edfas.org 23 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 25 NO. 4 SPM process with further control over the error accuracy. Finally, the logic diagram compares the incremental total depth reached and in case of reaching the user-defined tomogram dimension it ends the analysis. In this way, the user obtains a data-set ready for the 3D interpolation and all the readout scans used for the height control, in case they might be needed for further analysis. Figure 3a shows an example comparing two probes with different spring constants, using the RPM-3D to keep constant the RR at ca. 4 nm/scan for multiple scans. Note, both probes obtain a linear material erosion at a controlled rate irrespectively from the probe’s spring constant. This is enabled by the algorithm in Fig. 2b that is continuously adjusting the load force on the probe during the acquisition. To put this into perspective, compare the results of the same experiments with and without the use of the adjustable RR script, for the same structure under test, shown in the inset. In the plot of removal, Fig. 3b shows the details of RR and force applied during a 65 nm-deep scan (blue). It is visible how the RR can be maintained within the desired range with continuous adjustment on the load force that are applied by the RPM-3D. Without the adjustable RR script, a large deviation from a linear RR is obtained (red plot in Fig. 3b). The top inset shows the continuous adjustments introduced by the script to the piezo signal (i.e., load force) as the scalpel operation descends deeper into a sample with a controlled RR of 4 nm/scan. Note that during the first 3-5 scans the algorithm makes large adjustments as it tried to achieve the targeted RR, and later it focuses on small compensations of the load force. With this method, it is now possible to calibrate each scalpel scan to remove the desired amount of material, because of the “live” monitoring of the removal rate as demonstrated in Fig. 3, thus enabling stable removal control over prolonged dataset acquisition. The latter is shown in Fig. 4a for different values of RR obtained for an SiO2 film. Here, the RR control algorithm is used to maintain 3, 4, and 5 nm respectively for 10 scans in various samples, as reported in the x axis of Fig. 4a. The results indicate how the system can maintain a reasonable level of RR with an error of ca. 1 nm for multiple types of probes and samples including blanket films, simple metalinsulator-metal (MIM) films, and vertically poly-Si channels encapsulated into SiO2 (3D-NAND). Beyond controlling with good accuracy, the user-defined surface erosion speed, the “live” monitoring of the material RR adds an interesting new possibility to the RPM-3D now capable of detecting strong variation in the material hardness that might be induced by the presence of a physical interface within the stack. An example is reported in Fig. 4b, which shows the removal of a MIM stack, at a user-defined RR of ca. 6 nm/scan. As shown in the scanning electron microscopy image in the inset, the MIM device has a bottom interface buried about 60 nm from the top surface. Using the rate of variation of the material erosion it is possible to detect a spike in correspondence of the interface at ca. 50-60 nm from the top surface. In other words, the change in material hardness between the two layers is reflected in the sudden change in RR, that is continuously monitored by the RPM-3D offering a simple way to determine the position of such interface within the stack. This adds structural information in depth for the sample under Fig. 4 (a) RR control targeting 3, 4, and 5 nm of material erosion per scan. The plot shows a comparison be- tween different samples and multiple spring constants. Note, each datapoint represents ten scans obtained on the sample described in the x-axis and a new probe is used for every datapoint. (b) Interface detection capability offered by the ‘live’ monitoring of RR during the tomographic dataset acquisition. Note, the spike in RR signal detected in correspondence of the buried oxide-metal interface shown in the inset. (a) (b)
RkJQdWJsaXNoZXIy MTMyMzg5NA==