Mixed Digital Analog Computational Storage Power Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data storage systems for Machine Learning (ML) applications face challenges in achieving power efficiency while maintaining accuracy, particularly in edge ML devices with limited power sources.
Innovation Solution
A mixed digital and analog computational storage system that selectively directs computational tasks to either a digital processor or an analog in-memory compute unit based on parameters such as power, precision, and workload, leveraging the power efficiency of analog computing for tasks like convolution in neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If digital processing is used for all computational tasks, then precision and reliability are maintained, but power consumption increases significantly
Solution Approach 1:
The system segments computational tasks into two distinct domains: analog in-memory compute unit for power-efficient operations and digital processor for precision-critical operations. This segmentation allows each domain to handle tasks suited to its strengths, resolving the contradiction between power efficiency and precision by distributing workloads appropriately rather than using a single processing approach for all tasks.
Solution Approach 2:
The controller acts as an intermediary between the analog in-memory compute unit and the digital processor. It receives computational tasks, evaluates them based on precision requirements and available resources, and routes them to the appropriate processing unit. This intermediary coordination enables the system to maintain overall precision while leveraging the power efficiency of analog computing for suitable tasks.
2Use of energy by moving object
If analog computing is used for all computational tasks, then power consumption is reduced, but computational precision deteriorates
Solution Approach 1:
The system segments computational tasks into two distinct domains: analog in-memory compute unit for power-efficient operations and digital processor for precision-critical operations. This segmentation allows each domain to handle tasks suited to its strengths, resolving the contradiction between power efficiency and precision by distributing workloads appropriately rather than using a single processing approach for all tasks.
Solution Approach 2:
The controller continuously monitors the computational tasks and their requirements, using feedback about precision needs and resource availability to dynamically route tasks between analog and digital processing domains. This feedback mechanism ensures that analog computing is used maximally for power efficiency while maintaining sufficient precision through digital processing where needed.
3Use of energy by moving object
If a mixed digital and analog computational storage system is implemented, then power efficiency is improved, but device complexity increases
Solution Approach 1:
The controller is designed as a multi-functional unit that performs several critical functions: managing memory operations, evaluating computational tasks, determining routing decisions between analog and digital domains, and coordinating data flow. By consolidating these diverse functions into a single controller, the system achieves power-efficient mixed-domain computing without proportionally increasing overall system complexity.
4Use of energy by moving object
If computational tasks are routed to analog domain, then power consumption is reduced, but task routing decision complexity increases
Solution Approach 1:
The controller autonomously evaluates computational tasks and makes self-directed routing decisions based on predefined criteria for precision requirements and available resources. This self-service capability eliminates the need for external intervention in task routing, simplifying the decision-making process while maintaining power efficiency through intelligent workload distribution between analog and digital domains.
Data Source
AI summary
Various embodiments of this disclosure are directed to a mixed digital and analog domain approach to computational storage or memory applications. The mixed approach enables certain compute operations to be advantageously performed in the analog domain, achieving power saving. In some embodiments, an analog compute core is implemented based on a first set of memory elements that are made available with a second set of memory elements for digital data storage. A controller coupled to both sets of memory elements is able to selectively direct computational tasks to either the analog compute core or a digital processor coupled with the controller, based on one or more parameters including power, precision, and workload. In certain embodiments involving neural network tasks, the controller is configured to route certain tasks to the analog compute core based on neural network based factors such as network layer positioning and input signal type.