关于MC仿真mismatch参数的问题咨询

jcxwzy

咨询一下，目前.18工艺库的各个corner中包含mismatch参数；

但是在仿真中发现，如果选择single run并不会调用这部分参数；

所以是不是只有跑MC才能反映电路的random mismatch？

跑完MC之后发现运放的红圈电流镜部分贡献较多的mismatch，这个要怎么调整才能减小mismatch？

传说中的干饭王

mark!

小林同学

   

ipmsn5 发表于 2024-10-6 04:32
Mismatch 分为 system offset和random offset 两种
第一种是你的电路设计的问题，这个时候DC仿真就可以看出 ...

怎么操作看其每个管子的贡献度的？这儿不亮

ipmsn5

Mismatch 分为 system offset和random offset 两种
第一种是你的电路设计的问题，这个时候DC仿真就可以看出来，
第二种是工艺的随机误差，需要通过尺寸优化来提高。也就是MC仿真的时候看看哪些器件贡献的误差，然后进行改进。

shuxinglong

   

1033297102 发表于 2024-7-4 15:37
你好，smic工艺仿真offset贡献度是错的该怎么解决呢

我也遇到了同样的问题，请问楼主解决了吗

1033297102

   

david_reg 发表于 2022-1-12 11:39
如果电路不包含开关电容，即只需要跑DC就能看offset的话，也可以试试spectre的“dcmatch”分析，通常比mc快 ...

你好，smic工艺仿真offset贡献度是错的该怎么解决呢

wpc_0310

你好，可以看一下仿mismatch的testbench是啥样的吗？

jcxwzy

   

david_reg 发表于 2022-1-13 10:56
from cadence support
Problem
How does dcmatch analysis differ from monte carlo mismatch analysis?

刚才试着跑了一下dcmatch，不过这个结果都是0；

论坛看了下是有些参数没有调用？

jcxwzy

   

david_reg 发表于 2022-1-13 10:56
from cadence support
Problem
How does dcmatch analysis differ from monte carlo mismatch analysis?

你好，从这个解释来看，dcmatch只run一次，而且它的参数变化范围也没有MC大；

所以dcmatch跑出来的结果应该会比MC要好；可以反映一定的趋势；

david_reg

   

jcxwzy 发表于 2022-1-12 16:01
好的，明白，加入chopper应该是比较好的选择；

MC也有mismatch contribution来看是哪个管子的贡献比较大 ...

from cadence support
Problem

How does dcmatch analysis differ from monte carlo mismatch analysis?

Solution

The dcmatch analysis can be considered as a short cut for monte carlo mismatch analysis in dc. In monte carlo mismatch analysis, we do multiple runs with random number on circuit or model parameters and then examine the output. In dcmatch analysis, we capture the randomness via the dcmatch model parameters such as

+ mvtwl   =  5.740000e-09 mvtwl2  =  1.503000e-12 mvt0    =  0.000000e+00
+ mbewl   =  8.022000e-09 mbe0    =  0.000000e+00

and just do a single run, and examine the output. So the accuracy of dcmatch analysis relies on how good these mismatch model parameters representing the randomness, and how good the underline model equations are (which are provided by a foundry).

The second difference is dcmatch assumes that the randomness is varied in a small fashion such that the variations will not change the dc operating point too much. In monte carlo mismatch, the random numbers can swing large. Whenever the random variations cause significant dc operating point shift, the variations are considered large. But for dcmatch, since it is based on "noise-like" analysis algorithm, if you want to model mismatch with large variations, then the algorithm is inherently not so accurate. So, the dcmatch result assumes small random variations. Some people call dcmatch a local mismatch analysis, instead of global mismatch analysis.