characterization, modeling, and design of esd protection circuits

4.3. Device Failure Results 127
already discussed in Section 1.1 as well as by Polgreen [8]. In the TLP tests, failure is
defined as the point at which device leakages exceeds 1µA, while in the thermal-box
model failure is defined as the onset of second breakdown. A certain current density is
needed to cause a device to enter second breakdown, but widespread damage does not
follow instantaneously in narrow devices because there is not enough total energy in the
TLP pulse, and consequently narrow structures must be stressed with higher pulses than
predicted before damage is severe enough to create microamp leakage. Of course, the
absolute current to failure and power to failure increase with device width, but note that as
the width increases beyond 50µm, the failure current per width levels off (Fig. 4.49b)
while the normalized failure power continues to decrease (Fig. 4.49a), indicating that the
device voltage at failure, Vf , decreases with width. The decrease in failure voltage with
width is explained by the fact that the snapback resistance, which is roughly inversely
proportional to the width (Fig. 4.46), decreases with width more rapidly than the failure
current increases with width. In Section 2.4 and Table 2.1, the width was predicted to have
no effect on Vf (Vt2 ), but in Section 2.4 it was assumed that the failure current scales
directly with width, which is not the actual case. It would be beneficial to test even wider
structures to determine if there is a point at which the normalized power to failure levels
off.
In Section 4.1.4, the 100µm-wide structure was used for calibration of thermal failure
because microamp leakage was almost always created the first time second breakdown
was captured on the oscilloscope and thus there was no ambiguity in defining the failure
level. However, as seen in Fig. 4.49b another advantage of using wide structures for
calibration is that the measured failure current is proportional to device width for wide
devices and therefore more amenable to 2D simulation. In contrast, according to the
thermal-box model the intrinsic error between predicted 2D and 3D failure power (or
failure current) is independent of device width (Fig. 3.33). Again, the conflicting results
are due to the different concepts of failure and underline the importance of consistently
defining failure in experiments and simulations.
Experimental and simulated failure power vs. contact-to-gate spacing for 50/0.75µm
structures subjected to 200ns TLP stressing are compared in Fig. 4.50. As just stated, the
experimental failure level is defined as the power needed to create microamp leakage, but
for 200ns pulses this level usually coincides with the power-to-second breakdown. In the
simulations failure was defined, as described in Section 4.1.4, either by the time at which