IN-HOUSE COMPUTER PROGRAMS SOLVE MODELING PROBLEMS IN ELECTRONIC PACKAGING

FAYETTEVILLE, Ark. - Computer programs written "in-house" are more efficient than standard commercial programs for analyzing complex electronic packages, according to University of Arkansas researchers Panneer Selvam, Len Schaper and Aicha Elshabini.

Selvam, professor of civil engineering, and Elshabini, professor and head of the electrical engineering and computer science and computer systems engineering departments, conduct research at the U of A High Density Electronics Center (HiDEC). Schaper is director of HiDEC. Details of their study appear in the current issue of the International Journal of Microcircuits and Electronic Packaging.

"Because they are designed for specific problems, 'in-house’ programs can reduce the amount of data required, substantially reducing both storage space and computing time," said Selvam. "This allows the programs to run on much smaller computers, which reduces the computation costs."

Complex commercial modeling packages are like one-size-fits-all clothing in some ways. Because they must be constructed to accommodate the largest consumer, they usually don’t fit smaller customers very well. Since commercial programs must handle big problems, they require large amounts of data, storage space and computing time. In addition, modification can be difficult and produce unsatisfactory results.

Programs written in-house are more like tailored clothes. Because they deal with a specific problem, they can fit the problem precisely and produce a more elegant result. Since they must use only necessary data, they can be smaller and require less storage space and time to execute.

However, unlike tailored clothing, in-house programs are usually more cost-effective and efficient. For example, when Matt Gordon, professor of mechanical engineering, applied the commercial modeling program FLUENT to a problem, the solution for 1-dimensional flow required 10 days of run time on a Sparc20 workstation.

Selvam’s in-house program modeled the problem with much more complex and realistic 2-dimensional flow. Not only did it produce equally good results, it did so in less than 2 minutes on a desktop computer. The researchers estimate that the 2-dimensional problem would take more than a month of run time to solve using the commercial package on the Sparc20 workstation.

"This is particularly important in microelectronics packaging, where modeling is a vital tool for predicting electrical, thermal and mechanical performance of multichip modules," said Elshabini. "The trend is toward low cost, small size and high reliability. But the smaller the package size, the more severe are the thermal and mechanical issues that must be addressed."

Because microelectronics packaging is such a dynamic field, it requires another important feature of in-house programming - the ability to apply new computational techniques to a specific problem and to select the best technique for the problem. Just as computers change rapidly, the methods used to obtain solutions also change. However, because of the constraints in bringing a large modeling software package to market, commercial packages may use computational strategies that are many years old, making them less efficient and more cumbersome to use.

For example, finite-element and finite-difference modeling are good tools for modeling rectangular objects, such as microelectronic packages. An in-house program using either of these approaches is streamlined and efficient with low data and storage requirements. However, because they are designed to work on non-rectangular grids, commercial packages require much more data, storage and computing time to solve the same problem.

The move toward parallel computing highlights another important advantage of in-house programs. By dividing up a problem among many machines, parallel computing significantly increases the speed of solving a problem. Because of the nature of parallel computing, each program must be modified for the particular configuration of computers.

"To modify an in-house program to run in a parallel environment requires, maybe, adding one or two lines of code," explained Selvam. "But a commercial package cannot be modified to run in a parallel environment at all."