Matrix Space Memory with Concurrent Row Column Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies limit access to matrix elements by only allowing direct access along rows or columns individually, requiring transformations to access elements along both dimensions simultaneously for computations.
Innovation Solution
A Matrix Space is introduced, allowing direct access and storage of matrices and arrays by both rows and columns using individual RAM cells, enabling concurrent access and computation along multiple dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a matrix is stored using multiple vector registers in row major format, then row access is efficient, but column access requires transformation and is not directly readable in a single step
Solution Approach 1:
The patent introduces a transposed view dimension alongside the original row major storage dimension. By maintaining both the original matrix layout and its transpose, the system enables direct column access without transformation while preserving row access efficiency. This dimensional extension resolves the contradiction by adding a new access perspective without compromising the original efficient row access path.
2Ease of operation
If a matrix is stored using multiple vector registers in column major format, then column access is efficient, but row access requires transformation and is not directly readable in a single step
Solution Approach 1:
The patent adds a transposed view dimension to the column major storage structure. This allows the system to maintain efficient column access through the original layout while enabling direct row access through the transposed view, eliminating the need for transformation operations and reducing access complexity.
3Device complexity
If individual row or column access is allowed, then simple access pattern is maintained, but concurrent access along multiple dimensions is not enabled
Solution Approach 1:
The patent segments the matrix access capability into independent row access paths and column access paths through separate address decoding circuits. This segmentation allows concurrent access along multiple dimensions without interference, enabling parallel computation while maintaining simple individual access patterns for each dimension.
Solution Approach 2:
By introducing transposed views as an additional dimensional perspective, the patent enables concurrent row and column access operations to proceed independently. This dimensional extension allows multiple access operations to occur simultaneously without conflict, significantly improving parallel computation productivity.
4Adaptability or versatility
If transformation of row major matrix to transpose is performed, then column access becomes possible, but computational steps increase and efficiency decreases
Solution Approach 1:
The patent pre-computes and stores transposed views of the matrix alongside the original row major storage. This preliminary action of creating transposed views during storage or initialization allows column access to be performed directly without requiring transformation operations at computation time, thereby maintaining high computational efficiency while providing versatile column access capability.
Data Source
AI summary
Mechanisms to securely store and compute with a matrix of numbers or multi-dimensional array of values in a storage entity called a matrix space. A matrix space is configured to store matrices or arrays of values into arrays of volatile or non-volatile memory cells with concurrent accessibility in multiple dimensions. Any row or column or line of storage elements in the storage entity is directly accessible for writing, reading, or clearing via row bit lines and column bit lines, concurrently. The elements in rows of the arrays are selected or controlled for concurrent access using row address lines and the elements in columns of the arrays are selected or controlled for access using column address lines, and controlled by a hardware controller. Access control methods and mechanisms with keys are used to secure, share, lock, and unlock regions in the matrix space under software and/or hardware control.