2D Processing Array Layout for Memory-Free Data Permutation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data processing systems face challenges in efficiently performing matrix operations and convolution operations due to limitations in memory access and configurability of processing elements.
Innovation Solution
A two-dimensional array of processing elements is configured with corresponding two-dimensional arrays of data elements, where the dimensions match, allowing flexible control over processing operations and data flow direction, eliminating the need for memory accesses to permute data values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If data values are stored in separate memory structures requiring access and permutation, then data can be flexibly organized, but memory bandwidth requirements increase and processing efficiency decreases
Solution Approach 1:
The patent combines the data values and control values into a single two-dimensional array structure in memory. Each processing element corresponds to one data element in the array, eliminating the need for separate memory structures and the complex access/permission operations that would be required. This merging reduces memory bandwidth consumption while maintaining data organization flexibility.
Solution Approach 2:
The two-dimensional array structure serves multiple functions simultaneously: it stores data values, provides control information through the data element values themselves, and defines the processing configuration. This multi-functional approach eliminates the need for dedicated control structures and reduces overall memory bandwidth requirements.
2Device complexity
If processing elements have fixed configurations, then hardware complexity is reduced, but adaptability to different operations decreases
Solution Approach 1:
The patent makes the processing element configuration dynamic by using the values stored in the data elements themselves as control information. Rather than having fixed, hard-wired configurations, each processing element's behavior is determined by the data values it contains or receives, allowing flexible adaptation to different operations without increasing hardware complexity.
Solution Approach 2:
The data elements serve dual purposes: they contain the actual data to be processed and simultaneously encode the control information that determines how processing should be performed. This self-service approach eliminates the need for separate control logic, reducing hardware complexity while maintaining full operational flexibility.
3Ease of operation
If data permutation is performed through memory accesses, then data can be routed to correct processing elements, but processing time increases
Solution Approach 1:
The patent performs preliminary organization of data and control information in the two-dimensional array structure before processing begins. The data elements are pre-positioned in their final destinations with their control values already in place, eliminating the need for time-consuming runtime permutation operations through memory accesses.
Solution Approach 2:
The patent replaces the mechanical system of memory-based data permutation with a data-driven approach where control values embedded in the data elements themselves direct the processing. This substitution eliminates the time-consuming memory access cycle required for traditional data routing and permutation.
Data Source
AI summary
Apparatuses, computer programs and methods are disclosed, relating 2D arrays of data elements and a 2D array of processing elements. A first 2D array of data elements provides data values for processing by each processing element and a second 2D array of data elements provides control values controlling the processing. The 2D array of processing elements has a data flow direction across the 2D array of processing elements, and the data flow direction proceeds from a starting set of processing elements of the 2D array of processing elements. For each processing element not in the starting set of processing elements, the data processing operation preformed takes as an operand a respective data value provided by a neighbouring data element of the first 2D array of data elements and selection of the neighbouring processing element is controlled by a corresponding source control value in the second 2D array of data elements.