Embedded ML Inference in Storage SoCs Without Hardware Rework
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current storage devices face challenges in improving processing power and efficiency due to the complexity of managing variables and resources, with traditional methods often requiring costly hardware changes and external processing that can fail when communication is lost.
Innovation Solution
Implementing machine learning models within the System on a Chip (SoC) of storage devices to dynamically generate inferences and process data without additional capital investment, allowing for adjustable model complexity and processing time, and enabling efficient management of data attributes without manual firmware adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If additional processors or specialized components are added to improve internal processing power, then processing capability is improved, but manufacturing cost increases significantly
Solution Approach 1:
The existing SoC processor is made multi-functional by implementing machine learning inference capabilities alongside traditional storage control functions. The same processor core executes both conventional firmware and machine learning models, eliminating the need for dedicated ML hardware components while maintaining enhanced processing capabilities.
Solution Approach 2:
The system dynamically adjusts processing parameters by loading different machine learning models with varying complexity levels based on available resources and performance requirements. This allows the existing hardware to adapt its computational capacity without physical modifications, optimizing the trade-off between processing power and manufacturing cost.
2Power
If hardware design is reworked to add processing components, then processing capability is improved, but device complexity increases
Solution Approach 1:
The existing SoC is designed to perform multiple functions including traditional storage control and machine learning inference. By making the processor universal rather than adding specialized components, the hardware complexity remains unchanged while processing capability is enhanced through software-based machine learning models.
Solution Approach 2:
Instead of creating new hardware circuits for machine learning processing, the system uses software copies of processing logic implemented as machine learning models. These models replicate the functionality of specialized hardware through algorithmic computation, avoiding additional hardware complexity.
3Power
If processing is offloaded to external host system, then processing capability is improved, but system reliability decreases due to communication dependency
Solution Approach 1:
The storage device performs machine learning processing independently using embedded models within the SoC, rather than relying on external host systems. This self-service capability allows the device to execute inference tasks locally without communication dependencies, maintaining reliability while providing enhanced processing capability for tasks such as data classification and anomaly detection.
Data Source
AI summary
Methods are provided for tactically deploying machine learning operations within existing storage devices without the need for additional capital investment. Machine learning operations are specifically designed to locate and evaluate multiple types of data to complete an operation, including synthesizing missing data. These operations may be processed within a SoC of a storage device as embedded software. Storage devices designed to utilize machine learning methods within existing configurations can include a non-volatile memory for storing data, executable instructions, and a processor to conduct a variety of steps. The steps can include executing a plurality of applications stored in the non-volatile memory, and receiving a request for data, including measurements, from at least one of the applications. The steps can further determine if the requested data is suitable for substitution by an inference and subsequently select at least one machine learning model for generating a suitable inference.