A Read Margin Enhancement Circuit With Dynamic Bias Optimization for MRAM

Renhe Chen, Albert Lee, Zirui Wang, Di Wu, Xufeng KouIEEE Transactions on Circuits & Systems II Express Briefs 2024 · DOI

TL;DR

This paper introduces a novel circuit called the Dynamic Bias Optimizer (DBO) that dynamically adjusts the read voltage in MRAMs to maximize the sensing margin. This approach significantly improves data reliability by compensating for variations in manufacturing processes and operating temperatures, leading to a dramatic reduction in bit error rates.

What Problem Does This Paper Address?

Magnetic Random Access Memory (MRAM) is a promising non-volatile memory technology due to its speed, low power consumption, and high endurance. However, a key challenge in MRAM read operations is the relatively small difference in resistance between the two magnetic states (parallel and anti-parallel), known as the Tunneling Magnetoresistance (TMR) ratio. This small difference, coupled with variations arising from manufacturing processes (Process, Voltage, Temperature - PVT) and thermal fluctuations, leads to narrow "sensing margins." These narrow margins make MRAMs vulnerable to read errors caused by noise and circuit imperfections. While some solutions exist, they often add complexity, area, and power overhead. This paper aims to address this by providing a dynamic, on-chip solution to continuously optimize the read bias voltage for maximum sensing margin, thereby enhancing MRAM reliability without significant penalties.

Methods

The core innovation is the Dynamic Bias Optimizer (DBO) circuit. The DBO operates on a feedback principle to continuously track the optimal read bias voltage ($V_{OPT}$) that maximizes the sensing margin ($I_M$).

The DBO consists of three main blocks:

  1. Margin Extraction Circuit: This block measures the sensing margin ($I_M$) and converts it into a voltage margin ($V_M$). It uses active clamp circuits, current mirrors, a trans-impedance amplifier (TIA), and a source follower to accurately replicate cell currents and derive the margin.
  2. Control Logic Block: This block periodically samples the voltage margin ($V_M$) and compares it to a stored value ($V_S$). Based on the comparison, it generates control signals (like FLIP, UP_C, UP_F, DN) to guide the charge pump.
  3. Charge Pump Module: This module dynamically adjusts the reference read voltage ($V_{REF}$) based on the control signals from the logic block. It employs a coarse-fine tuning mechanism to quickly converge to the optimal voltage and then maintain it with fine adjustments.

The paper also theoretically derives the optimal bias voltage ($V_{OPT}$) as a function of the TMR ratio and a characteristic voltage ($V_h$) of the MTJ device, showing that $V_{OPT}$ is not constant and varies with device parameters and temperature.

Datasets

Not discussed in the paper. The evaluation is based on simulations using a 28-nm CMOS process design kit (PDK) and a physical MTJ device model.

Key Insights and Ideas

  • $V_{OPT}$ is Dynamic: The optimal read bias voltage ($V_{OPT}$) is not a fixed value but rather depends on the TMR ratio and the MTJ's characteristic voltage ($V_h$), both of which are susceptible to PVT variations. This fundamental insight drives the need for a dynamic tracking solution.
  • TMR vs. Bias Voltage Trade-off: While increasing read bias voltage can sometimes improve signal strength, the TMR ratio itself often decreases at higher voltages. This creates a non-monotonic relationship between bias voltage and sensing margin, necessitating an optimization approach rather than simply maximizing voltage.
  • Feedback Loop for Optimization: The DBO's feedback loop is crucial. By continuously comparing the current sensing margin to a reference and adjusting the bias voltage accordingly, it can adapt to real-time changes in the MTJ characteristics.
  • Coarse-Fine Tuning for Speed: The charge pump's coarse-fine tuning strategy allows for rapid initial convergence to $V_{OPT}$ and then precise, low-ripple adjustments in steady-state, balancing speed and accuracy.
  • Block-Level Optimization: Embedding DBOs at a block level within the MRAM array allows for localized optimization, addressing variations that can differ even within the same chip.

Key Results

  • High Tracking Accuracy: The DBO circuit demonstrated a tracking accuracy of above 98% for $V_{OPT}$ even when the optimal sensing voltage varied by up to 50%.
  • Significant Bit Error Rate Reduction: The DBO integrated into a 1Mb MRAM macro resulted in a 1 to 2 orders of magnitude reduction in bit error rate (BER) compared to a configuration without the DBO, across various PVT variations and temperatures.
  • Performance under Extreme Conditions: The DBO maintained high performance across a wide temperature range (-40 °C to 125 °C) and showed improved sensing margins (up to 20%) under rapid temperature drifts.
  • Power and Area Overhead: The DBO implementation incurred a 5% power overhead and a 6% extra layout area penalty for the 1Mb MRAM macro.

Limitations and Future Work

Limitations:

  • Simulation-Based: All results are based on simulations; physical silicon validation is not presented.
  • Power and Area Overhead: While modest, the 5% power and 6% area overhead might be a consideration for extremely constrained applications.
  • Comparator Sensitivity: The paper notes that at extreme temperatures, the comparator's reduced sensitivity might lead to slight deviations in $V_{REF}$ from the ideal $V_{OPT}$.

Future Work:

  • Advanced Searching Algorithms: The authors suggest applying more advanced searching algorithms to the DBO's feedback mechanism to further improve $V_{OPT}$ tracking efficiency, especially for high-density MRAM applications.
  • Physical Silicon Validation: Demonstrating the DBO's effectiveness on actual hardware would be the next logical step.
  • Integration with Other Memory Technologies: Exploring the applicability of this dynamic optimization technique to other memory technologies facing similar margin challenges could be beneficial.

Summary updated

Abstract

This brief introduces a read bias circuit to improve readout yield of magnetic random access memories (MRAMs). A dynamic bias optimization (DBO) circuit is proposed to enable the real-time tracking of the optimal read voltage across MRAM process variations and operating temperature fluctuation within an MRAM array. It optimizes read performance by adjusting the read bias voltage dynamically for maximum sensing margin. Simulation results on a 28-nm 1Mb MRAM macro show that the tracking accuracy of the proposed DBO circuit remains above 90% even when the optimal sensing voltage varies up to 50%. Such dynamic tracking strategy further results in up to two orders of magnitude reduction in the bit error rate with respect to different variations, highlighting its effectiveness in enhancing MRAM performance and reliability.
BibTeX
@article{chen2024read,
  author = {Renhe Chen and Albert Lee and Zirui Wang and Di Wu and Xufeng Kou},
  title = {{A Read Margin Enhancement Circuit With Dynamic Bias Optimization for MRAM}},
  journal = {IEEE Transactions on Circuits \& Systems II Express Briefs},
  year = {2024},
  doi = {10.1109/tcsii.2024.3367988},
}
Figure