12.12.2012 Views

characterization, modeling, and design of esd protection circuits

characterization, modeling, and design of esd protection circuits

characterization, modeling, and design of esd protection circuits

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

4.1. Calibration Procedure 105<br />

past VCC <strong>and</strong> the drain voltage stepped at values around VCC , so prior calibration <strong>of</strong> the<br />

drain current implies that the substrate characteristic should be fit only by adjusting the<br />

∞<br />

αn <strong>and</strong> λn coefficients (Eq. (3.28)). Similarly, the breakdown voltage, BVDSS , in Fig.<br />

4.40e is dependent upon the drain-substrate junction pr<strong>of</strong>ile, but calibration <strong>of</strong> BVDSS should concentrate on adjusting the ionization coefficients because the results <strong>of</strong> the drain<br />

<strong>and</strong> gate calibrations suggest that the junction model is already accurate. Adjusting the<br />

impact-ionization coefficients should not affect the drain, gate, <strong>and</strong> subthreshold<br />

characteristics because relatively high electric fields are not involved. However,<br />

introducing the II model to the drain-characteristic simulations does increase the drain<br />

current in the 0.5µm-gate structure up to 10% for Vds = 6V (well above VCC ) because the<br />

electric field is fairly high <strong>and</strong> the drain sinks most <strong>of</strong> the electrons generated by impact<br />

ionization.<br />

In MEDICI the default II coefficients are based on measurements <strong>of</strong> impact ionization in<br />

bulk silicon [63], but as discussed in Section 3.1 impact-ionization rates in MOSFETs are<br />

lower than in bulk silicon because II generation occurs near the surface, where the mean<br />

free path is lower, i.e., where the critical electric field <strong>of</strong> Eq. (3.27) is higher. Therefore,<br />

the final fitting values <strong>of</strong> the electron <strong>and</strong> hole mean free paths, λn <strong>and</strong> λp , are expected to<br />

be lower than the MEDICI defaults. In keeping with the philosophy <strong>of</strong> manipulating as<br />

few model coefficients as possible, only λn <strong>and</strong> λp were adjusted to calibrate the substrate<br />

∞ ∞<br />

<strong>and</strong> breakdown curves while the pre-exponential coefficients, αn <strong>and</strong> αp , were held<br />

constant. This approach works for calibration <strong>of</strong> the st<strong>and</strong>ard MOSFET characteristics,<br />

but it has a significant consequence on the snapback simulations that will be discussed in<br />

the next subsection.<br />

Calibration <strong>of</strong> the substrate curves was performed before that <strong>of</strong> the breakdown curves<br />

because the substrate current depends only on the electron II coefficients while BVDSS depends upon the hole coefficients as well as the electron coefficients. In Fig. 4.40d, Ib consists <strong>of</strong> holes diffusing from the high-field region under the drain side <strong>of</strong> the gate<br />

where they are generated by impact ionization (recall that Vds is around 3.3V during the<br />

stress, so the electric field is relatively high in this area). Since the device current consists<br />

almost entirely <strong>of</strong> electrons, only the electron II coefficients affect the level <strong>of</strong> substrate<br />

current. An explanation <strong>of</strong> the shape <strong>of</strong> the Ib-Vgs characteristic is given in [42]. Basically,<br />

the initial increase <strong>of</strong> Ib with Vgs is due to the deepening inversion layer which increases<br />

the drain current <strong>and</strong> proportionately increases Ib . At a critical value <strong>of</strong> Vgs , however, the

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!