as quantized values. The quantized values generation and processing can be performed using digital hardware performing quantized operation including digital shifts and quantized multiplication and additions. For each of the one or more of tensor dimensions adjacent to the grid point a quantized-floor-index is determined. A domain shifted floor index is determined for each quantized-floor-index. Weights are determined for each quantized-floor-index. Data values of the tensor points adjacent to the grid point are accessed and a weighted sum grid sample point is generated.