State Adaptive Predictive Programming for Non-Volatile Memory Cells
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing memory programming techniques face challenges in achieving tight threshold voltage distributions without sacrificing programming speed, as they often require verifying each memory cell's state, which slows down the programming process.
Innovation Solution
The implementation of state adaptive predictive programming, which applies program pulses to memory cells without verifying their target state, reducing the number of verify operations and allowing for both tight threshold voltage distributions and high programming speed by using adaptive program enable voltages based on the memory cell's state.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If verify operations are performed after each program pulse to ensure tight threshold voltage distributions, then manufacturing precision is improved, but productivity deteriorates due to reduced programming speed
Solution Approach 1:
The patent applies verify operations selectively rather than universally. Specifically, verify operations are performed only on memory cells that are likely to benefit from them based on predictive modeling, while other cells undergo predictive programming without verification. This partial application of verification maintains tight threshold voltage distributions for critical cells while avoiding the productivity penalty of verifying all cells.
Solution Approach 2:
The patent implements a feedback mechanism where verify operations provide information about actual threshold voltage attainment, which is then used to adjust and refine predictive programming parameters. This feedback loop enables the system to learn from verification results and improve the accuracy of predictive programming, reducing the number of cells that require full verification while maintaining overall distribution quality.
2Productivity
If verify operations are reduced to increase programming speed, then productivity is improved, but manufacturing precision deteriorates due to wider threshold voltage distributions
Solution Approach 1:
The patent performs preliminary predictive programming before final verification or locking of memory cell states. By using machine learning models to predict which cells will achieve target threshold voltages and adjusting program parameters accordingly, the system prepares cells in advance for optimal programming outcomes, reducing the need for extensive verify operations while maintaining precision.
Solution Approach 2:
The patent dynamically changes programming parameters such as program pulse voltage, pulse width, and number of pulses based on predictions from machine learning models. These parameter adjustments are made adaptively during the programming process to optimize both speed and precision, allowing the system to achieve tight threshold voltage distributions without requiring traditional extensive verify operations.
3Reliability
If traditional programming with full verification is used to ensure reliability, then reliability is improved, but loss of time increases due to extended programming duration
Solution Approach 1:
The patent uses feedback from verify operations to continuously refine predictive programming parameters and improve the accuracy of state predictions. This feedback mechanism allows the system to maintain high reliability by learning from actual verification results and adjusting future programming operations accordingly, reducing the time penalty associated with verification over multiple programming cycles.
Solution Approach 2:
The patent implements dynamic programming parameters that adapt based on real-time feedback and predictive modeling. Rather than using fixed verification schedules, the system dynamically adjusts which cells require verification and what verification parameters to use, optimizing the balance between reliability assurance and time consumption based on actual programming progress and predicted outcomes.
Data Source
AI summary
Techniques are provided for predictively programming of non-volatile memory, which may reduce the number of verify operations. In one aspect, a programming circuit is configured to program memory cells to a verify low voltage and to program a set of the memory cells to target states. The set comprises memory cells having a threshold voltage between the verify low voltage and a verify high voltage. To program the set of the memory cells to the target states, the programming circuit is configured to apply two or more program pulses to memory cells in the set without verifying whether the memory cells have reached their respective target states, including: apply a first and second program enable voltages to the bit lines associated with the memory cells having different strengths.