We have shown that CDM waveform measurement using 1 GHz/5 Gigasample/sec oscilloscopes is the largest source of system-to-system variation. This is consistent with findings by leading semiconductor manufacturers. The issue is temporal resolution, or the sampling rate of the oscilloscope.  The analogous parameter for visual phenomena is spatial resolution, as shown in Fig. 3.

Fig. 3.     Increasing the Spatial Resolution

The image is not clearly recognizable unless there is adequate spatial resolution. The same is true for the CDM waveform. The waveform cannot be measured correctly unless the oscilloscope has a high enough sampling frequency.

The current JEDEC specification specifies a 1 GHz/ 5 Gigasamples/sec oscilloscope to capture the CDM event. Digital oscilloscopes attempt to reconstruct waveforms using samples from an analog-to-digital converter and a sin(x)/x interpolation algorithm. The sampling rate imposes limitations on the ability of the instrument to render an accurate waveform. The Nyquist criteria state that a signal has to be sampled at least twice as fast as its highest frequency component in order to correctly reconstruct it. This is based on the following assumptions:

·     The waveform can be represented as a superposition of different frequency sine waves.

·     An infinite (or very large) number of samples is available.

·     The waveform repeats indefinitely.

Unfortunately, the CDM event does not fit these assumptions. First, the CDM waveform has higher frequency components, primarily due to extra components added to the JEDEC test head to shape the waveform. Next, the oscilloscope sampling rate does not allow for a large number of samples to be taken. Finally, the CDM event is a one-shot event; it does not repeat indefinitely.

Fig. 4.     Current (A) vs Time (ns), 1 GHz oscilloscope, interpolated and actual data points

When a 1 GHz oscilloscope is used to measure the CDM current versus time waveform, the display looks like the first picture in Fig. 4. Because the resulting waveform seems to be similar to the waveform shown in the CDM specification (Fig. 2), the waveform appears to be correct. If the interpolated data points are removed, however, and the actual samples are displayed, the waveform looks like the second display in Fig. 4 -- a considerably different picture. Two to three data points define the rise time. Three to four data points define the pulse width. The peak current will not always be captured correctly.

Although it is possible to model CDM test head characteristics and apply calculations to deconvolve the 1 GHz CDM waveform, the better approach is to use a higher bandwidth oscilloscope.