May 23, 2003
TQFP
Thermal management for semiconductor devices has become critical nowadays A poor thermal management A significant chip junction temperature : depreciate operating performance Considerable temperature gradient : induce substantial thermal stresses Underlying Solution To conduct thermal management of IC packages so as to effectively eliminate heat from packages Perform evaluation of the validity of existing correlation models for heat transfer coefficients Investigate the dependence of thermal performance on design parameters
TQFP Test vehicle: 100-lead, TQFP package Package size: 23.2 x 17.2 x 1.10 mm Die, die attach, die pad, gold wire, leadframe, and MC Test Board: single layer of thermal test board Thermal test board size: 114.3 x 101.6 x 1.57 mm Conforms to the JEDEC standard (EIA/JESD51)
Natural convection testing Based on JEDEC specification (EIA/JESD51) with a minor adjustment A cubic box with 1 ft length on each side Inner wall is coated with a thin layer of black paint Heat detected by IR thermometers is contributed directly from the radiation of the test sample One hole of diameter, 88 mm, is created on the top face of the box Surface temperature of the package can be measured through the hole using IR thermometers Thermal performance : J/A thermal resistance θ ja
Modeled by 1/4 of the package 19,257 brick elements/25,452 nodes Major components : Die, die attachment, die pad, MC, leadframe, and PCB Air Gap between the package and the PCB under the package 0.1 mm thick Neglect of convection/radiation effect Modeled by conduction mechanisms Air conductivity: 0.03
-PCB PCB contains a very intricate, internal structure Glass layers for reinforcement, FR-4 for isolation, and copper traces for electrical conduction Precise assessment of the thermal performance of the TQFP package Detailed modeling => Excessive modeling/computation time Equivalent model => Rule-of-mixture technique For the l-th layer of copper traces, the average thermal conductivity l ( 1 v) k fr vkcopper k ˆ + = 4 The in-plane/out-of-plane bulk thermal conductivity ( n n n n k t kˆ 1 t ( = t / kˆ t in plane = l= 1 l l l= 1 l l l kout of plane l= 1 l= 1 l
Surface temperature measurements by using an IR thermometer NEC IR Camera The black paint coating Temperature interpretation on 3D FE model using a self-development code Notes: Thermal analysis with given chip powers using ANSYS Providing uniform radiation of heat on the entire assembly Measurable range between -50 o C and 2000 o C Spatial resolution: 0.468 mm /Thermal resolution: 0.02 o C
The hottest spot locates in the central region The temperature in source side larger than the non-source side Source side Non-source side Source side Non-source side 0.404 w 0.801 w 0.602 w 1.022 w
Translate thermal images to nodal temperature of the 3D FE model using a numerical interpolation and extrapolation scheme 0.404 (w) 0.801 (w)
A thermal test die for measuring chip junction temperature 6.35 x 6.35 x 0.2974 mm A forward bipolar diode serving as a temperature sensor Diffused resistors serving as a heat generator Resistance of 23.2 (Ohm) TSP Calibration A thermocouple for measuring the ambient temperature Voltage(V) 3.70 3.30 2.90 2.50 10 50 90 130 Κ K=-9.41E-3 (V/ ) o Temp.( C) 1
A satisfactory agreement is presented between these two approaches 3-7% discrepancy in T j ; 8-10% difference in θ ja As power is larger than 0.4 w, the θ ja becomes stable The proposed methodology is substantially validated 0.404(w) 0.602(w) 0.801(w) 1.022(w) Ambient temp. 23.8 24.0 24.1 23.9 T j (the proposed) 40.30 47.95 56.08 63.51 T j (measurement) 41.7 50.4 59.0 68.0 Differences -3.4% -4.9% -4.9% -6.6% θ ja (the proposed) 40.80 39.79 39.92 38.76 θ ja (measurement) 44.30 43.83 43.49 43.06 Thermal resistance( C/W) o 60 40 20 Differences -7.9% -9.2% -8.2% -10.0% 0 0.00 0.40 0.80 1.20 Power(W)
To realize the locations of thermal barriers in the main thermal paths To calculate the surface-to-the-ambient heat flux of those components directly exposed to the air Thermal dissipation( / ) o o 100 80 60 40 PCB serves as the main external heat sink for the package 80% : PCB 12% : MC 7% : LF 0.6%: AG 0.4%: Other Two Main Heat Conduction Path: 20 0 Molding PCB Lead Air Compound Frame Gap Others Die->MC->Air Die-> LF-> PCB->Air
Practical and efficient to apply FE approximations for further parametric analysis with HT coefficients imposed as surface loads or the essential boundary Considerable limited benchmark data of the usefulness of these correlation models Additional studies: evaluate the validity of existing correlation models Most of these correlation models do not account for radiation effect Ellison (1989), Aghazadeh and Malik (1990), Mulgaonker et al. (1994), and Edwards et al. (1995) They are modified by Integration of the radiative HT coefficient correlation model introduced in Ridsdale et al. (1996)
Based on either an isothermal or an isoflux assumption Ellison s correlation model (1989) h = 0.83 f / ( T ) n c L ch Aghazadeh and Malik s correlation model (1990) ( ) 0.45 ( ) 0. 184 hc = 1.643 Lch T Mulgaonker et al. s suggestion (1994) : 8.5 w/m 2.o C Edwards et al. s correlation model (1995) a c [ D / P ] 0. 5 h = b [ ] 2 d h c 20.77 a a = 7.8 h c c Ridsdale et al. s correlation model (1996) q = h c = 7.8 ( Ba Pa ) / Ba [( B P ) / B ] 0. 5 = 43.4µ a a a ( h + h )( T T ) c r w a ( / L ) 0. 25 ( 2 2 + T )( T T ) hc = 1.581 T ch h = Bfe T + r w a w a
A simple numerical experiment is performed, based on the Ridsdale et al s correlation model About 9-12 % of disagreement No the radiation effect About 10-16% of disagreement No the convection effect Convection Only 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 49.21 60.91 72.37 84.55 T j (measurement) 41.7 50.4 59.0 68.0 Differences +18.0% +20.9% +22.7% +24.3% Radiation Only 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 49.57 61.89 73.97 86.77 T j (measurement) 41.7 50.4 59.0 68.0 Differences +18.9% +22.8% +25.4% +27.6% Both Effects 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 45.31 55.43 65.34 75.82 T j (measurement) 41.7 50.4 59.0 68.0 Differences +8.7% +10.0% +10.7% +11.5%
( ) Ridsdale et al. s correlation model 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 45.31 55.43 65.34 75.82 T j (measurement) 41.7 50.4 59.0 68.0 Differences +8.7% +10.0% +10.7% +11.5% θ ia (simulation) 57.18 52.22 51.47 50.80 θ ia (measurement) 44.30 43.83 43.49 43.06 Differences +20.0% +19.1% +18.3% +18.0% Modified Ellison s correlation model 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 43.87 53.08 62.05 71.41 T j (measurement) 41.7 50.4 59.0 68.0 Differences +5.0% +5.3% +5.2% +5.0% θ ia (simulation) 49.63 48.31 47.37 46.49 θ ia (measurement) 44.30 43.83 43.49 43.06 Differences +12.0% +10.2% +8.9% +8.0%
( ) Modified Aghazadeh and Mallik s correlation model 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 43.63 53.02 61.82 71.97 T j (measurement) 41.7 50.4 59.0 68.0 Differences +4.6% +5.2% +4.8% +5.8% θ ja (simulation) 49.04 48.21 47.08 47.04 θ ja (measurement) 44.30 43.83 43.49 43.06 Differences +10.7% +10.0% +8.3% +9.2% Modified Mulgaonker et al. s correlation model 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 43.87 53.75 63.55 74.03 T j (measurement) 41.7 50.4 59.0 68.0 Differences +5.2% +6.6% +7.7% +8.9% θ ja (simulation) 49.63 49.43 49.24 49.05 θ ja (measurement) 44.30 43.83 43.49 43.06 Differences +12.0% +12.8% +13.2% +13.9%
( ) Modified Edwards et al. s correlation model 0.404(w) 0.602(w) 0.801(w) 1.022(w) T j (simulation) 43.13 52.67 62.14 72.25 T j (measurement) 41.7 50.4 59.0 68.0 Differences +3.4% +4.5% +5.3% +6.3% θ ja (simulation) 47.80 47.63 47.48 47.31 θ ja (measurement) 44.30 43.83 43.49 43.06 Differences +7.9% +8.7% +9.2% +9.9%
( ) The Ridsdale et al s model presents the worst result against the experiment while the modified Edwards et al s model provides the most accurate. Ellison, Aghazadeh and Malik, and Edwards et al. s models result in the same accuracy to some extend. o o Error( / ) 25 20 15 10 5 0 Ridadale Ellison Aghazadeh and Mallik Average errors in θ ja Average error of thermal resistance Average errors in T j Average error of junction temperature Mulgaonker Edwards
( ) By neglecting the Ridsdale et al s model, reasonable results are obtained among the other four models: 3~9% of differences in the chip junction temperature and 8~14% in the J/A thermal resistance Similar degree of calibration discrepancy against the thermal test die approach as the proposed methodology did For a small device as TQFP These correlation models, except the Ridsdale al. s model, are fairly reliable Reasonable calibration is obtained by using a single, fixed convective HT coefficient over the entire surfaces of the assembly, as proposed by the modified Mulgaonker et al s model.
Compared with those taken by the IR thermometer, a large deviation in temp. distribution, particularly in PCB In reality, the conductance of PCB is anisotropic Modeling PCB by the Rule-of-Mixture technique Simplify the modeling process but may also screen out the anisotropy of thermal conductance of PCB Result in a poor estimation of local temperature distribution Simulation IR Thermography Simulation IR Thermography 0.404 w 0.801 w
A consistent dependence trend of the thermal performance w.r.t these four design parameters A consistent dependence trend of the thermal performance w.r.t these four design parameters A decrease of 2 o C/w of the θ ja : 25-30% increase of the thermal conductivity of the MC 25-30% increase of the in-plane thermal conductivity of the PCB 90% growth for the out-of-plane thermal conductivity of the PCB 41% enhancement for the die size Thermal resistance( C/W) o 80.0 60.0 40.0 20.0 K of EMC K,K of PCB x y K of PCB 0.0-120.0-80.0-40.0 0.0 40.0 80.0 The variation of the parameters( / 0) z Die size
PBGA
PBGA Test vehicle: 272 Pin (256 solder joints/16 thermal balls), planar MCM Package size: 27.0 27.0 2.33 mm Two dies, die attach, die pad, gold wire, solder joints, BT substrate and MC Test Board: Four layers of thermal test board Thermal test board size: 114.5 x 101.5 x 1.6 mm Conforms to the JEDEC standard (EIA/JESD51)
MCM Elison Risdale et al. Correlation Model h h c r = = 0.83 f Bef ( T T ( 2 w w T L + T ch 2 a a ) n )( T w ( W + T a ) / m 2 K) ( W / m 2 K) h = h c + h r
Nodes 21,319 Elements 24,586
( ) Elison Risdale et al. Correlation Model
MCM
1W Tj (Small die)( o C) Tj (Big die)( o C) Simulation 65.4 54.6 Experiment (Forward Voltage) 70.6 52.5 Difference(%) -7.4 4.0 Simulation (IR image) 66.3 56.5 Experiment (Forward Voltage) 70.6 52.5 Difference( ) -6.1 7.6 Simulation 65.4 54.6 Simulation (IR image) 66.3 56.5 Difference( ) -1.4 3.4
PBGA 100 80 TQFP Thermal dissipation( / ) o o 60 40 20 0 Molding PCB Lead Air Compound Frame Gap Others
PCB 59%~96% PCB PCB contains a very intricate, internal structure Glass layers for reinforcement, FR-4 for isolation, and copper traces for electrical conduction Precise assessment of the thermal performance of the TQFP package Detailed modeling => Excessive modeling/computation time PCB
T P G B EIA/JESD51-9 PCB
PBGA 256-pin PBGA Molding compound Chip Substrate Wireframe Solder joint PCB Thermal ball Heatspreader 27.00 24.00 4(REF) 27.00 24.00
PCB (benchmark)
PCB
A Cu air k = υk + ( 1 υ) k B C D 1 Cu air k22 = k 11 = k33 = υk + (1 υ) k υ 1 υ + Cu air k k A B C D A B C y, 2 x, 1 o 0.0 2 D y 42.5 2 o 1 x y o 19.9 1 x
power power FR FR ground ground yy xx k k k k k ε ε ε + + = = 4 4 power power FR FR ground ground zz k k k k ε ε ε + + = 4 4 1 Ground plane Power plane FR4 4 ) (1 + = FR i Cu i i k k k υ υ power ground i, =
T t T 1 r o Q zz z Qq k zz via zz = k = πr via zz 2 o T z t Q ( T T2 ) 1 zz z T Q 1 rr T2 Qq k rr via rr via T = krr r Q = rr πt( T 1 T 2 2 ) Q zz = πr 2 o q zz Q = 2πr rr o tq rr 2 r r
PBGA 18,313 solid elements 216 shell elements 21,362 nodes
PCB
IR Ellison(1989) Ridsdale (1994)
256 PBGA
57.00 61.73-7.7% Ta = 23.5 Power = 1.500 Watts 32 37 42 47 52 27
27 42 Benchmark Model Equivalent Model