Cut photolithography dose error from 15% to 2%

What I did

Inherited a working but inaccurate system, in which a microcontroller set each wafer's exposure dose from photodiode readings. Used oscilloscope captures and MATLAB analysis as the source of truth, debugged the dosing software, and redesigned the measurement circuits with new components for more accurate readings.

Result

Dose error fell from about 15% to 2%. Presented as posters at AAAS and NCUR 2018.