RRAM Crossbar Circuits With Feedback RTN Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Crossbar circuits with resistive random-access memory (RRAM) experience random telegraph noise (RTN) due to stochastic switching of resistance levels, leading to errors and instability, which limits inference accuracy in applications like in-memory computing and neural networks.
Innovation Solution
A method involving programming RRAM devices to a target conductance value, determining RTN values, and applying noise-reduction voltages until the RTN values fall within an acceptable range, using a controller to generate and apply these voltages iteratively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RRAM devices are programmed to target conductance values using standard programming processes, then the crossbar circuit can achieve the desired conductance levels for computation, but random telegraph noise causes the actual conductance to deviate from target values, reducing inference accuracy
Solution Approach 1:
The patent implements a feedback mechanism where the controller measures the actual conductance value of each RRAM device after programming, compares it to the target conductance value, and determines whether to apply additional noise-reduction voltages based on whether the deviation exceeds a threshold. This closed-loop feedback process continues until the conductance is within the acceptable range, thereby resolving the contradiction between achieving target conductance and maintaining inference accuracy despite RTN effects.
Solution Approach 2:
The patent changes the voltage parameter dynamically during the programming process. Initially, higher programming voltages are applied to set the conductance to target values. Then, based on feedback, lower noise-reduction voltages are applied iteratively to reduce RTN-induced deviations. This parameter change strategy allows the system to first achieve the desired conductance level and then refine it to improve inference accuracy.
2Reliability
If noise-reduction voltages are applied to suppress RTN in RRAM devices, then the stability and inference accuracy of crossbar circuits improve, but additional programming steps and voltage applications increase the complexity of the programming process
Solution Approach 1:
The system performs self-service by automatically determining whether noise-reduction voltages are needed based on feedback measurements. The controller autonomously decides whether to apply additional voltages by comparing measured conductance values to target values, eliminating the need for external manual intervention or complex external control circuitry. This self-service approach improves circuit stability while keeping the programming process complexity manageable.
Solution Approach 2:
The patent applies partial action by conditionally applying noise-reduction voltages only when necessary. Not all RRAM devices require noise-reduction voltages; the controller determines on a per-device basis whether the conductance deviation exceeds the threshold. This selective application of noise-reduction voltages improves circuit stability where needed while avoiding unnecessary operations, thereby balancing reliability improvement with programming process complexity.
3Measurement precision
If multiple iterative programming processes are performed to correct RTN deviations, then the conductance precision and inference accuracy improve, but the programming time and energy consumption increase
Solution Approach 1:
The patent applies partial action by performing iterative programming processes selectively rather than universally. The controller measures conductance after the initial programming process and only initiates additional noise-reduction programming steps if the deviation from target conductance exceeds a predetermined threshold. This approach improves conductance accuracy for devices that need it while avoiding unnecessary iterative processes for devices that are already within the acceptable range, thereby reducing overall programming time.
Solution Approach 2:
The feedback mechanism enables the system to determine the minimum necessary number of iterative programming processes for each device. By measuring conductance after each programming step and comparing it to the target, the system can stop the iterative process as soon as the conductance is within the acceptable range, avoiding unnecessary additional programming cycles. This feedback-driven approach improves conductance accuracy while minimizing programming time and energy consumption.
Data Source
AI summary
The present disclosure provides mechanisms for reducing and suppressing random telegraph noise (RTN) for a crossbar circuit. A processing device may perform a programming process to program the conductance of a resistive random-access memory (RRAM) device in the crossbar circuit to a target conductance value. The processing device may then determine whether a random telegraph noise (RTN) value associated with the RRAM device is within a predetermined range of acceptable RTN values. If the RTN value associated with the RRAM device is not within a predetermined range of acceptable RTN values, one or more noise-reduction voltages may be applied to the RRAM device until the RTN value associated with the RRAM device is within the predetermined range of acceptable RTN values.