Discharge capacitor simulation using Spectre. Discrepancy with silicon results.

[palmeiras]

Hello guys,

A capacitor composed by Native MOSFET and MiM was fabricated. The capacitor is isolated by a switch that is OFF, therefore, there is only leakage across the terminal of this switch. The discharge time is much lower than the transient simulation estimation using Spectre when using gmin = 10^-15. Ex: Simulation results: 1000 seconds, while silicon results 100 seconds.
For gmin = 10^-12: the discharge time is very small (10 seconds), while silicon results 100 seconds.

What is the reason of such discrepancy between the silicon and simulation results? What I can do in order to have a proper estimation​?

Thank you very much,

 

[erikl]

Probably either the Native MOSFET's or the switch's model (or both) aren't correct re. the parameters which determine the leakage value.

Anyway, for such analysis you should always use the lowest possible value for gmin (10^-15).

 

[palmeiras]

Is this discrepancy known? I would expected a smaller difference. Using gmin 10-15, the discharge takes more than 1000 seconds. I was not expecting a error factor more than 50.

 

[erikl]

Did you try worst corner models?

 

[palmeiras]

Hi erikl, thanks for your message. Yes, I did corner analysis.
I was wondering if SPECTRE can support very long simulation times (e.g. 1000 seconds..). I will try Ultrasim..

 

[erikl]

Did you already try the GEAR integration method? S. here: View attachment Integration_methods_1.pdf

 

[palmeiras]

yes.... none change. Thanks, erikl.