By Ralph Remsburg, Chief Technical Officer, Electronics Packaging Associates, San Diego, CA
Air bullets are little pulses of cooling air that are conducted to specific components without disrupting a fan’s airflow. Researchers at Influent Corporation and Electronics Packaging Associates confirmed that concurrent computational fluid dynamics (CFD) tools can predict the nuances of air flow and temperature behavior in an air bullet design.
Pulsed air bullet packets entrain a growing volume of air as they travel.
The bullets are emitted by a special actuator through ports designed to constrict jets of air in a controlled manner. An electrically powered piston oscillates within the actuator’s internal chamber, pushing air outward and inward alternately through a port in the housing on each side of the piston. The velocity imparted by the piston’s push must be sufficient to cause air pulses to escape the actuator ports and travel outward. No motion is wasted since a usable air bullet is dispensed through one port or the other by each new stroke.
The pulsating jet of air is amplified by “entrainment” as it moves away from the port. Entrainment is the result of friction between high-velocity air pulses as they move and the surrounding stationary air. This friction creates rotating vortices that follow each pulse, causing stationary surrounding air to be entrained into the pulsating air jet. In effect the vortices transform the pulse’s energy into the increasing flow rate symbolized by the widening cone of air. The overall mass of moving air and the cooling potential actually increase with distance. An air jet system can deliver cooling efficiency many times greater than an equivalent fan array when size and power consumption are considered.
CFD analysis can predict the thermal and flow performance an air bullet design will deliver. Mentor Graphics’ FloEFD Concurrent CFD analysis software is suited for performing these simulations. The Influent and Electronics Packaging Associates team chose the software to validate their concept for an air-bullet cooler.
To verify that a commercial CFD package could properly predict entrainment volume and flow divergence, the team compared FloEFD model outputs to the results of bench-top experiments. The jet port and the actuator enclosure were modeled to allow a variety of port configurations to be considered. The CFD tool allowed them to set boundary condition options such as velocity and pressure. Using any of these parameters along with the pulse frequency and port configuration (diameter and edge profile) provided sufficient model input to achieve a usable flow field output. The internal dynamics and volume of the actuator, and the stroke length did not have to be considered.
This shows the CFD model’s flow trajectories in the cross-stream plane at eight time points in a sinusoidal pulse cycle. Frame B shows the vortex pair beginning to form at the left edge. Frame C depicts the activity at the maximum port velocity, 80 m/s, while Frame E captures the view at the midpoint of the stroke when port velocity is 0 m/s. In Frame H the lower vortex has dissipated and the velocity is again returning to 0 m/s. The upper and lower halves of the vortex pair show evolving differences as their energy is expended on entrainment and decreased by friction with the ambient air. The trajectories calculated by the FloEFD CFD analysis (Frames A-H) closely matched bench-top measurement results. The simulation results tracked with the bench-top experiment. The engineering team derived a relationship between the two by defining the time-averaged transition from pulse flow to random flow as a function of the centerline velocity at any distance downstream of the port. This test and others proved that the CFD tool could reliably guide design work on the air bullet system.
In the evaluation project the basic mesh contained about 747,000 cells with cell size scaled according to the resolution required in each region of the model. In all of the models, the pulse frequency was set at 50 Hz and the time step for the calculation was 0.000025 s (800 time steps per air pulse). One of the earliest tests estimated flow divergence as a function of distance from the port opening.
The next step was simulation of an air bullet implementation that would work in the real world. In theory, an actuator of sufficient capacity can drive cooling air pulses through manifolds and out multiple ports. The design team was to confirm this theory and create a dedicated cooling source for a bank of 16 dual in-line memory modules (DIMMs) situated at the rear of a 1U rack server.
Simulating these elements was not a matter of simply entering the materials and geometries as usual. Because the actuator operates at 50 Hz, the CFD analysis requires a very fine time step: 25 microseconds. But the thermal reactions of hardware components follow a different time scale. A basic component with a heat sink may take 10 minutes to stabilize. This equates to 24,000,000 iterations, potentially requiring upward of one minute each to compute.
Knowing this, the design team chose to reduce the specific heat property of the solid components in their models to 1/1000 of the actual value. This reduced the thermal reaction times, and dividing by 1000 did not impact the final temperature results.
Results from CFD simulation “B.” The maximum DRAM case temperature with the jets off is 101.1° C, which exceeds the devices’ rating. With the jets active, the maximum case temperature drops to 81.5° C, a reduction of 19.6° C. The average temperature drop across all DRAMs is 11.2° C.
In simulation “A,” the near end of the circuit board mount contains 12 jets that shoot pulsed air between the fins of a heat sink mounted atop a 15 W IC package. The jet pulses were delivered at a 50 Hz frequency, just like the earlier entrainment experiments, and at a peak velocity of 30 m/s. The simulation was performed with and without the virtual jets. Ultimately, the CFD model showed a maximum temperature rise of 55° C with pulsed jets enabled and 195° C with convection cooling alone. The FloEFD tool’s result compared favorably with a bench-top test that produced a temperature rise of 59° C. The average temperature drop across all DRAMs is 11.2° C.
The airflow and temperature analysis models used by the FloEFD Concurrent CFD tool.
Simulation “B” depicts the proposed air bullet distribution hardware with manifolds and “slot runners” that direct air to the server’s DIMM array. Each manifold is connected to opposite sides of the jet actuator, resulting in the air pulses being 180° out of phase for alternate DIMM cards. The peak jet velocity at each of these ports averages 70 m/s.
Electronics Packaging Associates
Filed Under: Simulation, Software