An Investigation on Radiated Emissions from Heatsinks

Transcription

1 An nvestigation on Radiated Emissions from Heatsinks S. K. Das, T. Roy Sun Microsystems, nc. Menlo Park, California Abstract: With the increase in processor speed the power dissipation density of high speed asics are increasing. The most common method of dissipating the power produced in an asic is by using a heatsink. However, due to its close proximity to the high speed digital processor a heatsink also acts as an antenna. The radiating efficiency of the heatsink depends on its shape and clock frequency of the exciting asic. Quite often the energy radiated by a heatsink gets coupled to internal cables or nearby slots inside a computer enclosure. The cables and slots then emits the energy outside the box. The objective of this study was to investigate the radiated emissions from heatsinks. n this article, radiated emissions from a circular heatsink was investigated. The heatsink has been compared to a monopole. The radiated emission has been measured in a semi-anechoic chamber and numerically simulated. NTRODUCTON The speed of a processor is directly related to the integration density of the asic. With the development of sub-micron technology the integration density is getting pushed to new heights. Along with the package density, the power dissipation per square area is also skyrocketing. Since each of these semiconductor processors has a limited thermal range of operation, therefore cooling these processors has become a major challenge. The two major factors which help in cooling an asic are surface area of the processor and the rate of flow of the cooling fluid. Since the actual dye is very small in size, the effective surface area of the processor is increased by placing a heatsink in contact with the processor. The heatsinks are extremely good thermal conductor and with the fins on the heatsink the effective cooling surface area is increased. The most commonly used and cost effective cooling fluid is of course air. The airflow is provided by fans, whose speed determines the volume of air passing over the hot surface of the heatsink. However due to acoustic reasons these fans have limits on their rotational speed which subsequently limits the amount of air flow to a maximum permissible volume. With the fans running at their permissible speed the thermal designer quite often gets compelled to design larger heatsinks in order to meet the necessary cooling requirements. This however puts an EM engineer to quite a disadvantage, since the heatsink is usually placed as close as possible to the processor, which carries the maximum amount of high frequency energy in the entire system. The heatsinks are made of metal and usually consists of fin type structures. Depending upon the processor harmonics the heatsinlc dimension can become electrically long enough to start radiating. Due to its close proximity the energy Tom a processor gets coupled to the heatsink sitting on top of it. The radiated emissions from the heatsinks could couple to the internal cables or nearby slots. Quite often the processor noise coupled through these paths i.e. internal cables, vent panels, etc., could leak outside the enclosure. The radiated emission from heatsinks is being studied by Li et al [ 11 where the authors used FDTD method to study resonances and different modes supported by the heatsink structure. n this study the radiated emission pattern for a circular heatsink has been investigated. The radiation pattern depends on the geometry of the heatsink and excitation. The heatsinks have been modeled by using finite element EM) in frequency domain. The radiation pattern of an actual circular heatsink is also measured in a semi-anechoic chamber. The circular heatsink is compared to a monopole. Depending on the harmonic frequencies, the heatsink dimension will change from electrically small to quite significant fraction of the wavelength. This not only changes the radiating efficiency of the heatsink but also changes the pattern. Moreover the secondary coupling elements are usually located in the near-field of the heatsink. The objective of this study is to characterize a circular heatsink and to investigate its radiating performance. BACKGROUND A typical board is composed of a printed circuit board (PCB), asics, passive components like decoupling capacitors, ferrite beads, filters, terminating resisters, etc. Quite often there are sockets for plug-in daughter cards. The PCBs usually contain a stack of alternating conducting layers and insulating dielectric layers. The signals, power and ground planes are routed through these /9X/$ JEEE 784

2 conducting layers. Most of the components are placed on the top or bottom of the PCB. These components are either surface mounted on the PCB or mounted via through holes. Heatsinks are usually placed directly on top of the asics. The basic principle of cooling through heatsink is to place the heatsink as close as possible to the dye of the asic and provide a path for good thermal conduction between the dye and the heatsink. One way of achieving this, is by putting a thermally conductive epoxy on top of the dye and then placing a metal lid over the epoxy layer. f the heatsink is not very heavy then its quite common to attach the heatsink to this metal plate by means of thermally conductive glue. Since glue is not strong enough to stand rigorous shock and vibration tests, bigger heatsinks are either bolted on to the metal lid or screwed in through plastic fixtures sitting on top of the asic. Even though heatsinks have evolved to provide only thermal solution, it produces unwanted EM. The authors attempted to investigate the contribution of these thermal paths in creating an electrical coupling path between the heatsink and the asic or rest of the PCB. The authors studied several types of heatsinks and their various mounting strategies. For a typical scenario, where a heatsink is placed on an asic on a PCB, there are lot of EMl sources around the heatsink and energy can get coupled by various methods. Along with the main asic (which has the heatsink on top) adjoining active components e.g. oscillators could capacitively couple to the heatsink [2]. The traces routed on the top layer (adjacent to the heatsink) can also drive the heatsink. However, there is a good chance that the most probable driving source will be the main asic itself. n the main asic even though the silicon dye is not a conductor but there are lot of metal traces and vias inside the dye which carry high speed signals. Moreover the epoxy layer on top of the heatsink is electrically conductive. The top metal lid is usually tied with the asic ground which is again co~ected to the main ground in the PCB through ground pins. There is of course certain amount of package inductance existing between the two grounds which could put the two grounds at two different potentials. The asic ground is further linked with the top metal cover by means of vias and epoxy which pushes the lid to a slightly different potential than the asic ground. The heatsink is placed on top of the metal lid with either thermally adhesive glue or bolted lugs. n this investigation, the primary source of excitation is assumed to be the main asic itself. The possible sources are assumed to be the vias and traces which can be ~ t f Heatsink 1 EPOXY -7-l lid Figure 1. The General placement of a heatsink and a chip on a PCB is shown here. Some of the details like /O connectors am ignored. The epoxy layer between the dye and metal lid is shown by the hatched area. represented by electric and magnetic dipoles : 11. Since the ground inside the asic is electrically tied to the top metal lid, it is assumed that the heatsink and the top metal lid is basically driven with respect to the PCB ground. The simplified model of the asic and the heatsink are shown in figure 1. Different types of heatsinks were studied initially but finally circular heatsinks were chosen for more in-depth study. Structurally, this particular heatsink has similarity with a monopole. n this paper, the performance of this type of heatsink over the frequency range and its radiation pattern were investigated. The authors have tried to compare the heatsink s radiation performance with a monopole over a ftite ground plane. A second model of the heatsink over an asic could be created by taking the top metal lid into account. The metal lid acts a small patch in between the heatsink and the finite ground plane. The performance of the actual heatsink was also compared with a monopole on a patch over a finite ground plane. mermentalsm The simplified model shown in figure 1 is constructed for both experimental purpose and theoretical simulation. n order to represent a ground plane a rectangular piece of metal (54cm x 42 cm) was taken. n order to create the excitation source, an N-type connector was punched through the center of the rectangular metal plate and a subminiature type-a connector was hooked up to the N-type connector through an adapter. The subminiature connector was connected to a tracking generator through a 50 Ohm coaxial cable. The metal plate was placed on a J 4 7x5

3 42Omm Semi-Anechoic Chamber Figure 2. The top picture shows the finite ground plane and the excitation source. The close-up of the source is shown b&iw. The hatched lvea is the copper patch on the epoxy block. The center pin is soldered from the mm 11 Monopole Monopole He&lk over tip over chip Figure 3. The three differen models are shown here. wooden table on top of a turn table inside an aneehoic chamber. n order to mimic a chip and a heatsink a piece of epoxy was cut to the size of an asic with a small hole punched at the center for the center pm of the N-type connector. The metal lead was represented by apatchmade of copper tape and the heatsink was placed on the top. The copper patch was excited by soldering the center pin of the N-type corrector with the copper patch from the bottom side (see figure 2). The monopole over finite ground plane was created by replacing the center pin of the N-type connector with a conductor. The monopole over a patch was created by Figure 4. The measurement setup is shown here. replacing the heatsink with a wire soldered on top of the copper patch. The three different models are shown in figure 3. The radiated emissions were measured inside a 3 meter semi-anechoic chamber and the measurements were taken with a horn antenna placed at a distance of 3 meters (see figure 4). FlNlTEELEMENT MODEL The full three-dimensional electromagnetic field outside and inside the structure is calculated through a simulation technique which is based on finite element method. The simulation software Maxwell@ Eminence from Ansoft Corporations [3] is used to compute the fields. The full problem space is divided into many smaller regions and the field in each sub-region is represented with a local function called, basis function. n Eminence, the geometric model is automatically divided into a large number of tetrahedra, forming a finite element mesh. At each vertex, Eminence stores the components of the field that are tangential to the three edges of the tetrahedron. The vector field values at the selected edges, which is tangential to a face and normal to the edge are also stored. The field values inside the element are interpolated from the nodal values. This representation of field values essentially transform Maxwell s equation into a matrix one, which is solved by numerical technique. 786

4 Figure 5. The heatsink inside the pyramid and a close-up of the heatsink and its feed structure are shown here. As described in the previous section, the monopole and the heatsink were simulated as shown in fig 3. The feed structure was a coax and resided under the ground plane. To feed power to the coax, a 2D port was defined at the bottom of the coax and internally generated 2D field solutions served as the boundary condition for the port. The final field solution which was computed for the structure must match the 2D field pattern at the port. As we are interested in radiated field from the structure, it is imperative to define a surface where finite element mesh will be terminated. As the ground plane for our problem was very large compared to the structure itself, it was a daunting task to define the outer boundary so that number of unknowns stay relatively low without affecting the accuracy of the solution. A pyramid with its chopped top was chosen for the radiation surface, whose base coincided with that of ground plane. The base of the pyramid was defined to be perfect E boundary to simulate the ground plane (see figure 5). n the monopole problem, by using a pyramid boundary the number of tetrahedrons was reduced to 4265 tetrahedrons as opposed to 5750 when a box like structure was chosen for the absorbing boundary. Similarly, 8204 tetrahedrons were used for the monopole over chip and 8381 for heatsink simulation. The number of triangles for the 2D port was 75 for the coax feed to the heatsink. The no of matrix elements for the same was t is imperative to keep the solution domain small as to expedite the solution process and also not to exhaust memory resources. An automatic mesh refinement 1GB.z 3GHz 5GHz Figure 6. The radiated emissions from a 30.5mm monopole is shown here. The upper line is vertical polarization and the lower line is horizontal polarization. technique was used to refine the mesh in areas of highest error density. A discrete frequency sweep was employed to calculate the maximum radiated field at a certain distance namely 1 Om. The field pattern of the structure for the frequency of highest emitted radiation was also calculated. n simulation, the dominant mode excitation was used to do field computations. RESULTS The three different models shown in figure 3 are tested in an semi-anechoic chamber. The three models are excited by a tracking generator. The generator swept from 1GHz to 5GHz and the emissions were measured with a double ridge horn antenna. Both the vertical and horizontal polarized fields were measured in all three cases. The radiated emissions are shown in figure 6,7,8 The length of the monopole used in figure 6 is 30.5 mm and it becomes a quarter wave monopole at 2.459GHz. n the measured radiated emissions a fairly wide broadband emission could be observed around 2.35GHz. By doing a peak search the highest emission was observed at G-k For the second model, where a slightly smaller monopole was placed over the epoxy block, a high emission band was observed between 2.4hJHz to 3GHz. The peak search routine selected GHz. The 26mm monopole becomes quarter wavelength at 2.88GHz. A peak search on the radiated emissions from the heatsink yielded 1.575GHz. The radiated patterns were measured for the three models at the three highest emission frequencies found through peak search. The patterns were measured every ten degree and plotted in figure

5 - -1ciHz 3GHz 5GHz Figure 7. The radiated emissions from a 26mm monopole over the mock epoxy chip is shown here. The upper and lower lines are vertical and horizontal polarization. Figu~ 9. The measured vertically polarized radiation pattern for the three different models are shown here. P :. : : : : !...r...~......~..~...~...~...~... : ; : : : : : : %-aj _... in..... i.... i..... j : : : i C- --0 ~bwrnltepbti : &,-.A mnqmbcwchlp : : : o-----o hmt4rkovstchlp ~...~...~.- : i : ltiklz 3GHz 5GHz Figure 8. The radiated emissions from a circular heatsink over the mock epoxy chip is shown here. The upper and lower lines are vertical and horizontal polarization. The simulated results for the three models are shown in figure 10 and 11. The total magnitude of the E-filed from 2 to 3GHz is shown in figure 10 and corresponding radiated pattern of thez directed E-filed is shown in figure 11. CONCLUSON The measured radiated emissions (figure 6,7,8) were found to be the most efficient when the monopole length becomes quarter of a wavelength. Since the monopoles were directed along the z direction, the vertical polarized Figure 10. The simulated radiated emissions for the three models from 2GHz to 3GHz are shown here. The ~sults are in &V/m. emissions were found to be higher than the horizontal polarization. Though the highest emission frequency obtained through peak search (figure 6,7,8) did not coincide with exact quarter wavelength but the overall broadband emission was observed when the monopole length became quarter of a wavelength and this is expected from a monopole. n case of the circular heatsink (figure 8) the peak moved towards lower frequency, which may be caused by circular fins and thick central core of the heatsink The highest emission was observed when the vertical length of the heatsink became sixth or fifth of the wavelength. The radiated pattern for a monopole over an 788

6 d- o P- -S- CO- O...i...i mnopols~ aflmlc~ld S2iGHZ wrc@sovnchip 0 2.1GHz - hnishkwachlp B27oHz Figure 11. The simulated radiation pattern for the three different models are shown here. The results are in dewtm. infinite ground plane measured at a fixed height should not have any variation along &lirection. Though the measured radiated emissions show small variation (figure 9), the overall envelope is reasonably flat. The variations could be caused by reflections in the chamber and variation in the distance between the antenna and the exoerimental setuu. when the turn table was rotated. The ---Jr x. simulated radiated pattern also showed reasonably constant pattern as expected from a monopole. The authors are trying to refine their simulation and experimental models for more accurate measured and computed results. The authors are also trying to correlate the measured observations with the computational results. However, based on the trends observed in figure 6,7 and 8 it can be concluded that a circular heatsink can be simplified as a monopole and its radiation efficiency is highest when the monopole becomes closed to quarter of a wavelength REFERENCBS K. Li, CF. Lee, S.Y. Poh,R.T. Shin, J.A. Kong, Application of FDTD Method to Analysis of ElectromagneticRadiation from VLS Heatsink Configurations, EEE Trans. on EMC, Vol35, No. 2, pp , May S. Radu, Y. Ji, J. Nuebel, J.L. Drew&k, T.P. Van Doren, T.H. Hubing, dentifying an HM Source and Coupling Path in Computer System with Sub-Module Testing, EEE ht. Symp. EMC, Austin, TX, pp , August, Maxwell@ Eminence Usere s Reference, May, 1994, Ansoft Corporation, Pittsburgh, PA. 789!