The following is a commentary on the article
01/01/2000
The following is a commentary on the article "Moiré Interferometer: Assist in Electronic Packaging Development," which was published in the September 1999 issue of Advanced Packaging (pp. 44-50).
As the elder statesman in the field of moiré interferometry, I am compelled to help guide its practice. The authors and I share the same objective, and I applaud their efforts to popularize and propagate moiré interferometry.
The authors discuss the need for experimental analysis, especially for the mechanical deformation induced by changes of temperature. They introduce the moiré interferometer that was manufactured and marketed by IBM until the mid-1990s - it was called PEMI. The authors engage in the manufacture and marketing of another moiré interferometer called 4M, which they say is the same as the IBM PEMI except for one enhancement. The enhancement is a phase shifting system used for its potential to increase measurement sensitivity and enable automated analysis of the moiré data. Automatic multiplication and automatic numbering of moiré fringes is claimed without reservation. The automated analysis is the main focus of my review.
First, however, some background is useful for perspective. The PEMI design was adopted by IBM, but it originated at Virginia Tech and it was put into practice by my students.1 Fringe shifting by moving the standard grating (4M terminology) is described in our comprehensive textbook, and the 45° direction of movement was introduced at Virginia Tech.1 No patent protections were sought because we preferred wide access and propagation of the technology, which is continuing today. IBM spun off the PEMI business to its group manager, who established Photomechanics Inc.2 They produce a moiré interferometer called PEMI II, which is an entirely different design; it incorporates fringe shifting and various additional features.
Returning now to the issue of data reduction, please refer to Figure 3 of the September article under consideration. In images 3a, b, c and d, the fringe contours near the interface between the die and the die attach are ambiguous. Near the corner of the die, within the box drawn in Figure 3a, the contours have the form sketched in Figure 1a of this review. I assigned fringe orders progressively1 in regions unaffected by the ambiguities, assuming thermal expansion. A dilemma is observed.
Of course, a continuous fringe cannot have two different fringe orders; the displacement denoted by a fringe is always unique. Figure 1b illustrates the correct interpretation. It shows the continuity across the interface zone between segments of fringes of the same fringe order, although the width of the interface zone is exaggerated for clarity. In the sample, the fringes are very closely spaced in the narrow interface zone, indicating an extremely high level of shear strains.
The correct displacement contours are obvious to the discerning person. A reasonable estimate of the shear displacement gradient could be extracted from Figure 3c.
The region surrounding the interface zone under consideration is shown in Figures 3a, b, c and d, but that critical region is not shown in the results presented in Figures 3e and 3f. It appears that Figures 3e and 3f show enlarged images that represent a smaller portion of the sample, such that Figure 3e shows phase contours in a region above the interface zone, and Figure 3f shows displacement contours in a region barely extending to the interface zone.
Why were the interface results deleted? What went wrong? The problem appears to lie in the blind automation process. In my considered opinion, we must give up the dream of push-button analysis. A period of training and study is required. Engineering judgement is needed, and logical rules must be followed. The rules pertaining to the case at hand include the following:
Automated analyses should be restricted to regions of singular material properties; they should not cross boundaries between materials of different properties. [Why? Stresses (forces) are continuous functions across boundaries, so strains and displacement derivatives must be discontinuous wherever abrupt changes of properties (e.g., the elastic modulus) occur.] Otherwise, the filtering or smoothing performed by phase shifting algorithms misrepresent the data near discontinuities.
Accurate analysis by phase shifting algorithms requires definitive fringe patterns - definitive experimental data. Consequently, numerous pixels per fringe are required to record the data. If the fringes are too closely spaced to be recorded with numerous pixels per fringe, the data will be flawed. Thus, fringe shifting techniques are most effective for analysis of sparse fringe patterns, where a limited number of widely spaced fringes fill the field. It is under such conditions that the quoted increase of sensitivity can be realistic. Otherwise, any gain in sensitivity is more than offset by a loss of accuracy.
Electronic packaging can benefit enormously from moiré interferometry. In addition to effective instrumentation, a company must give time to committed personnel for meaningful study and preparation. Here, a key obstacle was fringe numbering. A key lesson seems to be Blind Analysis is Blind!
With due respect to the authors for their numerous positive contributions, I trust they will take pause to reconsider their philosophy on automatic data analysis and propagate effective techniques of interpretation.
Sincerely,
Daniel Post, professor emeritus
Virginia Polytechnic Institute and State University (Virginia Tech)