EPLC Sensing Scheme for CAM Search Timing and Power
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Content addressable memory (CAM) systems face performance bottlenecks due to random device variation (RDV) as technology scales to submicron geometries, leading to increased timing uncertainty and power inefficiency in search operations, particularly in the two-stage sensing scheme where the main-compare is delayed by conservative margining for variation.
Innovation Solution
The Early-Predict Late-Correct (EPLC) sensing scheme allows the main-search operation in a two-stage CAM search to start before the pre-search is fully complete, based on preliminary results, and may be terminated or completed based on final pre-search results, using an inverting Schmitt trigger with tunable hysteresis to optimize timing and power efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conservative margining is applied for random device variation in two-stage sensing, then reliability is improved, but search time increases and productivity deteriorates
Solution Approach 1:
The pre-search operation performs preliminary comparison of a subset of bits before the main-search. By completing the pre-search fully before initiating main-search, the system ensures reliable preliminary filtering while enabling early termination of main-search when pre-search fails, thus improving overall productivity without sacrificing reliability.
Solution Approach 2:
The sensing circuit uses self-referenced differential sensing where each sense amplifier compares its match-line voltage against a reference derived from the same circuit. This self-referenced approach inherently compensates for random device variation, maintaining reliability without requiring excessive conservative margining that would slow down the search operation.
2Reliability
If the main-search is delayed until pre-search completion, then sensing reliability is maintained, but search time increases and speed deteriorates
Solution Approach 1:
The system allows the main-search to proceed concurrently with the pre-search operation rather than waiting for pre-search completion. This overlapping execution enables the system to rush through the main-search in parallel, significantly reducing total search time while maintaining reliability through later correction of any premature decisions.
Solution Approach 2:
The system implements late correction by using the final pre-search results to validate or correct the main-search output. This feedback mechanism ensures that even if the main-search starts prematurely, the final result is corrected based on the complete pre-search, maintaining high reliability while enabling speed improvement through overlapping operations.
3Reliability
If full-swing sensing is used to maintain reliable operation, then sensing margin is sufficient, but power consumption increases
Solution Approach 1:
The system uses partial sensing by dividing the search into two stages: pre-search comparing a subset of bits and main-search comparing the remaining bits. This partial action approach allows the use of lower-swing, lower-power sensing circuits in each stage rather than requiring full-swing sensing across all bits simultaneously, reducing power consumption while maintaining sufficient sensing margin through the two-stage architecture.
Solution Approach 2:
The sensing operation is segmented into pre-search and main-search phases, each handling different portions of the bit comparison. This segmentation allows each phase to use optimized, lower-power sensing circuits tailored to its specific requirements, rather than using high-power full-swing sensing for the entire operation, thus reducing overall power consumption while maintaining reliability.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach improves CAM search performance by about 30% while maintaining low power consumption, achieving 1 Gsearch/sec throughput in a 32 nm High-K Metal Gate SOI process with minimal power impact, and is applicable to various two-stage memory operations.
Implementation Method 1
using an inverting Schmitt trigger with tunable hysteresis to optimize timing and power efficiency
Data Source
AI summary
Circuits and methods for performing search operations in a content addressable memory (CAM) array are provided. A system for searching a CAM includes a circuit that selectively activates a main-search of a two stage CAM search while a pre-search of the two stage CAM search is still active.