Compute-Near-Memory Protection With Combined ECC and Residue Codes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In compute-near-memory (CNM) or compute-in-memory (CIM) systems, traditional data protection mechanisms for storage and computation units introduce latency and waste power due to the need for decoding and re-encoding when moving data between these units, which are not clearly defined.
Innovation Solution
Implementing a combined two-dimensional (2D) code that merges error correction code (ECC) and residue checking to provide end-to-end data protection, ensuring data integrity with minimal impact on power usage or latency by pre-calculating residue codes and combining them with ECC codes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate ECC and residue protection mechanisms are applied in CNM systems, then data integrity is improved, but latency and power consumption increase
Solution Approach 1:
The patent combines separate ECC and residue protection mechanisms into a unified end-to-end protection system. The memory device applies both ECC codes and residue codes to data before storing in memory cells, and performs joint verification of both codes during read operations. This merging eliminates the need for separate protection stages, reducing latency while maintaining comprehensive data integrity coverage across the entire data path from memory to compute unit.
Solution Approach 2:
The unified protection mechanism serves multiple functions simultaneously: it provides error detection and correction (ECC), data validation (residue checking), and end-to-end protection across memory and compute units. By making the protection system universal and multi-functional, the patent avoids the overhead of multiple separate protection mechanisms while maintaining all necessary data integrity functions.
2Reliability
If separate ECC and residue protection mechanisms are applied in CNM systems, then data integrity is improved, but power consumption increases
Solution Approach 1:
The patent merges separate ECC and residue protection mechanisms into a unified system that operates simultaneously. The memory device applies both ECC and residue codes during write operations and verifies both during read operations in an integrated manner. This combining reduces the total power consumption compared to executing separate protection mechanisms independently, while maintaining comprehensive data integrity protection.
Solution Approach 2:
The protection mechanism performs preliminary encoding of both ECC and residue codes during the write operation before data is stored in memory cells. By pre-calculating and storing both protection codes upfront, the system avoids the need for repeated heavy computation during read operations, thereby reducing power consumption while ensuring data integrity.
3Reliability
If traditional protection mechanisms are used in CNM systems, then data corruption is protected against, but performance deteriorates due to decoding and re-encoding
Solution Approach 1:
The patent combines ECC and residue protection into a unified end-to-end mechanism that is applied once during data writing to memory and verified once during reading. This eliminates the traditional approach of separate decoding and re-encoding operations when data moves between memory and compute units, thereby maintaining strong data protection while improving performance by removing redundant processing steps.
Data Source
AI summary
A near memory compute system includes multiple computation nodes, such as nodes for parallel distributed processing. The nodes include a memory device to store data and compute hardware to perform a computation on the data. Error correction code (ECC) logic performs ECC on the data prior to computation on the data by the compute hardware. The node also includes residue check logic to perform a residue check on a result of the computation.