SRAM Screening for RTN Stability via Two-Stage Guardband Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern SRAMs with sub-micron feature sizes face issues with variability in electrical characteristics, leading to read and write failures due to device variability, negative bias temperature instability (NBTI), and random telegraph noise (RTN), which result in yield loss and false failures during manufacturing testing.
Innovation Solution
A method of testing SRAMs using a two-stage guardband approach, where memory cells are initially tested at a high guardband voltage to screen out predictable time-dependent degradation and RTN effects, followed by repetitive testing at a reduced guardband voltage to identify cells susceptible to RTN, thereby reducing yield loss and identifying stable cells for system use.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a single-stage functional test is used to screen memory cells, then testing time is reduced, but yield loss increases due to false failures from RTN effects
Solution Approach 1:
The testing process is divided into two distinct stages: a first functional test at a first bias voltage level to screen out predictable time-dependent degradation, and a second functional test at a second bias voltage level to detect RTN effects. This segmentation allows each stage to focus on specific failure modes, improving overall yield accuracy without significantly increasing total test time.
Solution Approach 2:
The first functional test at the higher bias voltage level performs preliminary screening to remove memory cells with predictable time-dependent degradation before the second test. This preliminary action reduces the number of cells that need to be tested in the second stage, improving efficiency while maintaining reliability.
2Reliability
If a high guardband voltage is used in functional testing, then time-dependent degradation is screened out, but RTN effects are not detected
Solution Approach 1:
The testing method changes the bias voltage parameter between two stages: the first test uses a higher bias voltage level that screens out time-dependent degradation, while the second test uses a lower bias voltage level that makes RTN effects visible. This parameter change allows the test to detect different failure modes in different stages.
3Object-affected harmful factors
If a low guardband voltage is used in functional testing, then RTN effects are detected, but time-dependent degradation is not screened out
Solution Approach 1:
The testing process is segmented into two stages with different bias voltage levels. The first stage uses a higher voltage to screen time-dependent degradation, while the second stage uses a lower voltage to detect RTN effects. This segmentation ensures both failure modes are addressed without requiring a single test to do both simultaneously.
4Reliability
If repetitive testing at reduced guardband voltage is performed, then RTN-susceptible cells are identified, but testing time increases
Solution Approach 1:
The first functional test at higher bias voltage performs preliminary screening to remove memory cells with predictable time-dependent degradation before the second test. This preliminary action reduces the number of cells that need to be tested repeatedly in the second stage, thereby reducing total testing time while maintaining reliable identification of RTN-susceptible cells.
Data Source
AI summary
A screening method for testing solid-state memories for the effects of long-term shift and random telegraph noise (RTN). In the context of static random access memories (SRAMs), each memory cell in the array is functionally tested with a bias voltage (e.g., the cell power supply voltage) at a severe first guardband sufficient to account for worst case long-term shift and RTN effects. Cells failing the first guardband are then repeatedly tested with the bias voltage at a second guardband, less severe than the first guardband; if the tested cells pass this second guardband, the suspect cells are considered to not be vulnerable to RTN effects. Over-screening due to an unduly severe guardband is avoided, while still screening vulnerable memories from the population.