BOOKCOMP, Inc. — John Wiley & Sons / Page 986 / 2nd Proofs / Heat Transfer Handbook / Bejan 986 HEAT TRANSFER IN ELECTRONIC EQUIPMENT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [986], (40) Lines: 1080 to 1130 ——— 1.48141pt PgVar ——— Short Page * PgEnds: Eject [986], (40) the lateral conductivity and the transverse (through the thickness) conductivity is sufficient to achieve the desired accuracy. As in all discretized thermal models, the error inherent in this technique is inversely proportional to the volume represented by each resistor. Because the copper and dielectric layers in the board may not be continuous, due to the presence of traces and vias, it is very difficult to model the conduction in this layer in full detail. If n c is the number of cutouts (holes for interconnection and etched copper) in the copper layers and d c is the diameter of each cutout, a conduction factor for conduction normal to the copper layers may be defined as κ n = WL −n c (πd 2 c /4) WL (13.54) where W and L are the width and length of the printed circuit board, respectively. The factor κ n simply represents the fraction of the conduction area missing due to the cutouts. Similarly, a conduction factor for the in-plane direction may be defined as κ p = W −n c d c W (13.55) The thermal resistance of each of the layers making up the printed circuit board in the normal direction can be written as R n x = δ x κ p k x LW (13.56) where δ x and k x are the thickness and thermal conductivity of layer x. Similarly, the thermal resistance of each of the layers in the in-plane direction can be written as R p x = L κ p k x δ x W (13.57) The thermal resistances of the various layers are in series in the normal direction and in parallel in the in-plane direction. Using the law of electrical resistances, the total resistance R n t in the normal direction can be written as R n t = δ t k en LW = δ x k x LW (13.58) where δ t is the total thickness of the interposer or motherboard and k en is the equiv- alent normal thermal conductivity. Similarly, the total resistance R p t in the in-plane direction can be written as 1 R p t = k ep δ t W L = k x δ x W L (13.59) BOOKCOMP, Inc. — John Wiley & Sons / Page 987 / 2nd Proofs / Heat Transfer Handbook / Bejan LENGTH-SCALE EFFECTS ON THERMOPHYSICAL PROPERTIES 987 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [987], (41) Lines: 1130 to 1145 ——— 0.927pt PgVar ——— Short Page PgEnds: T E X [987], (41) Equations (13.56) and (13.57) can be used to estimate the equivalent in-plane and normal thermal conductivities for the motherboard and interposer layers. Thermal Vias Existing electrical vias, or plated through holes, as well as “thermal vias” embedded in the board, enhance the transverse thermal conductivity of a PCB and help to reduce the resistance to heat flow in the direction perpendicular to the plane of the PCB. Thermal vias are also used in chip scale packages (CSPs) and ball grid array (BGA) packages to provide thermal paths of reduced resistance from the chip directly to the PCB. Due to the geometric complexity of most PCBs, a combination of analytical and experimental approaches is usually needed to estimate Figure 13.21 Temperature distribution in an epoxy–glass substrate with and without copper planes. (From Howard et al., 1984.) BOOKCOMP, Inc. — John Wiley & Sons / Page 988 / 2nd Proofs / Heat Transfer Handbook / Bejan 988 HEAT TRANSFER IN ELECTRONIC EQUIPMENT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [988], (42) Lines: 1145 to 1175 ——— -0.3pt PgVar ——— Normal Page PgEnds: T E X [988], (42) the thermal conductivity of a particular board, but much can be learned from the analysis of a PCB with a relatively simple wiring pattern. When a large number of vias are present, it is possible to determine the thermal conductivity in the normal direction by assuming that the higher-conductivity vias provide the dominant thermal conduction path for heat. The thermal conductivity of the dielectric can therefore be neglected and the thermal conductivity of the dielectric layer in the normal direction can be computed using k n = (1 − κ n )k cu (13.60) where the factor k n is computed using the via diameter and eq. (13.54). The conduc- tivity in the in-plane direction for the dielectric layers can still be assumed to be that of the dielectric alone. Effect of Trace Layers Use of additional in-plane (or artificially thickened) cop- per layers can serve to substantially increase the effective lateral thermal conductivity of a PCB, as achieved by Hewlett-Packard in the “finstrate” technology of the mid- 1980s. The reduced in-plane thermal resistance can help transport heat to the edges of the board and/or dramatically reduce the local temperature rise of a high-power component mounted to the PCB. The beneficial effect of such trace layers can best be illustrated with a specific example using a finite-element model and numerical simulation to generate the temperature distributions (Lewis et al., 1996). The following results were obtained with a 7.47-mm-square chip size, dissipat- ing 1 W, mounted on various PCBs (Howard et al., 1984). The results for case 1, using an epoxy–glass substrate without any copper, are shown in Fig. 13.21a in the form of a three-dimensional plot of the temperature distribution in the substrate. The poor thermal conductivity of the epoxy glass is seen to yield a 160°C maximum temperature rise under the center of the chip. With a 1-oz copper trace, maximum temperature rise under the center of the chip is reduced to 21.8°C, whereas with 2-oz copper it is 13.9°C and with 4-oz copper it is 8.9°C, as shown in Fig. 13.21b, c, and d, respectively. 13.4 CONVECTIVE PHENOMENA IN PACKAGING 13.4.1 Printed Circuit Boards in Natural Convection Despite increasing performance demands and advances in thermal management tech- nology, natural convection air cooling of electronic equipment continues to command substantial attention. The simplicity, reliability, and low cost of air natural convec- tion make this cooling mode a popular choice for individual IC packages, populated printed circuit boards, and the heat sinks attached to modules, power supplies, and chip packages. The heat transfer coefficient for natural convection can be related to the temperature difference between the surface and the ambient fluid, along with the fluid’s thermal and fluid properties, as seen in the correlations listed in Tables 13.4 and 13.5, respectively. BOOKCOMP, Inc. — John Wiley & Sons / Page 989 / 2nd Proofs / Heat Transfer Handbook / Bejan CONVECTIVE PHENOMENA IN PACKAGING 989 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [989], (43) Lines: 1175 to 1205 ——— 7.097pt PgVar ——— Normal Page PgEnds: T E X [989], (43) Figure 13.22 Composite correlation for parallel isothermal plates. Vertical heated channels formed by parallel PCBs are a frequently encountered configuration in natural convection cooling of electronic equipment. The historical work of Elenbaas (1942) provides the foundation for much of the effort dealing with natural convection in such smooth, isothermal, parallel-plate channels. His exper- imental results for isothermal plates were confirmed via numerical simulation by Bodoia and Osterle (1964), who showed that the results could also be used for the constant heat flux condition. Other researchers have extended this pioneering work to include asymmetrically heated channels, including the case of the single insulated wall (Aung, 1972; Aung et al., 1972; Miyatake et al., 1973; Miyatake and Fujii, 1972). These many studies revealed that the value of the convective heat transfer coefficient lies between two extremes associated with the separation distance between the plates or the channel width. For wide spacing, the plates appear to have little influence upon one another, and the heat transfer coefficient in this case achieves its isolated plate limit. On the other hand, for closely spaced plates or for relatively long channels, the fluid attains the fully developed velocity profile and the heat transfer rate reaches its fully developed value. Intermediate values of the heat transfer coefficient can be obtained from a judi- cious superposition of these two limiting phenomena, as presented in the composite expressions proposed by Bar-Cohen and Rohsenow (1984). Figure 13.22 shows one such correlation for natural convection heat transfer from isothermal parallel plates. Composite correlation for other situations such as symmetrically heated isothermal or isoflux surfaces are available in the literature (Kraus and Bar-Cohen, 1983). A compilation of these natural convection heat transfer correlations for an array of vertically heated channels is listed in Table 13.8. The heat transfer from isothermal plates is expressed in terms of the Elenbaas number, defined as BOOKCOMP, Inc. — John Wiley & Sons / Page 990 / 2nd Proofs / Heat Transfer Handbook / Bejan 990 HEAT TRANSFER IN ELECTRONIC EQUIPMENT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [990], (44) Lines: 1205 to 1246 ——— 0.87271pt PgVar ——— Normal Page PgEnds: T E X [990], (44) TABLE 13.8 Natural Convection Heat Transfer Correlations for an Array of Heated Vertical Channels Condition Fully Developed Limit Composite Correlation Symmetric isothermal plates Nu 0 = El 24 Nu 0 = 576 El 2 + 2.873 √ El −1/2 Asymmetric isothermal plates Nu 0 = El 12 Nu 0 = 144 El 2 + 2.873 √ El −1/2 Symmetric isoflux plates Nu 0 = El 48 Asymmetric isoflux plates Nu 0 = El 24 Symmetric isoflux plates based on midheight temperature Nu 0 = El 12 Nu 0 = 12 El + 1.88 (El) 2/5 −1/2 Asymmetric isoflux plates based on midheight temperature Nu 0 = El 6 Nu 0 = 6 El + 1.88 (El) 2/5 −1/2 El = C p ρ 2 gβ(T w − T amb )b 4 µk f L (13.61) In eq. (13.61), b is the channel spacing, L the channel length, and (T w − T amb ) the temperature difference between the channel wall and the ambient or channel inlet. The equations for the uniform heat flux boundary condition are defined in terms of the modified Elenbaas number, which is defined as El = C p ρ 2 gβq b 5 µk 2 f L (13.62) In eq. (13.62), q is the heat flux leaving the channel wall(s). A different type of asymmetry can occur if adjacent channel walls are isothermal but at different temperatures or isoflux but dissipating different heat fluxes. For the case where the walls are isothermal but at different wall temperatures T w1 and T w2 , Aung (1972) defined an asymmetry parameter as r T = T w1 − T 0 T w2 − T 0 (13.63) in which T 0 is the air temperature at the channel inlet. Then the heat transfer could be calculated using the parameters listed in Table 13.9. In the case of symmetric isoflux plates, if the heat flux on the adjacent walls is not identical, the equations in Table 13.8 can be used with an average value of the heat flux on the two walls. BOOKCOMP, Inc. — John Wiley & Sons / Page 991 / 2nd Proofs / Heat Transfer Handbook / Bejan CONVECTIVE PHENOMENA IN PACKAGING 991 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [991], (45) Lines: 1246 to 1276 ——— 0.22913pt PgVar ——— Normal Page PgEnds: T E X [991], (45) TABLE 13.9 Nusselt Number for Symmetric Isothermal Walls at Different Temperatures r T Nu 0 /El 1.0 1/24 0.5 17/405 0.1 79/1815 0.0 2/45 Source: Aung (1972). 13.4.2 Optimum Spacing The composite relations presented in Section 13.4.1 may be used in optimizing the spacing between PCBs. For isothermal arrays, the optimum spacing maximizes the total heat transfer from a given base area or the volume assigned to an array of PCBs. Figure 13.23 shows the maximum power dissipation from a PCB card cage as a function of the PCB spacing. It is clear from the figure that there is an optimum PCB spacing at which maximum power can be dissipated in the PCBs for the same PCB-to-ambient temperature difference. In the case of isoflux parallel-plate arrays, the power dissipation may be maxi- mized by increasing the number of plates indefinitely. Thus, it is more practical to define the optimum channel spacing for an array of isoflux plates as the spacing that will yield the maximum volumetric heat dissipation rate per unit temperature differ- ence. Despite this distinction, the optimum spacing is found in the same manner. Figure 13.23 Variation of total power dissipation from a PCB array with PCB spacing. BOOKCOMP, Inc. — John Wiley & Sons / Page 992 / 2nd Proofs / Heat Transfer Handbook / Bejan 992 HEAT TRANSFER IN ELECTRONIC EQUIPMENT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [992], (46) Lines: 1276 to 1312 ——— 2.23035pt PgVar ——— Normal Page PgEnds: T E X [992], (46) TABLE 13.10 Optimum Spacing for Natural Convection Cooled Arrays of Vertical Plates or Printed Circuit Boards Condition Optimum Spacing Symmetric isothermal plates b opt = 2.714 P 1/4 Asymmetric isothermal plates b opt = 2.154 P 1/4 Symmetric isoflux plates b opt = 1.472R −0.2 Asymmetric isoflux plates b opt = 1.169R −0.2 Source: Kraus and Bar-Cohen (1983). The optimal spacings for different conditions are tabulated in Table 13.10. The parameter b opt in Table 13.10 represents the optimal spacing, and the plate to air parameter P is given as P = C p (ρ f ) 2 gβ∆T 0 µ f k f L (13.64) where ∆T 0 is the temperature difference between the PCB and the ambient. The parameter R in Table 13.10 is given by R = C p (ρ f ) 2 gβq µ f k 2 L (13.65) These smooth-plate relations have proved useful in a wide variety of applications and have been shown to yield very good agreement with measured empirical results for heat transfer from arrays of PCBs. However, when applied to closely spaced PCBs, these equations tend to underpredict heat transfer in the channel due to the presence of between-package wall flow and the nonsmooth nature of the channel surfaces. 13.4.3 Printed Circuit Boards in Forced Convection The thermal analysis for PCB arrays exposed to forced airflow can be performed in a manner similar to that described for natural convection. If the PCB spacing is large enough that the boundary layers developing on adjacent boards are not expected to interfere with each other, the heat transfer correlations developed for the isolated flat plate in Table 13.4 can be used. Note that in doing so, it is implicitly assumed that the PCBs are densely populated and that the gap between adjacent components on the PCB is small enough such that the developing boundary layers are not interrupted. Additionally, for such cases, even though heat is generated within discrete compo- nents on the board, it is possible to assume that there is a uniform heat flux condition on the surface of the components because the component spacing is very small. When BOOKCOMP, Inc. — John Wiley & Sons / Page 993 / 2nd Proofs / Heat Transfer Handbook / Bejan CONVECTIVE PHENOMENA IN PACKAGING 993 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [993], (47) Lines: 1312 to 1349 ——— 0.31216pt PgVar ——— Normal Page PgEnds: T E X [993], (47) TABLE 13.11 Nusselt Number for Developing Laminar Flow in Isoflux Channels x + = 2 x/D h Re ·Pr Nu 0.002 18.50 0.010 9.62 0.020 7.68 0.10 5.55 0.20 5.40 ∞ 5.39 Source: Heaton et al. (1964). the spacing between the PCBs is small, the boundary layers from the adjacent boards are expected to merge. In such cases the heat transfer coefficient can be estimated using the channel flow correlations discussed later in this section. The channels formed by PCBs are generally relatively short and the ratio of chan- nel length to the spacing between the boards is rarely larger than 100. In such cases, distinct velocity and temperature boundary layers are expected to develop over much of the length of the PCB. Consequently, correlations or tables for developing flow solutions must be used to correctly estimate the heat transfer from the PCBs. Heaton et al. (1964) determined the local Nusselt numbers expected during developing lam- inar flow between isoflux walls. These are listed in Table 13.11. It is clear from the tabulated values that the heat transfer coefficient in the developing region can be two to three times higher than that obtained using the fully developed flow correlations. Similar data for a variety of flow geometries have been summarized by Kays and Crawford (1993). It is often necessary in the thermal analysis of electronics equipment to evaluate the heat transfer coefficient for a discrete component located on a poorly conducting PCB, positioned at a downstream location in a channel. Thermal performance of such a discrete component can be assessed by assuming a uniform heat flux condition on the component surface and assuming that the boundary layer development begins at the leading edge of the component. The Nusselt number values tabulated in Table 13.11 can now be used to evaluate the heat transfer coefficient. For forced laminar flow in long or very narrow parallel-plate channels, the heat transfer coefficient attains an asymptotic value under fully developed conditions (the fully developed limit), which for symmetrically heated channel surfaces is approxi- mately equal to h = 4k f D h (13.66) where D h is the hydraulic diameter defined in terms of the flow area A and the wetted perimeter of the channel P w : BOOKCOMP, Inc. — John Wiley & Sons / Page 994 / 2nd Proofs / Heat Transfer Handbook / Bejan 994 HEAT TRANSFER IN ELECTRONIC EQUIPMENT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [994], (48) Lines: 1349 to 1373 ——— 0.28415pt PgVar ——— Normal Page * PgEnds: Eject [994], (48) TABLE 13.12 Nusselt Numbers for Fully Developed Laminar Flow in Channels Cross Section Aspect Ratio Nu for T Constant q Constant Rectangular 1 2.98 3.61 2 3.39 4.12 4 4.44 5.33 Infinite 7.54 8.23 Infinite with only one 4.86 5.38 wall heated Circular — 3.66 4.36 Triangular — 2.35 3.00 Source: Kays and Crawford (1993). D h = 4A P w (13.67) Table 13.12 provides the values of the fully developed Nusselt numbers Nu and heat transfer coefficient h for several channel geometries and for both isoflux (q const) and isothermal (T const) conditions. For the case where adjacent channel walls each have a different uniform value of heat flux, Kays and Crawford (1993) also provide Nu 1 = 5.385 1 − 0.346(q 2 /q 1 ) (13.68) In the inlet zones of such parallel-plate channels and along isolated plates, the heat transfer coefficient varies with the distance from the leading edge and can be com- puted using the flat plate correlations provided in Tables 13.4 and 13.5. For calculation of the junction temperatures of components mounted on PCBs and cooled by forced convection, Witzman (1998) proposed T j − T a = ∆T a + ∆T jc + ∆T ca (13.69) where ∆T a is the rise in the temperature of air as it flows over the PCB, ∆T jc is the temperature drop between the silicon junction and the outer surface of the PCB- mounted component, and ∆T ca is the temperature drop between the component sur- face and the local ambient temperature. With increasing component power dissipation and more densely populated PCBs, it is important to determine the heat transfer co- efficient on the component surface accurately in order to compute ∆T ca . Witzman (1998) and Graham and Witzman (1988) obtained the local heat transfer coefficients through careful empirical measurements. They found that in most cases the temper- ature varied axially, as well as laterally, along the surface of the PCB, creating a high-temperature area in the center of the PCB rack. Based on their measurements for 20-mm board spacing and uniform component power dissipation ranging from 10 to 30 W, Witzman (1998) proposed BOOKCOMP, Inc. — John Wiley & Sons / Page 995 / 2nd Proofs / Heat Transfer Handbook / Bejan JET IMPINGEMENT COOLING 995 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [995], (49) Lines: 1373 to 1405 ——— -0.21452pt PgVar ——— Normal Page PgEnds: T E X [995], (49) T j − T a = 76.1q 0.8 1 + 0.23(n − 2) 0.62 (13.70) where n is the position of the component (n = 1 is the bottom row), T j − T a the average temperature rise of the nth component above the ambient temperature, and q the component power dissipation. Moffat and Ortega (1988) recommended that a heat transfer coefficient h ad based on the adiabatic temperature of the heated element be used when utilizing correlations for heat transfer determined from actual measurements of simulated electronic cool- ing systems. This recommendation is based on the fact that h ad is independent of the geometry and flow and does not depend on the upstream flow pattern. Additionally, h ad can be measured by heating up only one element on the PCB and allows for superposition of solutions. The superposition formulation for computing the junction temperature of the ith component on a PCB can be written as T j − T in = q i h i A i + j j=1 θ ij q j h j (13.71) where T in is the inlet coolant temperature and θ ij is the thermal wake function, defined as θ ij = T i − T in T j − T in (13.72) In eq. (13.71), j represents the components upstream from component i. The first term in eq. (13.71) calculates the temperature rise of the component above its own adiabatic temperature and the second term adds the difference between the adiabatic temperature of the component and the inlet temperature of the coolant. Moffat and Ortega (1988) have also summarized an extensive set of experimental data for several geometric arrangements in a channel. The convective resistance in the fluid region adjacent to the wall in turbulent flow is nearly independent of the exact thermal boundary condition on the channel wall or on any discrete component. Consequently, the correlations listed in Table 13.4 can be used to obtain a fairly accurate estimate of the heat transfer coefficient in turbulent flow for both isothermal and isoflux wall boundary conditions. The constants and eigenvalues required to estimate a heat transfer coefficient in developing turbulent flow have been summarized by Kays and Crawford (1993). 13.5 JET IMPINGEMENT COOLING 13.5.1 Introduction In recent years the use of impinging fluid jets for thermal management of electronic components has received extensive attention. The high heat transfer coefficients that can be attained in this cooling mode, the ability to vary and control the heat transfer . BOOKCOMP, Inc. — John Wiley & Sons / Page 986 / 2nd Proofs / Heat Transfer Handbook / Bejan 986 HEAT TRANSFER IN ELECTRONIC EQUIPMENT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [986],. al., 1984.) BOOKCOMP, Inc. — John Wiley & Sons / Page 988 / 2nd Proofs / Heat Transfer Handbook / Bejan 988 HEAT TRANSFER IN ELECTRONIC EQUIPMENT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 [988],. 1983). A compilation of these natural convection heat transfer correlations for an array of vertically heated channels is listed in Table 13.8. The heat transfer from isothermal plates is expressed