- die 2:
package: pbga
power: 1,5 w
die dimensions: 8 x 8 x 0,82 mm
 

other powers: 1,693 w
air flow: no airflow

then, the model has been drawn also using some packages downloaded from flopack, the web site of flomerics. particularly, we used flopack for the two main dies, the crypto pcb and the pci pcb. in tables 1 and 2 dimensions and materials of the main components of this model are displayed.

table 1: dimensions of the main components used in model 023

table 2: thermal conductivity of the main materials used in model 023.

fig. 2: the model 023 of the cryptographic coprocessor
(a 3d view of the model in the "visualization window" of flotherm)

fig. 3: the model 023 of the cryptographic coprocessor
(top view in the "drawing board window" of flotherm)

fig. 4: the 4758 pci cryptographic coprocessor in a wireframe view (model 023)

for this model, a grid array of 111 x 84 x 37 in the x, y and z directions (344988 cells) was used.

table 3 displays the six different cases that were considered. the alternative was related to the kind of metal, coating and protection that were proposed both for the inner and the external cover of the enclosure. these simulations were run with the previous release of flotherm (version 2.2), but also the new 3.1 version has been tested with this project. specifically, the new available feature "command centre" has revealed all its potentiality in this parametric study.

table 3 summarizes all the combinations studied, whereas table 4 gives the final results.

model 023: results

simulations reached the convergence, without any problems, in some hundreds of iterations. results obtained from runs performed for the model 023 of the 4758 pci cryptographic coprocessor are summarized in table 4 (see table 3 for the description of runs). from that table it is clear that the combination of materials used in run 1 is the solution that provides the lowest temperature in all the monitored points.

the best results have been obtained with run 1. it presented the following materials: an external cover made of copper, a coating of tin and a protection of a typical epoxy resin. the inner cover made of copper had just the protection made of the same epoxy resin.

table 3: combination of different materials used for each run in the
thermal simulation of model 023.

table 4: final results for model 023.

simulations made with flotherm were executed in parallel with experiments in order to be able to fix the model and to compare final temperatures. in table 5 there is an example of the really good relation between the two different approaches. those values have been obtained with an ambient temperature of 23°c.

fig. 5: an image of final results from a run of model 002.

table 5: comparison of results from experiments and flotherm at tamb. = 23°c

model 002: setting of the simulation

in order to create a model that accomplished the maximum level of security (fips level 4), the previous cryptographic coprocessors was reviewed. the new model, named 002, differed from model 023 both in the encapsulation concept and in some hardware particulars. in this article, the hardware changes are not faced.

the encapsulation has been reinforced because this model presented both the tamper detection mesh and the potting resin. this fact caused a series of thermal problems that have been solved after having found good thermally conductive gap fillers. furthermore, in order to get lower the maximum junction temperature of die 1, a no cap version of this device has been used. a thermal pad has been put between the chip and the inner cover.

the next figure presents a view of the mesh of the model 002:

fig. 6: mesh of the 4758 pci cryptographic coprocessor (model 002).

in table 7 there is the description of the main dimensions of the model; there are some differences respect to the model 023 because of the presence of the potting resin and the tamper detection mesh that wrapped the inner cover.

table 6: main materials properties of model 002.

fig. 7: layout of the model 002.

table 7: main dimensions of model 002.

the initial conditions set for this model are the same of those ones set for model 023 (see page 3). in table 8 there are the thermal conductivity coefficients of the main materials used.

table 6 indicates the choice that has been done for materials, coating and protection of this model.

table 8: main materials properties of model 002.

fig. 8: the model 002 of the cryptographic coprocessor
(x side view in the "drawing board window" of flotherm)

at the end of the setting, a grid array of 126 x 102 x 44 in the x, y and z directions (565488 cells) was employed.

fig. 9: the model 002 of the cryptographic coprocessor (3d view in flovisual)

model 002: results

first runs gave some problem of convergence that has been partially solved by using a subdomain inside the inner cover. this subdomain has been set with the same pressure of the external ambient and with a temperature equal to 68°c.

the following table displays the final results obtained for model 002:

table 9: final results for model 002.

in figure 10 a thermography has been displayed. this is just an example of the different experiments that have been executed to compare experimental with numerical results. at the beginning, these comparisons were useful for setting the thermal simulation; then they have confirmed results obtained using flotherm as already said for the previous model.

fig. 10: thermography of the 4758 pci cryptographic coprocessor.
(in this image there is a model not described in this article)

preliminary study of a new cryptographic coprocessor device

due to our well-known experience in cryptography applications in the ems industry, we had been requested of performing several preliminary studies concerning new cryptographic cards over this last year. each time, thermal aspects were very interesting and challenging because of the peculiarity of cryptographic cards, i.e. high power sources embedded in closed boxes. this gives a particular importance to the contribution that thermal simulations provide in the very beginning phase of a thermal evaluation.

what is presented in this article is the thermal analysis performed, during the preliminary study of a new cryptographic card, for a customer interested in cryptographic applications. we were asked for a card that should have to satisfy the fips pub 140-1. based on previous experiences, we realised a two metal boxes card with the potting resin and the tamper detection mesh.

in order to forecast the thermal behavior of the card, a thermal simulation was set and realised by means of flotherm 3.1.

in the next page follow the setting data regarding this project:

- ambient temperature: 42°c
- ambient pressure: 1 atm

- air properties:
thermal conductivity
(at 30°c): 0,0261 w/m×k
viscosity (at 30°c): 1,84e-5 n×s/m2
density (at 30°c): 1,1614 kg/m3

- crypto card position: horizontal

- die 1:
package: wirebonded tbga
power: 5.0 w
die dimensions: 11 x 11.3 x 0.70 mm

- die 2:
package: wirebonded tbga
power: 4.0 w
die dimensions: 18 x 18 x 0.30 mm

- die 3:
package: wirebonded tbga
power: 5.0 w
die dimensions: 15.14 x 13.3 x 0.30 mm

- other power sources: 7.03 w
- air flow: no airflow
- total power source: 21.03 w
- overall dimensions: 320 x 113.4 x 139.2 mm

thermal model has been prepared with the help of some packages downloaded from the flopack web site (all the three dies and the pcb). in tables 10 and 11 dimensions and materials of the main components of this model are displayed.

in order to make these simulations nearer the reality, some cylinders of copper have been drawn through the substrate under the main devices. in this way, we tried to simulate the presence of thermal vias. but this experiment didn't give an appreciable difference in final temperatures if compared to the simulation without them. the reason is probably that a big number of thermal vias would be necessary in order to create a significant path of heat. this fact makes it not opportune because it would mean to have an excessively high number of cells.

at the end of the setting, a grid array of 147 x 138 x 71 in the x, y and z directions (1440306 cells) has been obtained.

fig. 11: z side view of the cryptographic coprocessor device described in this article.

fig. 12: y side view of the cryptographic coprocessor device described in this article.

table 10: main dimensions.

table 11: thermal conductivity of the main materials.

fig. 13: drawn of a wirebonding tbga in the web site of flopack.

final results have been obtained without big difficulties.

table 12: final results.

also for this model, the comparison between results obtained by means of flotherm and in the experimentations has been satisfactory.

fig. 14: plane of temperatures over the cryptographic coprocessor device.

fig. 15: plane of temperatures over the cryptographic coprocessor device.

conclusions

this article speaks about the achievements of the italian laboratory of celestica in the thermal analysis of a challenging product: a cryptographic coprocessor device.

in particular, in this paper three different models have been described.

our laboratory is able to realise cryptographic coprocessor devices that accomplish the maximum level of security prescribed for transmitting and exchanging information: the fips level 4. due to this required level of protection (both from the mechanical and the electronic point of view), inside these embedded cards a high temperature is generally reached with obvious problems for the reliability of some devices.

for this reason, an accurate thermal analysis is necessary.

flotherm has been used with success: main problems (at the beginning some simulations didn't converge) have been solved by using a subdomain inside the embedded card.

the comparison between numerical and experimental results has validated the use of a cfd based software like flotherm for such a kind of study.

acknowledgements

the author would like to thank all the people who supported his work and those ones who gave him the possibility to grow his experience in thermal simulations.

references

1) flotherm version 2.0 introductory course, flomerics limited, hampton court, surrey, united kingdom, 1997.
2) flotherm version 2.0 introductory training course, flomerics limited, hampton court, surrey, united kingdom, 1998.
3) flotherm version 2.1 release notes, flomerics limited, hampton court, surrey, united kingdom, 1998.
4) flotherm version 3.1 release notes, flomerics limited, hampton court, surrey, united kingdom, 2000
5) modelling ic packages using flotherm & flopack, lecture notes, flomerics limited, hampton court, surrey, united kingdom, 1999.
6) modelling ic packages using flotherm & flopack, tutorials, flomerics limited, hampton court, surrey, united kingdom, 1999.
7) thermal measurements in electronics cooling, kaveh azar, crc press llc, boca raton, florida, 1997.
8) surface-mount plastic packages - an assessment of their thermal performance, mali mahalingam, member ieee, ieee ieee transactions on components, hybrids, and manufacturing technology, vol. 12, no. 4, 1989.
9) methodology for thermal evaluation of multichip modules, balwant s., bruce m. guenin, ronald j. molnar, ieee transactions on components, packaging and manufacturing technology, part a, vol. 18, no. 4, december 1995.
10) plastic package thermal performance, r. tiziani, v. motta, sgs - thomson microelectronics package development department.
11) thermal analysis of a chip on board, de moerloose, j. temmerman, proceedings of the 1997 ieee 13th annual semiconductor thermal measurement & management symposium, december 1997.
12) compact models for accurate thermal characterization of electronic parts, h. vinke, clemence j. m. lasance, associate member, ieee, transaction on components, packaging and manufacturing technology - part a, vol. 20, no. 4, december 1997.
13) la gestione dei problemi termici, david tatchell, "elettronica integrata", magazine, february 1997.
14) investigation of thermal enhancements on flip chip plastic bga packages using cfd tool, lee, tien-yu tom, proceedings of asme, heat transfer division, 1999.

about celestica

with over 31,000 employees worldwide, celestica operates 36 manufacturing and design facilities in the united states, canada, mexico, the united kingdom, ireland, italy, the czech republic, thailand, hong kong, china, malaysia and brazil. celestica provides a broad range of services including design, prototyping, assembly, testing, product assurance, supply chain management, worldwide distribution and after-sales service. its customers include industry leading original equipment manufacturers (oems), primarily in the computer and communications sectors.

for further information on celestica, visit its website at http://www.celestica.com.