SSD Thermal Throttling Prediction using Improved Fast Prediction Model
SSD Thermal Throttling Prediction using Improved Fast Prediction Model
复制标题
使用改进的快速预测模型进行 SSD 热节流预测
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
H. Takiar
中科院分区：
文献类型：
--
作者：
Hedan Zhang;Ernold Thompson;Ning Ye;Dror Nissim;Steve Chi;H. Takiar
固态硬盘（SSD）产品的热管理非常具有挑战性，特别是M.2，因为它的小尺寸和高能量密度。更好地了解SSD的热预测有助于优化性能，避免热失控问题。使用传统的计算流体力学（CFD）模拟方法来预测用户自定义功率输入下的温度是非常耗时和繁琐的。例如，要预测持续三个周期的典型热节流曲线，CFD模拟可能需要几天到一周的时间才能完成。CFD方法也不利于集成到SSD性能模拟器中。因此，一个快速、鲁棒且易于使用的SSD热预测模型对于应对这些挑战变得越来越重要。本文采用了一种较早的基于部分分式电路修正热阻抗的快速预测模型方法。对于最近的M.2固态硬盘产品模型，当一次仅为一个封装供电时，可以相应地为NAND提取常数对。r平方值在曲线拟合中表现良好。建立了一个详细的笔记本环境下的固态硬盘CFD模型，并通过实验进行了验证。从CFD模拟结果中提取常数对。使用这些参数的温度分布与CFD结果完全一致，r平方值接近1。利用线性叠加原理，将预测的温度分布与任意循环功率模式下的CFD结果进行对比验证。结果一致性在预期范围内。然后将该方法应用于预测M.2 SSD的1000秒热节流。研究了不同的节流温度。结果表明，提高温度阈值的上限和下限都有助于提高SSD的热性能。
Thermal management is very challenging for Solid State Drive (SSD) product especially for M.2 due to its small form factor and high energy density. A better understanding of thermal prediction of SSD can help optimize performance and avoid thermal runaway issue. Using traditional Computational Fluid Dynamics (CFD) simulation approach to predict temperature under user defined power input is very time consuming and tedious. For example, to predict a typical thermal throttling profile lasting three cycles, for CFD simulation it could take from days to a week to finish. CFD approach is not favorable to be incorporated into SSD performance simulator either. Therefore, a fast, robust, and easy to use thermal prediction model for SSD is becoming more and more critical to meet these challenges. This paper has used an earlier fast prediction model methodology with modified thermal impedance using partial fraction circuit. For a recent M.2 SSD product model, constant pairs can be extracted accordingly for NAND when powering one package only at a time. R-square values have been shown to be good in curve fitting. A detailed SSD in a laptop environment CFD model has been established and validated through experiment. Constant pairs were extracted from the CFD simulation results. Temperature profile using these parameters was perfectly aligned with CFD results with R-square values close to 1. Using linear superposition principle, the predicted temperature profile was compared with CFD results under arbitrary cycling power mode for verification. Result consistency is within expected range. This approach was then applied to predict M.2 SSD thermal throttling for 1000 seconds. Different throttling temperatures have been studied. Results have shown that increasing temperature threshold for both upper and lower limit may help improve SSD thermal performance.