Interview: advantages of multiple imaging for PERC inspection

What is the locations of the manufacturers for which this solution has been provided?

Eric Rüland: The first systems were installed in Taiwan, but during 2015 and recently further installations were made in Korea, Southeast Asia and the U.S.

You mention pseudo defects in your product literature, does this mean that using a single light wavelength for inspection can lead to cells being discarded or incorrectly classed during production? How commonly can this occur?

Eric Rüland: Pseudo defects or "overkill" always happens, but it is a clear matter of system quality if it limits the performance. In real evaluations ISRA Vision / GP Solar achieved << 0,05% overkill for real production. We already get calls of some customers, if more than five cells out of more than 100,000 per day are sorted incorrectly.

In the case we are talking about with PERC rear sides, it is more a question of, "is my system capable to detect the print even if the optical appearance of my samples varies slightly with time?" If it cannot detect clearly where there is print paste on the wafer, it will have difficulties to decide whether the print quality is good or not.

Standard inspection systems use standard illumination and are limited on single image inspection. This can lead to much higher pseudo error rates compared to the ISRA Vision / GP Solar’s lightdome, multi-view technology because illumination homogeneity and isotropy is very high.

I can provide some more explanation on this: For example, overkill will always happen if you inspect samples with features very close to the good/bad limit ("tolerance bound"), with very close meaning the scattering range or repeatability of a measurement system. It is easy to see if a sample is exactly on the tolerance limit, for example the edge length 157 mm with 156 +/-1 being your acceptable range, measuring the sample 100 times will, theoretically, give you 50% good and 50% bad results.

Assuming the sample is very close to the tolerance limit, such as 156.99, then doing a "repeatability test", in nearly 50% of the measurements, the sample will still be classified as bad. If the sample is 156.94, then nearly all measurements will give a good result. The better the repeatability, the lower the number of samples in a given quality distribution which are close to the tolerance limits. If you do the same test with a system with +/- 100 µm accuracy, then you get possible overkill even for samples which are at 156.9 mm.

How many units using this technique have been deployed by ISRA Vision?

Eric Rüland: ISRA Vision / GP Solar shipped around 150 systems already for innovative process control which means more than 2.5 GW of PERC production capacity worldwide.

Can the new technology be retrofitted to existing technology or must it be installed as a new standalone tool?

Eric Rüland: This depends on the hardware installation. But even our customers of our first version systems, delivered more than eight years ago, are able to upgrade to latest technology.