TL;DR
This paper introduces a novel FinFET SRAM cell design (AXP) that significantly improves robustness against Bias Temperature Instability (BTI) at high supply voltages and maintains high yield across a wide range of supply voltages, including low voltages. By employing strained pMOS transistors and predischarged bitlines, the AXP cell demonstrates superior read static noise margin (SNM) degradation and better overall yield compared to conventional SRAM designs.
What Problem Does This Paper Address?
Modern integrated circuits, including Static Random-Access Memory (SRAM) cells, are increasingly susceptible to performance degradation due to various factors. One significant challenge is Bias Temperature Instability (BTI), a phenomenon that causes a gradual increase in transistor threshold voltage over time, adversely affecting circuit stability and performance. As transistors shrink and operating voltages are scaled down, the electric fields within the gate oxide can become more intense, exacerbating BTI effects. Furthermore, process variations inherent in semiconductor manufacturing can lead to significant deviations in device characteristics, impacting the yield of SRAM cells, especially at lower supply voltages where stability margins are already reduced. Conventional SRAM cell designs, typically using 6-transistor (6-T) architectures with nMOS access transistors, struggle to maintain adequate read stability and yield under these conditions. This paper aims to address these limitations by proposing a FinFET SRAM cell that is more resilient to BTI and exhibits higher yield across different operating conditions.
Methods
The authors propose a new SRAM cell architecture, termed AXP, which deviates from the conventional 6-T design (AXN) in two key aspects:
- Strained pMOS Access Transistors: The AXP cell utilizes strained pMOS transistors for its access transistors. This strain, achieved through embedded SiGe in the source/drain regions, is known to improve hole mobility and, importantly, enhance robustness against Negative Bias Temperature Instability (NBTI) in pMOS transistors. The paper leverages this property to mitigate BTI-induced degradation of the read static noise margin (SNM).
- Predischarged Bitlines: In the AXP cell, the bitlines are predischarged during the read and hold states. This is a departure from the conventional approach where bitlines are typically precharged to Vdd. This predischarge strategy, combined with pMOS access transistors, is designed to improve read stability.
The study employs mixed-mode device/circuit simulations using Silvaco's Atlas and Athena tools to evaluate the proposed AXP cell against conventional AXN cells (both unstrained and strained pMOS variants). The simulations cover a range of supply voltages (0.5V to 1V) and consider both symmetric and asymmetric BTI stress conditions. The impact of process variations on read SNM and write margin is also investigated using statistical analysis based on Gaussian distributions for channel length and silicon thickness.
Datasets
Not discussed in the paper. The study relies on device and circuit simulations calibrated with experimental data from existing literature for BTI parameters.
Key Insights and Ideas
- Strained pMOS for BTI Resilience: The use of strained pMOS transistors is crucial for suppressing read SNM degradation due to BTI, particularly at higher supply voltages. The strain helps to reduce the impact of threshold voltage shifts caused by NBTI.
- Predischarged Bitlines Enhance Read Stability: The combination of predischarged bitlines and pMOS access transistors in the AXP cell leads to improved read stability, especially when compared to conventional designs under BTI stress.
- Trade-offs in Straining: While straining improves BTI robustness, it can slightly decrease the read current in conventional AXN structures due to a smaller Id-Vg slope. However, in the proposed AXP structure, straining, coupled with the two-fin pull-up design, results in the highest read current.
- Symmetric vs. Asymmetric Stress: Symmetric stress (data flipping regularly) causes less degradation in read SNM compared to asymmetric stress (data stored for long periods) for most structures, as it leads to more uniform device degradation.
- Process Variation Impact: The study highlights that while BTI is a concern at high voltages, process variations become critical for yield at lower supply voltages. The proposed AXP cell demonstrates superior cell sigma yields, indicating better robustness against variations.
- Workfunction Tuning for Yield: The paper suggests that workfunction tuning can be employed to improve the cell sigma yields of other structures, but the proposed SAXP structure still maintains an advantage.
Key Results
- Read SNM Degradation: For a supply voltage of 1V and after three years of symmetric aging, the strained proposed structure (SAXP) shows only a 6% degradation in read SNM. In contrast, unstrained proposed (AXP), unstrained conventional (AXN), and strained conventional (SAXN) structures exhibit 14%, 12%, and 11% degradation, respectively.
- High Yield at Low Voltages: The AXP cell achieves read and write cell sigma yields greater than six for supply voltages as low as 0.5V. Other structures often fall below this threshold at such low voltages.
- Read Current: The SAXP cell (with two fins) exhibits the highest read current among all structures, with the lowest degradation due to aging.
- Write Margin: The write margin of the SAXP structure is higher than that of the AXP. The write margin degradation with supply voltage scaling is also less pronounced in the proposed AXP structure.
- Static Power: All structures, including the proposed SAXP, consume less than 10nW per cell static power, attributed to the near midgap gate workfunction.
Limitations and Future Work
- BTI Data for SiGe FinFETs: The authors note that direct BTI data for SiGe source/drain FinFETs were not available, and they relied on BTI degradation results from [17] for typical surface orientations, assuming strain does not create additional trap density generation.
- Two-Fin Pull-up for SAXP: To achieve comparable read SNM to other structures, the SAXP cell's pull-up transistors were designed with two fins. While this improves read SNM, it slightly deteriorates write stability. The authors mention that for the one-fin SAXP cell, the write margin is already high enough, so this increase in fins doesn't make it unacceptably low.
- Future Work: The paper implicitly suggests that further optimization through workfunction tuning could enhance the yields of other structures, but the SAXP remains superior. The study focuses on 22nm FinFETs, and extending these findings to smaller technology nodes could be a future direction. Investigating the impact of other process variations beyond channel length and silicon thickness could also be explored.
Summary updated