Aug_EDFA_Digital

edfas.org 29 ELECTRONIC DEV ICE FA I LURE ANALYSIS | VOLUME 23 NO . 3 near the fiber input/output, normally used in laser applica- tions. This reduces the number of photons that can actu- ally reach the sensor and is one of the major bottlenecks to keeping the possibility of switching fibers, depending on the application (e.g., photon counting, OBIC/OBIRCH). Once again, this set-up is only for developing proof of concept. APPLICATION TO INVERTER CIRCUIT IN AN FPGA As a proof of concept, this process is applied to signals acquired froma Xilinx Kintex 7 FPGA [17] mounted on a Skoll Kintex 7 board fromNumato Lab. [18] The DUT is packaged in lid-less flip-chip BGA, requiring little sample prepara- tion for backside analysis. The substratewas only thinned down by several tens of microns to remove laser marking on the die. For this test, the circuitwas programmed for an inverter function. The value received fromone of its IOswas invert- ed and routed back to another IO. The input was biased with a square signal (1.2 V, 5 MHz, and 50% duty cycle). A 1 MHz periodic pulse of 100 ns synchronous with the inverter input signal was used for referencing of the detection time of photons. An area of interest (AOI) gen- erating photons was first identified using InGaAs camera Fig. 2 Summary of themachine learning based approach for photon emission image interpolation. (a) 3D scatter plot of a single cluster of photons. (b) Histogram count. (c) Top view of the histogram count. (d) Estimated probability density function. (e) Top view of the probability density function. Fig. 3 Schematic of the measurement setup. Fig. 4 Identification of the area of interest for further investigation. (a) Emission image overlay onto the device micrograph. Both images were acquired using the InGaAs camera and the 50X objective lens (NA 0.4). The red square indicates the area chosen for TRE investigation. (b) Zoom-inof the emission image. (a) (b) (a) (b) (c) (d) (e)