在现有模型库里进行修改可以用来做monte carlo分析吗？

fumes

比如把库里的Vth值删掉，在网表里重新定义一句.param Vth=aguass(0.7,0.1,1),这样做法对吗？
倒是可以仿，结果也符合预期，就是不知道这符合不符合实际情况，非要用工艺厂提供的
专用库吗？它的测试数据不也是给的这些值吗？就是改了一下语法结构而已

fhchen2002

Sometimes the VTH is not specified the way you expected.
(Most more sophiscated MOS models don't directly assign device threshold voltage to VTH.)
Then you loose the freedom of assigning VTH as Gaussian variable.
Under such circustances, you want to set another SPICE parameter (that is part of the equation that derives VTH) as Gaussian.  For instance, you may assign the parameter "delvto" as Gaussian, and still achieve the same goal (of varying device threshold voltage in your Monte Carlo simulations).

Not limited to VTH variation, you may also want to make beta (i.e., u * Cox * W / L) of the MOS transistor a Gaussian.  As one of the practical approaches, beta variation can be 100% assigned to W variation, since in ordinary analog design, W usually is much larger than W_minimum, and W > L (actually not true for low-power analog design).

fumes

回复 2# fhchen2002


    明白了，谢谢

fumes

Sometimes the VTH is not specified the way you expected.
(Most more sophiscated MOS models don't di ...
fhchen2002 发表于 2010-12-29 20:07

                               
登录/注册后可看大图

    在我这个库里，只有delvt0p是可变的，delvt0n这个参数是个常数，在各个工艺角下都不变，但
阈值电压在各个工艺角下还是变的，而且也没有beta这个参数，那我要控制Vth的变化也只能改变
Vth0n和Vth0p了

fhchen2002

Beta (other people might call it K) is the product of mobility (u), oxide thickness (tox), W_eff and divided by L_eff.  

   (K = u * Cox * W_eff / L_eff)

So you want to make beta a random variable (say, with standard deviation of 2%)  is equivalent to setting W a random variable (and with standard deviation of 2%).

HNwh2011

YESYES

jeffej

hard to know !!!

ralphtwtw

明白, 谢谢~!!