Column-Based Data Layouts for Clustered Memory Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large-scale similarity searches in content-based systems face challenges due to high computational and memory intensity, especially in databases with billions of entries, where search speed does not meet current throughput demands, and stochastic associative memories provide imperfect results due to imperfect data conversion across domains.
Innovation Solution
The implementation of a compute device with column-read enabled memory that uses a three-dimensional cross-point architecture, allowing bit-level addressability and efficient data access through media access circuitry, which enables column data layouts that reduce the number of read operations and improve read performance by avoiding multiple reads on cluster edges and modulo penalties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If hashing methods are used to perform stochastic associative searches, then search speed is improved, but search accuracy is degraded
Solution Approach 1:
The patent introduces a hybrid search architecture that uses hashing methods as an intermediary for initial fast filtering, then transitions to exact space searches for final accurate matching. This mediator approach allows the system to benefit from both the speed of hashing and the accuracy of exact space searches, resolving the contradiction between search speed and search accuracy.
2Productivity
If the database is partitioned into clusters to improve search speed, then throughput is improved, but memory access complexity increases
Solution Approach 1:
The patent divides the database into multiple clusters, each with its own representative vector, enabling parallel search operations that improve throughput. The segmentation is organized hierarchically with cluster-level indexing and within-cluster data structures, which manages the complexity through structured organization rather than random access patterns.
Solution Approach 2:
The patent transforms the memory access pattern by organizing data in a column-based layout where related data elements are stored together in vertical columns rather than traditional row-based horizontal storage. This dimensional change in data organization enables more efficient access patterns for clustered data, reducing the complexity of memory access while maintaining high throughput.
3Speed
If column-based data layouts are used to reduce read operations, then read performance is improved, but data storage organization complexity increases
Solution Approach 1:
The patent implements a column-based data layout where data is organized vertically by feature dimension rather than horizontally by data sample. This dimensional reorganization allows the system to read entire columns of related features in single memory operations, dramatically improving read performance. The complexity is managed through systematic columnar organization and indexing structures that map logical data access to physical storage locations.
Data Source
AI summary
Technologies for media management for providing column data layouts for clustered data include a device having a column-addressable memory and circuitry connected to the memory. The circuitry is configured to store a data cluster of a logical matrix in the column-addressable memory with a column-based format and to read a logical column of the data cluster from the column-addressable memory with a column read operation. Reading the logical column may include reading logical column data diagonally from the column-address memory, including reading from the data cluster and a duplicate copy of the data cluster. Reading the logical column may include reading from multiple complementary logical columns. Reading the logical column may include reading logical column data diagonally with a modulo counter. The column data may bread from a partition of the column-address memory selected based on the logical column number. Other embodiments are described and claimed.