Parallel Select Circuit for Ordered Data Stream Processing

Solution Overview

Problem

Existing data processing systems require large amounts of memory and processing power to handle large volumes of ordered input data, leading to increased costs, size, and power consumption, particularly in applications like 3D beam forming and statistical analysis.

Innovation Solution

A hardware architecture that enables highly parallelized processing of streams of ordered input data using select circuits with interval generation circuitry to determine the number of input data values to be skipped, reducing the need for indexing and overhead, and incorporating transform and reduce circuits for efficient data processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional memory-based processing is used to handle large volumes of ordered input data, then data processing capability is achieved, but memory requirements and processing power requirements increase significantly

Engineering Contradiction:

Improvedata processing capabilityVSAvoidmemory requirements

Core Design Contradiction:

ProductivityVSQuantity of substance

Solution Approach 1:

The patent divides the processing system into multiple parallel select circuits that each handle a portion of the input data stream. Instead of loading all data into memory and processing sequentially, the system segments the data stream and processes multiple segments simultaneously through parallel circuit paths, reducing peak memory requirements while maintaining high processing throughput.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent transitions from sequential processing (single dimension) to parallel processing (adding a temporal/spatial dimension). By using multiple select circuits operating simultaneously on different portions of the data stream, the system processes data in parallel across multiple circuit paths, effectively adding a dimension of concurrency that reduces the need for large memory buffers.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Productivity

If conventional memory-based processing is used to handle large volumes of ordered input data, then data processing capability is achieved, but processing power requirements increase significantly

Engineering Contradiction:

Improvedata processing capabilityVSAvoidprocessing power requirements

Core Design Contradiction:

ProductivityVSPower

Solution Approach 1:

The patent divides the processing system into multiple parallel select circuits that each handle a portion of the input data stream. Instead of loading all data into memory and processing sequentially, the system segments the data stream and processes multiple segments simultaneously through parallel circuit paths, reducing peak memory requirements while maintaining high processing throughput.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent transitions from sequential processing (single dimension) to parallel processing (adding a temporal/spatial dimension). By using multiple select circuits operating simultaneously on different portions of the data stream, the system processes data in parallel across multiple circuit paths, effectively adding a dimension of concurrency that reduces the need for large memory buffers.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Ease of operation

If data values are stored in memory and indexed for processing, then data retrieval is achieved, but overhead and processing time increase

Engineering Contradiction:

Improvedata retrieval capabilityVSAvoidprocessing overhead

Core Design Contradiction:

Ease of operationVSLoss of time

Solution Approach 1:

The patent pre-organizes the data into ordered streams with predetermined interval patterns before processing begins. Select circuits are pre-configured with interval values that determine which data points to select, eliminating the need for runtime indexing calculations. This preliminary organization allows direct selection of data points based on pre-computed intervals, reducing processing overhead and access time.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements skip-based selection where select circuits directly jump to and retrieve specific data points based on pre-determined interval patterns, rather than sequentially accessing or indexing through all data points. This skipping mechanism allows the system to rush through irrelevant data points and directly access only the required values, significantly reducing access time and overhead.

Inventive Principle:

Principle #21Skipping (Rushing through)

Data Source

PatentUS9691374B2Processing a stream of ordered input data

Publication Date: 2017.06.27 THE RGT UNIV OF MICHIGAN

AI summary

A data processing system is provided for performing processing operations upon an ordered stream of input data values to form an ordered stream of output data values. A select circuit (18) includes select interval generation circuitry (34) which determines a number (interval number) of input data values between each data value to be selected for output from among the ordered stream of input data values. This interval number varies with position within the ordered stream of input data values. The select circuit (18) can thus perform selection of input data values in accordance with an interval number which may be varied, for example, in accordance with a linear piecewise approximation of an desired curve or, in other embodiments, in a piecewise quadratic variation approximating a desired curve. The processing techniques may be used, for example, in beam forming application, such as 3D beam forming of ultrasonic images.