ACIM Weight Degradation Compensation via Local Digital Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analog compute-in-memory (ACIM) circuits experience degradation over time due to operational characteristics and manufacturing variations, leading to reduced accuracy in machine learning tasks, and conventional methods to address this involve remote storage and periodic rewriting, which increases data movement and resource overhead.
Innovation Solution
A device and method that utilize a digital weight storage unit local to the ACIM circuit to detect and mitigate degradation by reprogramming degraded analog weights with corresponding digital weight references, eliminating the need for host system collaboration and reducing resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional remote storage methods are used to address degradation, then weight accuracy can be maintained, but data movement and resource overhead increase
Solution Approach 1:
The patent stores digital weight references locally in the ACIM device before degradation occurs. When degradation is detected, the correct weight values are already available in local storage, eliminating the need for time-consuming remote retrieval and reducing overall resource overhead.
Solution Approach 2:
The patent creates a local copy of the weight storage unit within the ACIM device. This local copy contains digital weight references that can be quickly accessed and applied when degradation is detected, avoiding the need to repeatedly query the remote host system for weight values.
2Reliability
If periodic rewriting of weights is performed, then accuracy degradation can be mitigated, but computational resources and time are consumed
Solution Approach 1:
The patent performs preliminary action by storing digital weight references in advance within the local storage unit. When degradation is detected, the correct weight values are immediately available for rewriting, eliminating delays associated with remote data retrieval and reducing the time required for weight correction.
Solution Approach 2:
The ACIM device performs self-service by autonomously detecting degradation and rewriting weights using locally stored references. This self-contained operation eliminates the need for continuous host system involvement, reducing computational overhead and time consumption while maintaining accuracy.
3Reliability
If host system collaboration is used for weight management, then weight accuracy can be maintained, but device independence and efficiency are reduced
Solution Approach 1:
The patent creates a local copy of the weight storage unit within the ACIM device, enabling the device to autonomously manage its own weights. This local copy contains all necessary digital weight references, allowing the ACIM to operate independently without requiring continuous collaboration with the host system while maintaining weight accuracy.
Solution Approach 2:
The ACIM device achieves self-service capability by storing and managing its own weight references locally. When degradation occurs, the device can autonomously detect and correct weights using its internal storage, eliminating the need for host system intervention and enhancing both independence and efficiency.
Data Source
AI summary
Certain aspects of the present disclosure provide techniques for performing compute in memory (CIM) computations. A device comprises a CIM module configured to apply analog weights to input data using multiply-accumulate operations to generate an output. The device further comprises a digital weight storage unit configured to store digital weight references, wherein a digital weight reference corresponds to an analog weight of the analog weights. The device also comprises a device controller configured to program the analog weights to the CIM module, cause the CIM module to process the input data, and reprogram one or more analog weights that are degraded. The digital weight references in the digital weight storage unit are populated with values from a host processing device. Degraded analog weights in the CIM module are reprogrammed based on the corresponding digital weight references from the digital weight storage unit without reference to the host processing device.