Computational Memory Out-of-Phase Refresh for AI Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI computation hardware faces bottlenecks due to data congestion and performance reduction caused by data movement between memory and CPU, exacerbated by issues like read disturb, retention loss, and endurance in existing memories, which are compounded by time-consuming data refresh operations that degrade inference accuracy and performance.
Innovation Solution
An electronic device with a computational memory comprising a first and second memory array, where both arrays store the same data and perform operations out of phase, allowing one to read while the other refreshes, thereby ensuring continuous data availability and accuracy without wasting time on refresh operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data refresh operation is performed to compensate retention loss, then data accuracy is improved, but time consumption increases and computing performance decreases
Solution Approach 1:
The memory array is divided into multiple banks, allowing parallel operation of refresh and read tasks across different segments. This segmentation enables simultaneous execution of refresh operations on one bank while read operations occur on another bank, eliminating the time penalty of sequential refresh operations.
Solution Approach 2:
Refresh operations are performed in advance during idle periods or in parallel with computation tasks, ensuring data is refreshed before it is needed for computation. This preliminary action prevents data loss without interrupting the computational workflow.
2Power
If data is stored in memory and transferred to CPU for computation, then computing capability is improved, but data congestion occurs and computing performance is reduced
Solution Approach 1:
The patent merges storage and computation functions into a single in-memory computing system. Compute-in-memory units are integrated directly within the memory array, allowing data to be processed in place without transfer to external CPU, thereby eliminating data congestion and improving overall computing performance.
3Speed
If data is continuously read from memory, then computing speed is improved, but read disturb and retention loss occur
Solution Approach 1:
The system implements periodic refresh operations at predetermined intervals to counteract retention loss and read disturb effects. These periodic actions are scheduled to occur during computation tasks, ensuring data integrity is maintained without continuously interrupting the computational process.
Data Source
AI summary
An electronic device is provided. The electronic device comprises a computational memory. The computational memory comprises a first memory array and a second memory array. The first memory array and the second memory array contain same data. The first memory array and the second memory array are configured to perform operations in an out of phase manner.