edfas.org 33 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 25 NO. 4 To provide one example, consider a SEM image taken with a beam current of 400 pA and a dwell time per pixel of 1 microsecond. We expect the noise within each pixel of the image to be equivalent to 140 mVRMS DUT voltage noise when imaging metal lines, and 320 mVRMS DUT voltage noise when imaging silicon. A user trying to measure 1 V signal variations on a DUT between two SEM images could be expected to see a VC signature with an SNR of 5 for metal and 2 for silicon under these conditions. VOLTAGE CONTRAST APPLICATIONS Figure 2 shows the voltage-contrast signature of a 5 MHz 0.8-VPP waveform. A continuous DC beam was parked on the signal trace and an oscilloscope was used to average the output signal from the secondary-electron detection system one million times. A 100-pA beam current could be expected to give an SNR over 50 in this case, and the SNR is such that some of the noise observed in the signal may be actual noise on the data line. Probing with continuous DC electron beams is very useful for observing DUT frequencies below the bandwidth of the detector system, which historically has been limited by the scintillator decay time of the secondary electron detector; typical secondary detectors have bandwidths of 5 to 10 MHz, though work is underway to use a faster scintillator. When probing signals above the secondary detector bandwidth, pulsed electron beams must be used, and the waveform reconstructed from equivalent-time sam- pling.[3,8] The bandwidth of such pulsed systems is no longer determined by the detector but rather by the minimum achievable pulse duration. Our current systems support 2 GHz waveform analysis. Figure 3 shows such an equivalenttime sampled 666 MHz waveform. Acquiring such waveforms requires anywhere from seconds to many minutes, depending on the test loop rate, since only one sample is acquired per test loop in this mode. Figure 4 shows the ability to perform spectral analysis of signals. For frequencydomain analysis, a periodic stream of electron pulses samples the DUT potential, and Fourier techniques return the presence, or absence, of signal (e.g., a clock signal) over a region. Additionally, if the tester and measurement systems are phase locked together, such Fourier techniques can return amplitude as well as phase information, as demonstrated in Fig. 4. The SNR is not extremely high in this example, though it could be made better with a longer image acquisition time, but it is sufficient to observe transistor switching events on an IC. Fig. 3 Waveform at 666 MHz taken in pulsed equivalent-time sampling mode. E-beam pulses of approximately 150 ps FWHM are directed at the DUT, once per tester loop, and several hundred such pulses are summed together for each equivalent time sampling of the waveform in order to achieve acceptable SNR in the resulting waveform. After a sufficient number of pulses have been averaged, the process repeats at the next equivalent sample time, and in turn, this repeats until the entire waveform is captured. Such waveform captures require acquisition times on the order of several minutes, depending on the test loop repetition rate. Fig. 2 Waveform taken from a test chip operating at 5 MHz, the signal having 800 mVPP amplitude. A continuous electron beam of 100 pA was directed at the sample, and an oscilloscope averaged one million samples taken directly from the detector. Such waveform captures require acquisition times on the order of a second.
RkJQdWJsaXNoZXIy MTMyMzg5NA==