Computational SSD Error Correction for On-Die Semantic Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional AI/ML implementations of semantic searching in storage devices face inefficiencies due to data movement between the host processor, host memory, and storage device, leading to bandwidth and energy waste, and a bottleneck for the storage device controller handling both I/O and computing operations.
Innovation Solution
Distribute computing to each NAND flash die of the SSD with on-die AI/ML processing units that perform computation and comparison operations, with the SSD controller aggregating results, reducing data movement and alleviating the bottleneck.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If data is repeatedly moved between the host processor, host memory, and storage device for AI/ML processing, then the storage device can perform semantic searching, but substantial bandwidth and energy are wasted
Solution Approach 1:
The patent segments the AI/ML processing functionality by distributing computing resources to each NAND flash die through on-die processing units. This allows semantic searching to be performed locally on the storage device without repeatedly moving data between host processor, host memory, and storage device, thereby eliminating substantial bandwidth and energy waste while maintaining semantic searching capability.
Solution Approach 2:
The patent adds a new dimension of computation by integrating AI/ML processing units directly into the storage device's die structure. This dimensional shift from external host-based processing to embedded on-die processing enables semantic searching to be performed in-place, eliminating the need for repeated data movement and reducing energy consumption.
2Adaptability or versatility
If the storage device controller handles both I/O and computing operations for AI/ML processing, then semantic searching can be performed, but a bottleneck is created for the controller
Solution Approach 1:
The patent segments the computing workload by distributing AI/ML processing across multiple on-die processing units, each capable of independent computation. This segmentation allows the storage device controller to handle I/O operations while the distributed on-die units perform computing operations in parallel, eliminating the bottleneck that would occur if the controller had to handle both types of operations sequentially.
Solution Approach 2:
The patent introduces on-die processing units as intermediary components between the storage media and the controller. These intermediary units perform AI/ML computing operations locally, freeing the controller from having to handle computationally intensive tasks and allowing it to focus on I/O operations, thereby maintaining high throughput for both functions.
3Adaptability or versatility
If data is buffered in memory for the controller to handle AI/ML processing, then computation can be performed, but additional memory buffer is required
Solution Approach 1:
The patent segments the computational workload across multiple on-die processing units that can perform AI/ML operations directly on the storage die. This eliminates the need to buffer data in external memory, as each on-die unit can process data locally, thereby reducing the quantity of memory buffer required while maintaining full computational capability.
Solution Approach 2:
The patent enables the storage device to serve its own computational needs through integrated on-die processing units. These units perform AI/ML operations directly on the storage media without requiring external memory buffering, allowing the system to be self-sufficient and eliminating the need for additional memory buffer resources.
Data Source
AI summary
Devices and methods to implement semantic searching on SSD through a computational SSD system that distributes computing to each NAND flash die of the SSD while the SSD controller handles the results aggregation with new on-die computation logic circuits to provide on device file semantic search are disclosed herein. The computational SSD system can read file feature vectors from multiple dies to the SSD controller, and if needed, these feature vectors may be buffered in DRAM and controller handles distance computing. Local, on-die AI/ML processing units may perform, for example, computation and comparison operations and pass the processing scores and results to the SSD controller. The SSD controller aggregates results from all dies and returns the result to the host. The feature vector store size, circuitry and number of on-die AI/ML processing units may be configured as needed to adapt to different tasks, system constraints, and/or feature vector sizes.