Case Study

Outliers

The Problem:

A customer has a sensor application requiring a specific red LED spectrum. They pay premium to bin LED die by dominant wavelength to +/- 1nm. Most colored LEDs are sold by dominant wavelength or by peak wavelength.Then after receiving the die, they are retested to bin by dominant wavelength to +/- 0.5nm. Even after finer grain binning, unacceptable variations in the red LEDs are still found for their application. They could not figure out a way to find the right die for their application.

C-Parameters Quantify the Shape of the SPD

With C-Parameters, each SPD of the LED die is broken down into two gaussians represented by six real numbers. Dominant wavelength is actually calculated by taking the spectrum and finding the CIExy (which is an uneven color space) then finding the closest point on the edge of the CIExy curve from the center. After so many transformations, color content from the physics based spectrum is lost. The customer eventually used centroid to determine suitable LEDs for their sensor application. Centroid again loses the color content, as centroid is only the wavelength at which half of the power underneath the spectral power distribution occurs. But with C-Parameters, the whole spectrum is considered.

Where Centroid Fails to Determine Outliers, C-Parameters Distinguish Outliers

C-Parameters is the solution to determine the variations that are unacceptable for the application. To the left, is boxplot of the binning according to centroid that is previously used, which gives no evidence of the variations that they actually see in their application. However, with the same die, sorted by C-Parameters and by just comparing CP_C1 among all of the die, the variations naturally start falling out, distinguishing the problem die.

Same Centroid, Different C-Parameters

The spectrums are seen to be obviously different, but they have the same centroid. It's already hard enough to find the right die in a lot of hundreds of die, C-Parameters make the difference.

Viewing the SPDs of these "outliers" makes it abundantly clear WHY these die behave differently when packaged.