How to derive a hand calculating model

[youyang]

From a hspice (bism3/level49) lib, how everybody here to derive a hand calculating model, say, the large signal Ids model in sat/resistant area, and the small signal parameters such as Gm, Ro, etc.

The method I has been taking is to derive the indispensable parameters like U0, Cox, Vth, etc. directly from the Hspice lib file, then substitue them into the basic models based on the typical square-law.

Although this method is easily handled, it may produce large deviations when the second-order effects show up manifestly, especially with the ever-decreasing process-size.

My question is that in analog circuit design, given a process small enough, say, 0.18um, should the designer at his best abate the effects produced by short-channel-velocity-saturation by making Vov samller and L larger, so that the square-law may also valid to give us a good hand model. In other words, is the short-channel effect adverse to analog circuit and its design? Is it worth reducing its effects at the cost of a larger L?

If not, how to get a one-order model in the present of obvious short-channel effect?

any reply is appreciated! thanks a lot!
youyang

 

[pfd001]

hand calculation is always not accurate. Normall, we don't use the smallest channel length for analog design. The channel length is select according to your spec.

 

[youyang]

hand calculation is always not accurate. Normall, we don't use the smallest channel length for analog design. The channel length is select according to your spec.

hi,pfd001
thanks for your replying

According to you, the channel length is selected according to your spec, but my spec is not including this aspect. Can you give a detailed example how the channel length is selected, with your experience?

youyang

 

[kivvzhou]

usually L=3~5Lmin in our project

 

[mthomas]

Hi Youyang,

There is another solution to deal with hand calculation without using square law. This technique is called gm/Id. You can find information about it on:

**broken link removed**

I am just begining to learn it so I can't tell you much about that. What I can tell you:

* You can use table lookup model in your hand calculation, so you don't have to extract parameters from model.

* This technique is valid in all level of inversion of MOS.

* You still have to determine L with other considerations. For example parameters which influence the choice of L:
- For a current mirror, you have to lower the channel length modulation effect so L have to be large enough.
- Matching
- Output resistance (when L increase, output resistance increase)
- Cgd, for example for a current mirror if L is too high you had a pole-zero doublet in your bandwith which influence overall stablity
- I am forgetting lot of thing, to be completed....

 

[gte582w]

check out Sansen's new book which talks about three
region of MOS operations and the hand calculation models.
Also you can take a look at MOSVIEW for a truely unified
model valid for all regions.