NVM-Integrated SRAM CIM Cell for Local Data Recall
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Compute-in-memory (CIM) systems face inefficiencies due to the volatility of SRAM, which requires data retrieval from distant memory sources, leading to high power consumption and performance bottlenecks during data recall and CIM operations.
Innovation Solution
Integration of non-volatile memory (NVM) into SRAM CIM cells, enabling data storage and recall within the NVM, reducing power consumption and improving performance by eliminating the need for data transfer from distant memory sources, while also reducing area overhead in integrated circuits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If SRAM is used for compute-in-memory operations, then fast data access is achieved, but data volatility causes data loss when power is removed
Solution Approach 1:
The patent combines SRAM and NVM into a single hybrid memory cell structure, allowing the system to leverage both the fast access speed of SRAM and the non-volatile data retention of NVM. The SRAM portion provides rapid data access for compute operations while the NVM portion maintains data persistence across power cycles, effectively merging the advantages of both memory types to resolve the contradiction between speed and reliability.
Solution Approach 2:
The hybrid memory cell is designed to perform multiple functions: it operates as volatile SRAM during active compute-in-memory operations for high-speed access, and as non-volatile memory for data persistence when power is removed. This multi-functionality allows the same physical structure to address both the speed requirement during operation and the reliability requirement during power-off states.
2Reliability
If data is stored in distant memory sources, then non-volatile storage is achieved, but power consumption increases during data retrieval
Solution Approach 1:
By merging SRAM and NVM into a single hybrid cell located at the same memory location, the patent eliminates the need to physically transfer data from distant memory sources. The data resides locally in both volatile and non-volatile portions of the same cell, allowing immediate access without energy-intensive data movement operations while maintaining data persistence.
Solution Approach 2:
The NVM portion acts as an intermediary that retains data locally without requiring external memory sources. Instead of fetching data from distant storage, the system directly accesses the NVM component within the hybrid cell, using it as a local buffer that eliminates the need for long-distance data transfer and associated power consumption.
3Reliability
If SRAM cells are expanded to include NVM, then data retention is improved, but area overhead increases
Solution Approach 1:
The patent merges SRAM and NVM structures into a shared physical footprint, where both memory types coexist within the same memory cell area. This integration approach allows the hybrid cell to achieve non-volatile data retention while utilizing the same silicon real estate as traditional SRAM cells, thereby avoiding significant area overhead despite the added functionality.
Data Source
AI summary
A memory device including a static random-access memory that includes two cross-coupled inverters and an access transistor having a gate connected to a word line. The memory device further includes one or more logic gates electrically coupled to the static random-access memory, and a non-volatile memory electrically coupled to the static random-access memory and configured to store data and be read using the static random-access memory, wherein the non-volatile memory is connected on one side to the access transistor and on another side to the two cross-coupled inverters.


