edfas.org 37 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 26 NO. 2 that reveals areas of corrupted logic built from the processed dLVP data. SIL images could be color coded to form composite image overlays showing where logical anomalies exist, or where unique digital waveform shapes or signatures are or are not present (waveform imaging), similar to laser voltage trace but not confined by the hardware capabilities from using a scope in fast acquisition mode.[8] A memorable question that came up after first introducing dLVP at ISTFA 2022 was: “How does this technique differ from laser assisted device alteration (LADA)?” The simple answer is that the techniques are very different but complementary. LADA reveals the locations of devices sensitive to a few picoseconds of laser induced timing change that modulate the test result, whereas dLVP gathers optical based pass versus fail internal waveforms to look for distinctions. Could dLVP resolve differences along LADA paths? Yes and no. For purely timing driven failures it will depend on the temporal resolution limits of the available hardware and system jitter. For other types of level sensitive failures it will depend on the voltage resolution of the waveform. The author believes that the complementary nature of dLVP and LADA will enable follow-on combination techniques that give breakthroughs for analyzing these soft failures, where LADA is used to reveal the critical path feeding into a capture flop, and dLVP indicates where there’s logical corruption inside and propagated downstream from the same flop. In closing, and upon reflection, the biggest limitation to dLVP as it was introduced in 2022 is that it was only applied to soft failure modes. However, in the short time since its introduction, additional efforts to improve and develop the technique revealed the power of data processing, sparking ideas applicable to working with hard failures. Looking to the field of astronomy offers inspiration and possible methods that may apply to FA and dLVP as well. Those familiar with the James Webb Space Telescope have witnessed some of the best images ever recorded. Many of which were achieved using a technique known as pixel dithering,[9] where deconvolution techniques are used to remove unwanted noise and other non-ideal detector related issues. Similar schemes Fig. 7 5 nm FinFET example of a dLVP waveform set showing clock, pass, fail, and the differential in yellow. LVP data (via averaging). The major downside is that this confines the waveform content to only 1000 points and limits available negative delay to ~2 times the waveform span. In practice, post triggering with a 10 ps sampling resolution (to look for a glitch, for example), limits the maximum available negative delay from the trigger position to only negative 20 ns. The workaround is to switch the scope from fast acquisition mode to normal mode, but then trigger collection efficiency plummets from roughly 76% down to < 1% (for a typical 20 μs test loop). Therefore, even when using the most expensive oscilloscopes, the tradeoff needed to achieve adequate negative delay while sampling at high temporal resolution is that it takes 76 times longer to average a typical waveform. Fortunately, a new 32 GS/s digitizer from Guzik exists that provides a solution to the negative delay problem and adds a new feature called “classifying marker,” making it ideal for ultra-high resolution dLVP. At least two vendors now offer the digitizer as a detector option for their latest probe systems. Having such a fast digitizer opens exciting possibilities toward generating real time waveforms, achieved through a combination of oversampling and data processing techniques such as analog boxcar averaging.[7] Rapid formation of useable waveform data opens the door to new image pixel mapping applications generated from such waveforms. From the click of a software button, users could choose to view the LADA speed path overlaid to a reflected laser image, or alternatively switch to viewing a pixel map
RkJQdWJsaXNoZXIy MTYyMzk3NQ==