Bandwidth Adaptive Memory Compression Controller
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory compression schemes face challenges in efficiently managing bandwidth utilization, particularly in determining whether to access data in compressed or uncompressed mode, leading to high latency and large on-chip structures for storing compression states.
Innovation Solution
A bandwidth adaptive memory controller that dynamically determines the bandwidth limited condition based on memory bus utilization or queue depth, switching between compressed and full access modes to optimize data retrieval, thereby reducing latency and conserving bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If memory compression schemes use large on-chip data structures to store compression state, then the ability to determine compressed/uncompressed state is improved, but the device complexity and on-chip memory requirements increase
Solution Approach 1:
The patent extracts the compression state information from separate on-chip data structures and embeds it directly within the memory data itself. Specifically, the most significant bit (MSB) of the compressed data block is used to indicate whether the block is compressed or uncompressed, eliminating the need for large external state storage structures.
Solution Approach 2:
The patent makes the memory data bits serve multiple functions: the MSB simultaneously serves as both data (when uncompressed) and as the compression state indicator. This multi-functionality reduces the need for dedicated state storage resources while maintaining reliable compression state determination.
2Loss of energy
If the system encodes compression state in main memory and accesses data assuming compressed first, then bandwidth usage is reduced, but the access latency increases for uncompressed data
Solution Approach 1:
The patent segments the data access process into two paths based on the compression state indicator: a fast path for uncompressed data (where the MSB indicates no compression) and a decompression path for compressed data. This segmentation allows uncompressed data to be accessed with minimal latency while still enabling bandwidth savings for compressed data.
Solution Approach 2:
The patent changes the interpretation of the data bit parameters based on the compression state. The MSB is interpreted differently depending on whether compression is detected, allowing the system to optimize access patterns dynamically without incurring penalty latency for uncompressed data.
3Loss of time
If the system always accesses full-size blocks, then access latency is minimized, but bandwidth utilization efficiency decreases when compression is applicable
Solution Approach 1:
The patent implements a dynamic access strategy where the memory controller adapts its access pattern based on the compression state indicator in the data. When the MSB indicates compressed data, the controller accesses only the reduced-size block, dynamically adjusting bandwidth usage to match actual data requirements and improving overall bandwidth efficiency.
Data Source
Figure 1~2
Figure 3
AI summary
Data is retrieved from system memory in compressed mode if a determination is made that the memory bus is bandwidth limited and in uncompressed mode if the memory bus is not bandwidth limited. Determination of the existence of the bandwidth limited condition may be based on memory bus utilization or according to a depth of a queue of memory access requests.