Hi,

Negative capacitance means the MOSCAP is leaking charge instead of storing it.

The reason why it is happening only at HF is because the oxide has more traps, similar to Si3N4 If oxide was porous or had pin-holes, then it'd have been leaky at LF also.

You can try to anneal it in FGA (Forming gas Anneal) @ 4000C for 20 minutes.

Alternately, you can try improvising the process by which you grow or deposit the oxide. Grown oxide will be better than deposited one, but then you can only grow native oxide of the substrate, for example on Si you can grow SiO2. If the problem still persists, change the oxide if the oxide is not of particular interest of study.

Few ways to know if the oxide is good are:

1) Do Ellipsometery measurements to measure oxide thickness and refractive index and cross verify the thickness with the electrical thickness calculated from Cacc of LFCV curve. if thickness matches means oxide thickness is good, but then check the refractive index for that particular oxide thickness, it should match the ideal refractive index of that oxide, which you can search on internet. For example, Al2O3 has 6 oxidation states and the refractive index varies from n0 = 5 to 8 for different oxidation states.  For perfectly stoichiometric Al2O3 the refractive index will be 5.98, so the physical thickness(one measured from ellipsometery) obtained at this refractive index should match the one from electrical measurements. If it does not, then the oxide is not good.

2) Do XPS or XRD to know the quality of the oxide.

Yaksh Rawal, thanks for your answer. In the experiment I am trying to measure the trap density for this particular device. So, if the device has too many traps, which method will be particularly useful to measure the density of interface traps? Also, I am using Terman's method, but it needs me to go up to 1MHz which I am not able to. So, will using a frequency of 500KHz give me a proper approximation of the traps?

Hi,

Terman's method is based on the stretch-out of HFCV curve, so you'll first need to fix the process flow or the oxide layer in your MOS device to get a good HFCV curve.

Regarding 500MHz, the CV curve should look like HFCV curve, if it does you can use Terman's method. 

But the fact that you're getting negative capacitance means that interface traps are not dominant as compared to other oxide defects that makes you moscap resistive. To check this, you can take IV of your device, for good normal moscap the IV should look similar to diode IV, which means it should be non-linear. If you get a straight line kind of IV or a vary noisy IV with high current values means that you tried to make moscap but you ended up making a resistor. which can also mean that oxide is of poor quality. 

Alternately, you can also use DLTS for interface trap density.

There is this one book on MOS Cap, purely, 917 pages only on MOSCAP (This book is bible of MOSCAP, everything that can go wrong in a moscap is explained in detail in this book)

E. H. Nicollian and J. R. Brews, MOS Physiscs and Technology, John Wiley, New York (1982).