edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 25 NO. 4 32 a path that eventually leads to collection by the detector, thereby altering the VC signal. 2) An electron beam parked at a certain probe location on a DUT generally causes a contamination buildup at the probe location. Often that contamination is hydrocarbon-based and leads to a darkening of the sample and a reduced VC signal. Sometimes, however, a spot of contamination can build up, causing image and VC brightening, perhaps because when a contamination spot attains some height, secondary electrons generated within the contamination have an easy escape path in many directions. 3) Grain, material, or edge structures that produce contrast in SEM images can partially contribute to the VC signature. In the presence of image contrast at a probe location, beam drift, or slight electrostatic-induced beam shifts due to changing voltages at or near the probe location, can intermix VC signals with standard image contrast, leading to a perceived change in the measured VC signal with time. These three examples, as well as many other effects unrelated to voltage contrast, may masquerade within a VC signal, sometimes increasing, and sometimes decreasing, the measured VC signal. TYPICAL RESULTS Voltage contrast has been used for many applications in IC debug. In the 1990s, Schlumberger produced an IDS10000 series of e-beam probers that could measure electrical waveforms on then-current technology.[4] Since that time, however, IC devices have shrunk two orders of magnitude, and device operating voltages dropped 5x. As a result, any modern e-beam probing solution should have resolution requirements closer to 10 nm than to the micron-sized features of the 1990s. This has created challenges in instrument design. For example, reduced beam currents are required to support 100x better imaging, reduced DUT voltage swings result in lower VC signals, and acceptable levels of image shift when chopping the e-beam also drop in proportion.[3] These constraints presently limit waveform measurements to frequencies up to 2 GHz, however the work done on the IDS 10000 in the 1990s showed that this technique could support waveform measurements up to 6 GHz. This clearly demonstrates longterm development opportunities for the e-beam roadmap, and we can expect more publications on these developments in the future. Various papers describe developments in electron-beam probing of active modern ICs.[5-9] In addition to describing waveform capture, which is the subject of this paper, several of these references describe static imaging applications on active devices, such as logic-state mapping whereby high and low logic levels in IC circuitry appear at different contrast levels in SEM images.[3,5-7,9] Yet more uses for SEM technology with ICs abound, such as techniques that use passive voltage contrast on inactive devices to locate various design faults.[10] A classic case of passive VC involves the identification of circuit opens, where an impinging e-beam will charge such an open circuit positively (or negatively) depending on whether the electron yield is greater (or less) than unity, leading to a dark (or bright) spot at such locations in the SEM image. VOLTAGE CONTRAST SENSITIVITY The remaining question concerns how accurately a DUT voltage can be measured using this technique. To this end, consider the signal detection process. A beam of electrons strikes the DUT, causing backscattered and secondary electrons to be emitted from the DUT. A fraction of these electrons is captured by the SEM column, wherein a mirror mechanism routes only mid-energy secondary electrons into the detector. The detector comprises a scintillator maintained at +10 kV potential such that hundreds of photons are created as each now-energized secondary electron strikes it. These photons are routed by a light pipe into a photomultiplier tube (PMT), and conditioning electronics thereafter convert the PMT output current into a video signal. The dominant noise source in the PMT output signal is limited by the shot noise of primary beam current, though this shot noise is enhanced by a noise factor having to do with the many stochastic processes that occur between when a primary electron first strikes the DUT and when the video signal emerges from the detection system. Rather than attempting to calculate this noise factor from its many parts, it is easier to empirically determine the sensitivity of any SEM to voltage contrast and adjust system parameters to maximize this sensitivity. The result, as detailed in Thong, relates the expected RMS noise in any SEM measurement (i.e., video signal output) to the number of primary electrons used to make the measurement, according to: (Eq 1) where C is the empirically determined spectrometer constant for the SEM, e is the electron charge, and N is the number of primary electrons used in each sample measurement.[3] Thong quotes a typical spectrometer constant of C = 5 × 10-9 V√(As), whereas in 2019 the authors reported values for IC metal lines and silicon of CMetal = 4 × 10-9 V√(A s) and C Si = 9 × 10-9 V√(As).[8] As detailed earlier in this paper, many environmental variables may change results for any given measurement, but our results of 2019 remain valid today. VOLTAGE CONTRAST WITHIN ELECTRON MICROSCOPY (continued from page 29)
RkJQdWJsaXNoZXIy MTMyMzg5NA==