Reading time ( words)
By Chris Heard, Amphenol
This case study examines a leading-edge chassis design from the thermal management perspective. Through increasingly detailed simulation environments, the system designer was able to optimize airflow, verify the design's feasibility, and circumvent possible thermal problems as the design progressed.
A network equipment supplier needed a backplane for a new communications product. Thermal management was going to be a major challenge. The 14U chassis was required to dissipate about 1700 W to deliver the rich array of features provided by the new product. In the early stages of the project, the network equipment supplier had little more than the number of cards and approximate power dissipation for each card. Unable to estimate the junction temperature of the individual components, the focus was on delivering sufficient volume of air to each card while limiting the rise in temperature between the air inlet and outlet.
The goal was to provide each card with airflow greater than 400 ft./min. while limiting the rise in air temperature at the outlet to 15°C, based on an ambient temperature of 50°C. Simple guidelines helped determine the objectives for the early-stage concept design. Network equipment building standards (NEBS) generally require airspeed of at least 300 linear ft./min. (lfm), as heatsinks tend to lose effectiveness with air tranversing below that speed. In this case study, the temperature rise limit was set at 810°C to provide leeway for later stage design changes.
The first step was to analyze airflow through the chassis with a Web-based hydraulic resistance simulation tool.* This returned results based on a 2D representation of the chassis, making calculations instantaneous. It provided an overall estimate of the air moving through the chassis and exit air temperature, but did not estimate the details of the airflow seen by individual cards.
Various chassis sizes and fan arrangements were analyzed using the hydraulic software. The client had originally requested that the height of the chassis be limited to 13 rack units (U) or 24.5" in height; smaller was better. With 18 cards in a 13U chassis, card pitch is between 0.8 and 0.9" unusually restrictive from an airflow standpoint. At 13U, airflow through the chassis was too low. With the size increased to 14U, 8, 9, 10, 11 and 12" depths were simulated. The 11" depth was the smallest at which airflow goals could be achieved.
In fan simulation, the goal was to find a fan size capable of optimizing the total airflow through the enclosure. For this application, a 92 × 38 mm fan** showed good performance, a reasonable cost, good reliability, and availability in the timeframe required by the customer. However, the hydraulic software has a key limitation. It calculated the total airflow through the chassis but not able to determine the distribution of airflow. It was not able to predict if each card would receive the specified amount of air using the fan specified.
Using CFD for Thermal SimulationComputational fluid dynamics (CFD) software*** was brought in to model the entire geometry of the chassis. It allowed the designer to specify where the power would be dissipated. The simulation tracks the flow of air and transfer of heat throughout the chassis. The software's accuracy had been verified with prior projects, and customer knowledge transfer was ensured with compatible systems.
Using a form-based menu system to define PCB geometries saved time in creating the CFD model, avoiding the process of modeling the full geometry of the PCBs. Instead, PCB parts were created by defining the length and width and each layer of the board. For each layer, percentage of copper and dielectric were specified. Each board was then filled with 32 25.4-mm2 devices designed to simulate the type of field programmable gate arrays (FPGAs) typically used in data communications equipment. For each, these FPGA models were arranged in the same evenly spaced layout around the board. Besides saving modeling time, this technique also reduced the amount of time required to perform the simulation. Using the same component layout on each board further reduced the overall number of cells in the model, cutting out more simulation time.
Properties of the fan were defined in a similar program dimensions were entered as well as the curve that defined its performance. The design was then ready for simulation that predicted the airflow and heat transfer throughout the chassis.
Figure 2. Slot 7 air speed.
The simulation predicted airflow for each slot. The airflow for Slot 7, shown in Figure 2, is important because this slot dissipates 150 W. Overall, the simulation shows the airspeed within the chassis ranges from 500 to 800 lfm, but the arrows moving in a circular direction at the lower right-hand side of the plot indicate a recirculation zone in this area. The card guides in this area of the chassis are clearly causing the problem. This problem could be solved by modifying the geometry of the card guides. Recirculation was also seen at the top of the card just below the hubs of the fans, which appear as blue rectangles in the figure. This might have caused a problem if a high-powered component was placed near the area where recirculation was occurring. The problem could be addressed by moving the fans further from the board. At this early stage in the design process, these relatively minor issues were not yet addressed.
Figure 3. Slot 7 air temperature.
Figure 3 shows the air temperature at slot 7. The exit air temperature is 54.5°C, which is acceptable. The hottest point is 61.2°C.
Figure 4. Chart shows operating point of the fan overlaid on the complete fan curve.
The left side of Figure 4 shows the operating pressure of the chassis design, 0.29" of water, overlaid onto the fan curve. As shown in the chart, the fan is operating just above the knee of the fan curve. Ideally, the operating point would be below the knee of the fan curve because a change in pressure would result in a smaller change in airflow. In a cutting-edge design such as this, however, it is increasingly common to operate above the knee.
ConclusionThis project demonstrates how thermal simulation can be used in the early stages of a project to minimize expensive late-stage changes and reduce time to market. While it was not possible to resolve every issue with the limited information available at this stage of the process, thermal simulation gave the network equipment manufacturer confidence to move forward knowing that the basic thermal design was sound any thermal issues that arose later should be relatively easy to resolve. Once more design information became available, the manufacturer simulated the design again, made minor changes, and successfully brought the product to market without any thermal issues.
AcknowledgementsCFD simulations were carried out with the FloTHERM CFD software from Mentor Graphics Corporation (Mechanical Analysis Division), 4 Mount Royal Ave., Suite 450, Marlborough, MA 01752; (508) 357-2012; firstname.lastname@example.org; www.mentor.com/mechanical.
* An internally developed tool called the Chassis Thermal Analyzer.** Sanyo Denki Model 9G0948H102.*** FloTHERM computational fluid dynamics (CFD) software from the Mentor Graphics Corporation, Mechanical Analysis Division.
Chris Heard, TCS engineer, Amphenol, may be contacted at Amphenol Backplane Systems, 18 Celina Avenue, Nashua, NH 03063; (603) 883-5100; www.amphenol-abs.com.