Browse Prior Art Database

Algorithms for MOSFET Compact Model Quality Checks

IP.com Disclosure Number: IPCOM000148210D
Original Publication Date: 2007-Mar-29
Included in the Prior Art Database: 2007-Mar-29
Document File: 2 page(s) / 32K

IBM

Abstract

This invention relates to the field of compact modeling of semiconductor devices and to the problem of quality checking a model for physical behavior over the complete range of allowable device sizes and bias conditions. Non-physical behavior of a model can cause convergence problems during circuit simulations and yield inaccurate simulation results. Our invention is a set of unique algorithms for measuring MOSFET compact model quality with respect to negative gm or gmb as well as the shape of the Vt vs L curve.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 55% of the total text.

Page 1 of 2

Algorithms for MOSFET Compact Model Quality Checks

Algorithm for gm and gmb quality checks:
Run a series of MOSFET compact model simulations over the conditions to be tested, and measure

1.

2.

3.

4.

Algorithm for Vt vs L curve shape quality checks:
Identify all significant local minima and maxima.

1.

2.

3.

4.

5.

gm and gmb.
Typically the conditions to be tested cover a 6-dimensional space in Device Length, Device

a.

Width, Operating Temperature, Vds, Vbs, and Vgs. The space is tested by running a series of nested sweeps for each dimension. A sweep may typically have anywhere between 2 to 50 values, but could have significantly more if required.
gm and gmb are determined by direct output from the model (if available) or by numerically

calculating the derivatives by simulating a second point with a small deviation in Vgs (for gm calculation) or Vbs (for gmb calculation).

Calculate the values of the ratios of gm/I and gmb/I.

b.

Compare these ratios to a minimum criteria value.

a.

b.

For a minimum criteria = 0, the test becomes equivalent to checking for negative gm or gmb.

For a minimum criteria < 0, some values of negative gm or gmb will pass the test provided their

magnitude is small relative to the current (I) and the selected criteria. This allows more freedom for optimizing the model fit to data while ensuring that any negative derivatives will be too small to cause convergence problems in a jacobian based algorithm.

Flag an error if the value of gm/I or gmb/I is less than the minimum criteria.

a.

Compare the value of each point to the value of the preceding point and determine if the slope of

the curve is increasing, decreasing, or flat. If the magnitude of the change in value between the points is below some chosen threshold value, the curve is considered flat. A threshold value > 0 is chosen to eliminate numerical noise that can occur in the calculations and to eliminate very slow changes in Vt vs. L (which are considered inconsequential).

If the slope is either increasing or decreasing, compare the direction of the slope to the direction

of the most recent previous slope that was also either increasing or decreasing.
If the change in slopes is from increasing to decreasing, there is a local maxima at this

I.

II.

b.

location in the curve.

If the change...