XFP光學無線電收發機的元器件及板級熱建模技術

Component and Board-Level Thermal Modeling Techniques for XFP Optical Transceivers

 

摘要Abstract

在10吉比特小型插塞式光學收發器的多源協議中，定義對于最多只包含16個功率為1.5,2.5和3.5瓦模塊的單板結構，溫度應保持在70攝氏度以下。The Multi-Source Agreement (MSA) for the 10 Gigabit Small Form Factor Pluggable (XFP) optical transceiver specifies that case temperature remain below 70 ºC for single board configurations of up to sixteen 1.5, 2.5, and 3.5 W modules.由于溫度作用涉及很多不同的設計因素，如空氣的流速，流向，散熱片的設計方式以及其它元器件的布局等等，因此這一定義是設計方面的重大挑戰。This specification can present a significant design challenge, as case temperature is a strong function of many varied design elements such as flow velocity, flow direction, heat sink design, and the placement of other components.熱分析軟件可快速詳盡地探知光學收發器內部機械裝置以及它與板內其它元件交互的熱傳遞情況，從而大大的簡化了熱設計的過程。Thermal analysis software can greatly simplify the thermal design process by providing fast, detailed insight into the mechanisms of heat transfer for the optical transceiver and its interaction with other components on a board.本文主要介紹在XFP光學收發機元器件及板級分析中所采用的熱建模技術。重點講述建立詳細緊湊的熱模型的方法。This paper covers thermal modeling techniques for component and board-level analyses of XFP optical transceivers. A focus is placed on detailed and compact form modeling methodologies.

 

介紹Introduction

為光纜網絡公司發布新一代產品的市場機會的評估在數月內，有時在幾星期內就可完成。The window of opportunity for launching new generation products is measured in months, sometimes weeks for optical networking companies.與此同時，在規范產品性能上所帶來的壓力就成為工程設計者要面對的新的更加嚴峻的挑戰。At the same time the pressure to push the limit on product performance introduces new, increasingly severe engineering challenges.

將10G比特率以太網部件和系統投放市場的潮流使那些提供熱設計的實際建模軟件對這種高級熱設計技術進行了再投資。The rush to introduce 10 Gbps Ethernet components and systems to the market has placed a premium on advanced design techniques for thermal design such as those offered by virtual prototyping software.熱分析主要集中在其中一個部件上，XFP光學收發器。但是，XFP有一個70攝氏度的溫度限制。This thermal study focuses on one of these components, the XFP optical transceiver. The XFP, however, has a case temperature limit of 70 °C.由于電路板印刷管腳更為細小，可并排放置16個XFP模塊，因此要對這種排放進行細致分析以確保不超過70攝氏度的溫度限制。With a smaller board footprint, the ability to place up to 16 of these modules in a row must be carefully analyzed to ensure that the limiting temperatures are not exceeded.

FLOTHERM熱分析軟件可用于測試環境下的XFP光學接收器的建模。FLOTHERM thermal analysis software was used to model the XFP optical transceiver within the test environment.FLOTHERM為XFP建立了兩種不同的模型表述。Two separate representations of the XFP were created in FLOTHERM.建立并測試的第一種模型是XFP的詳細表述模型，它包括了內部器件的詳細描述，諸如電路板，元器件以及激光組件等等。The first model created and tested is considered a detailed representation of the XFP, including interior details such as the board, components, and laser assemblies.第二種XFP模型是在同等溫度條件下對詳細模型的簡化描述。通過改變氣流速度，測試電路板構型和功率給出兩種模型的比較結果。The second model of the XFP is a compact representation that is designed to reproduce equivalent case temperatures as the detailed model. Results involving comparisons between the detailed and compact models are presented for varying airflow rates, test board configurations, and powers.

 

 

建模Modeling

無論是XFP的詳細模型還是簡化模型，都必須能夠對給定的氣流環境中的溫度情況進行預測。The detailed and compact models of the XFP must both serve the same purpose, to predict accurate case temperatures in a given flow environment.但是，簡化模型在一些方面要優于詳細模型。其中最重要的一個優勢是它沒有表現出任何的自身敏感性信息。這就使簡化模型可以滿足任何對XFP熱模型的要求。However, there are several advantages that the compact model has over the detailed model. The most important advantage is that the compact model does not reveal any sensitive proprietary information. This enables a compact model representation to be given to anyone who requires a thermal model of the XFP.

使用簡化模型的另外一個顯著優勢是可以減少問題解決所需的時間。任何分析軟件對復雜幾何構型的分析總比對簡單幾何構型的分析花費的時間多。用簡化模型可以在很大程度上節約時間。Another large advantage for using compact models is reducing solution time. In any analysis software, more complex physical geometry will take longer to solve than simple geometry. By using a compact model, significant amounts of time.

光學收發器是根據MSA中提供的信息建立起來的。這些信息包括部件的整體尺寸和構形，放置散熱器的空間和測試環境的結構。The optical transceiver was constructed based on the information provided in the MSA. This included the overall size and configuration of the cage, the space available for the heat sink, and the configuration of the test environment.散熱器是10 x 4的齒狀結構。XFP詳細模型的外觀圖如圖1所示。The heat sink was constructed as a 10 by 4 pin fin heat sink. The outer view of the detailed XFP is shown in Figure 1.箱體，散熱器和模塊的傳導率可從早先研究概括在MSA中的數據直接獲得。The conductivities of the cage, heat sink and module were taken directly from a previous study summarized in the MSA.內部結構的詳細情況如圖2，假設有一塊主板，板上有三個部件以及一組激光組件。The interior details, shown in Figure 2, were constructed by assuming a single main board with 3 components as well as estimates of typical laser assemblies.

 

建立簡化模型是為了去掉內部細節，簡化箱體設計。箱體連接器集成在一起，給出一個傳導系數來表示等量熱阻。箱的底部同樣簡化成具有等量熱阻的單一模塊。散熱器以一種FLOTHERM軟件提供的簡化方式來表示。XFP的簡化模型整觀圖如圖3所示。The compact model was constructed in order to remove the interior details, and simplify the cage design. The cage connector tabs were lumped together and given a conductivity that represented an equivalent thermal resistance. The cage bottom was also simplified to a single block with an equivalent resistance. The heat sink was replaced with a compact representation that is available within the FLOTHERM software. An overall view of the compact XFP is shown in Figure 3.

Figure 3: Overall view of the compact XFP.

測試模型在MSA提供的信息基礎上創建。其結構如圖4所示。The test configuration was created based on information provided by the MSA. This configuration is shown in Figure 4.測試板的尺寸為406.4mmx 279.4mm。8個部件置于板上，散出的熱量使空氣溫度上升10攝氏度。所有空氣在40攝氏度時從風道一端進入由另一端排出。尖刃處沒有空隙，空氣不會散失。The test board has a size of 406.4 mm (16 in) by 279.4 mm (11 in). Eight components were also placed on the board and given a heat dissipation that resulted in a 10 ºC temperature rise of the air. All of the air enters at 40 °C at one end of the wind tunnel, and exits out the other end. There are no openings in the bezel to allow the air to escape.

Figure 4: Wind tunnel test configuration showing eight XFP’s.

 

基本模型模擬結果Baseline Result

構建簡化模型完全依據從詳細模型的基本解決方案中所獲取的結果。The construction of the compact model is entirely dependent on the results obtained from a baseline solution involving the detailed model.詳細模型必須提供熱量由內部器件導入箱體的位置以及在每一位置上的熱傳導量。The detailed model must provide the locations that heat is conducted into the case from the interior components, as well as the amount of heat conducted at each location.

初始的詳細模型解決方案包括：將一個XFP放置在測試板上。散失功率1.5瓦。通過風道的氣流速度為1.016米/秒。溫度由箱體上的四個離散點確定。這四點分別是組合部件的頂部，兩側和尾部。The baseline detailed model solution involved placing a single XFP on the test board. The detailed model dissipated 1.5 W. The flow through the wind tunnel was 1.016 m/s (200 lfm). Case temperatures were determined at four discrete points on the cage. The cage temperatures determined were the top, the two sides and rear of the cage assembly.

內部器件和箱體外部的熱傳遞路徑應該是確定的。用像FLOTHERM這樣的分析軟件對這些位置的熱量傳導進行測量是非常容易的。The conduction paths between the heat generating interior components and the outer cage should be well defined. In analysis software such as FLOTHERM, the amount of heat being conducted into the cage at these locations can be measured quite easily.例如，在分別位于XFP板上和箱體頂部的兩個器件之間放置一個隔離墊。因而箱體和隔離墊的接觸面就是主要的熱傳導位置，通過這個面的熱傳導量很容易度量。For example, a gap pad has been placed between two of the components on the XFP board and the top side of the cage. Therefore the interface between the cage and gap pads is a primary conduction location, and the amount of heat conducting through that face is easily measured. Based on this test condition, the case temperatures ranged from 49.2 to 51.0 °C. The flow of air through the wind tunnel is shown in Figure 5. The high conductivity of the cage, along with the large surface area of the heat sink both contribute to a large percentage of the heat being convected to the air instead of conducted to the board.

Figure 5: Flow of particles past a single detailed XFP.

The most important piece of information, the heat conducted into the cage, was also determined. Using this information, a compact model can now be completed. Instead of using interior heat generating components as in the detailed model, non-physical heat sources can now be created at the same locations. This includes the gap pad/cage interface, and the laser assembly/module interface. The remaining heat that did not directly conduct into the cage is spread throughout the interior volume in order to convect to the cage.

Using this compact model, it was placed in the exact same wind tunnel as before. In this test, the cage temperatures ranged from 50.0 to 50.8 °C, which is not significantly different than the detailed model results. The largest percentage difference in the change in temperature was less than 9%. A comparison between the detailed and compact model results is shown in Table 1. The largest advantage, however, can be seen in the solution time. Where the detailed model took over 10 hours to converge, the compact model was able to converge in 29 minutes, over an order of magnitude less. The reduction in grid count was also significant. While the detailed model had over 413,000 grid cells, the compact model had only 115,000 grid cells.

Top	Bottom		Left		Right
Detail	49.2	51.0		50.3	49.8
Compact	50.0	50.8		50.6	50.3
% Change	8.50	1.89		3.40	5.18

 

Table 1: Case temperature comparison for a single XFP in 1.016 m/s airflow.

Parametric Studies

While an initial comparison of the cage temperatures between a detailed and compact representation of the XFP were very close, there is no guarantee that this would be true in different test conditions. The advantage gained from the reduction in solve time would be greatly diminished if every different test condition required a detailed model baseline result. The best situation is if the result gained from the initial baseline test could be used in different conditions. Common differences between separate test configurations would include the flow rate of air, the number of XFP’s, and the test board. Therefore a parametric study was conducted to investigate the effect of these changes.

In each test configuration, both compact and detailed XFP modules were tested. This ensures that the cage temperatures for the compact model would always have a detailed result to compare with. However, the compact models that are used are based entirely on the heat conduction data obtained from the baseline case. Ideally, a detailed model test situation would only have to be solved for once, and using that result, the compact model can be placed in a variety of different test configurations to present reliable cage temperatures.

Parametric Study 1

The first parametric study involved altering the flow rate through the wind tunnel. The baseline test condition was 1.016 m/s (200 lfm). Two additional flow rates were tested at 1.778 m/s (350 lfm) and 2.54 m/s (500 lfm). The compact model was based entirely on the data obtained from the initial baseline case.

For the 1.778 m/s case, the case temperatures for the detailed model ranged from 46.6 to 48.9 °C. The compact model resulted in case temperatures that ranged from 47.9 to 49.0 °C. A comparison is shown in Table 2. Again, the compact model solution is a conservative representation of the case temperature, but still not much different from the detailed model result.

Top	Bottom		Left		Right
Detail	46.6	48.9		48.0	47.6
Compact	47.9	49.0		48.8	48.5
% Change	19.74	1.55		9.93	11.77

 

Table 2: Case temperature comparison for a single XFP in 1.778 m/s airflow.

For the 2.54 m/s case, the case temperatures for the detailed model ranged from 46.4 to 48.6 °C. The compact model resulted in case temperatures that ranged from 46.6 to 47.8 °C. These results are outlined in Table 3. While the compact solutions are not all on the conservative side, on a percentage basis they are still within 10% of the detail model solution.

Top	Bottom		Left		Right
Detail	46.4	48.6		47.7	47.3
Compact	46.6	47.8		47.5	47.3
% Change	3.72	9.71		2.49	1.29

 

Table 3: Case temperature comparison for a single XFP in 2.54 m/s airflow.

Parametric Study 2

The second parametric study focused on the number of XFP’s that are placed on the test board. All of the previous tests were configured with only one module. A series of tests involving eight transceivers was computed using two different airflow rates, 1.016 m/s and 2.54 m/s (200 and 500 lfm).

The eight transceivers were placed in a single row, with a pitch of 46 mm. This pitch distance was taken directly from the MSA. While results for all 8 transceivers are available, only the case temperatures of the sixth transceiver will be presented here for brevity’s sake. The flow of air particles is shown in Figure 6.

Figure 6: Flow of particles past eight detailed XFP’s.

At a flow rate of 1.016 m/s, the sixth detailed transceiver had case temperatures ranging from 60.5 to 62.0 °C. For the compact models, the sixth transceiver had case temperatures ranging from 61.2 to 61.9 °C. These results are summarized in Table 4. As can be seen, even though the compact model is still based on the baseline result, the results for an XFP placed in the interior of a row of eight devices still present very accurate results. The percentage difference in case temperatures are all less than 4%.

Top	Bottom		Left		Right
Detail	60.5	62.0		61.3	61.0
Compact	61.2	61.9		61.7	61.5
% Change	3.56	0.13		1.98	2.45

 

Table 4: Case temperature comparison for the sixth XFP in 1.016 m/s airflow.

The amount of time saved is also even more significant for this case. The detailed model of the eight XFP’s involved a grid cell count of 1,645,020 and a solve time of several days. The compact model, in contrast, had a grid cell count of 310,200 and a solve time of approximately 6 hours.

At a flow rate of 2.54 m/s, the sixth detailed transceiver had case temperatures ranging from 53.6 to 55.8 °C. For the compact models, the sixth transceiver had case temperatures ranging from 54.6 to 56.0 °C. As shown in Table 5, the comparison between compact and detailed models shows that the compact models give very accurate results.

Top	Bottom		Left		Right
Detail	53.6	55.8		54.8	54.5
Compact	54.6	56.0		55.5	55.3
% Change	7.05	0.93		4.87	5.51

 

Table 5: Case temperature comparison for the sixth XFP in 2.54 m/s airflow.

Parametric Study 3

In all of the test configurations discussed so far, the composition of the wind tunnel test board and its eight components has remained constant. In this study, the effect of changing this board was investigated with regards to the amount of heat being dissipated, and the composition of the board itself.

The power being dissipated by the other generic components on the test board had been adjusted in order to give an air temperature rise of 10 °C. Each of the eight components dissipated the same amount of heat. As a result, the test board itself would have been heated substantially, thereby reducing the amount of heat being conducted into the board from the transceivers. The amount of power being dissipated by the components was reduced to zero in order to determine the effect on the case temperatures.

The tests were performed with one XFP in 1.016 m/s airflow. For the detailed model, the case temperatures ranged from 45.4 to 45.8 °C. The effect of the component heat on the case temperature can be seen quite clearly. The case temperatures here are much cooler than the initial baseline case due to the reduced temperature of the board. For the compact model, the case temperatures ranged from 45.3 to 45.7 °C. As the results show in Table 6, the effect of having a cooler test board still resulted in the compact model giving very accurate case temperatures.

Top	Bottom		Left		Right
Detail	45.4	45.8		45.6	45.5
Compact	45.7	45.3		45.5	45.5
% Change	5.77	8.52		1.02	0.91

 

Table 6: Case temperature comparison with zero test board power dissipation.

The second part of this study was to alter the composition of the board. For all of the test cases, the board configuration was set to 10% copper and 90% FR4. For this test, the percentage was changed to 20% copper and 80% FR4. In this case, the detailed model case temperatures ranged from 51.3 and 54.0 °C. The compact model case temperatures ranged from 52.4 to 54.0 °C. Again as shown in Table 7, even though the compact model was created based on a different test board configuration, it still presents very accurate results.

Top	Bottom		Left		Right
Detail	51.3	54.0		53.0	52.4
Compact	52.4	54.0		53.6	53.2
% Change	9.20	0.42		4.20	6.30

 

Table 7: Case temperature comparison for different test board composition. Parametric Study 4

Up to this point the focus has been on the 1.5 W version of the XFP transceiver. The last parametric study is to determine if a 3.5 W version of the XFP transceiver can be similarly represented in a compact model with equally accurate results.

To create a compact representation of a 3.5 W transceiver, a detailed 3.5 W transceiver was first tested in a 2.54 m/s environment. The process to create the compact model was exactly the same as in the initial baseline case involving the 1.5 W transceiver.

For the 2.54 m/s flow over one XFP, the detailed model had case temperatures ranging from 51.4 to 54.7 °C. The compact model had case temperatures ranging from 52.1 to 53.3 °C. As can be seen in Table 8, the results still agree very well.

Top	Bottom		Left		Right
Detail	51.4	54.7		53.3	52.8
Compact	52.1	53.3		53.1	52.8
% Change	5.72	9.99		1.50	0.20

 

Table 8: Case temperature comparison for a single 3.5 W XFP.

The 3.5 W XFP was also tested with eight XFP’s in 2.54 m/s airflow. Again, only the sixth transceiver will be compared. For the detailed model, the case temperatures ranged from 61.1 to 64.5 °C. For the compact model, the case temperatures ranged from 64.3 to 65.8 °C. These results are summarized in Table 9.

Top	Bottom		Left		Right
Detail	64.5	62.9		62.7	61.1
Compact	65.8	65.4		65.2	64.3
% Change	14.71	5.09		10.71	11.23

 

Table 9: Case temperature comparison for the sixth 3.5 W XFP.

Conclusion

The viability of using compact models of XFP optical transceivers has been analyzed. The use of compact models hides all of the important proprietary information while dramatically reducing solve times. While the XFP optical transceiver is well suited for compact modeling due to its large heat sink and high conductivity cage design, the compact modeling procedures outlined here are applicable to other complex components as well. The most important element of creating such a compact model is to fully understand the conduction paths into the cage and board, and to represent those paths with equivalent thermal resistances