gm/id design technique

Status
Not open for further replies.
V

VASANSNS

Junior Member level 1
Joined
Apr 7, 2008
Messages
15
Helped
1
Reputation
2
Reaction score
0
Trophy points
1,281
Visit site
Activity points
1,430
Hello everybody,
I am a newbie in the field of analog design. I accidentally stumbled upon the gm/id methodology and found it to be pretty useful and many reviews say that it is comparatively better than the traditional design considering overdrive voltages especially for hand calculations. But i am having issues starting with this design. For example when i don't know what is the aspect ratios is need, i don't know how to generate the gm/id vs id/(W/L) curve. Also i read that the curves can be obtained from models or from experimental data. How do i obtain them or generate them from the BSIM model. If anybody has a document explaining it step by step i kindly request you guys to share it with me. Any other documentation is also welcomed .
Regards
Vasan SNS
 

You don't really have to bother about the aspect ratio. gm/Id, Id/W, gm/Cgg, gm/gds don't depend much on W. They do depend on L, though. So, just take a transistor with some W, say 5um and extract the curves for different L.
I think there are a lot of discussions how to do it here on this forum. You can also search lectures by prof Boser from Berkeley where he talks about the gm/Id methodology and its details. Lectures are freely available but they are frm about few years back.
 
VASANSNS said:
If anybody has a document explaining it step by step ...
Click to expand...
Here's a workshop presentation of Paul Jespers about this methodology. The book mentioned on page 53 isn't yet published AFAIK.
 



oermens said:
Not sure if this link will work...
Click to expand...
Works fine!

oermens said:
Actually I can see it in springerlink (dated dec 1 2009), but my university springerlink account does not give me access to the chapters, only front and back matter.
Click to expand...
Same for me. I think this is standard. They can't distribute gifts.

oermens said:
each chapter is 25$!!
Click to expand...
Actually it's cheaper to buy the book. And probably meant so.
 

There's no magic about gm/ids. I prepared a screenshot here to explain

In principle gm/ids is nothing else but picking a vgs_eff (which is vgs-vth)
As vth in BSIM3/BSIM4 is just a fitted parameter (not an exact value you can
count on) you can easily end up in problems, the same is of course also true
if you're checking for saturation margin ...
Back to gm/ids. It's well known that a vgs_eff around -50mV sets a device in
weak inversion. Usually this is used for differential input pairs to get lowest
offset voltage and lowest noise. This can also be used in low power circuits
Setting vgs_eff around 50-100mV gets us in moderate inversion which is also
a good setting for input pairs. a vgs_eff of 200mV sets us in mod/strong inversion
which is a good operating point for differential pairs with higher bandwidth
and currentmirrors at low supply voltage. Finally, a vgs_eff around 500mV yields
in strong inversion which is best for low offset, low noise current mirrors. However
500mV is hard
If you take a look at the attachment I have marked al four op's with markers plus
the vth threshold (~480mV) . It can easliy be seen that gm/ids is quite high
(around 20 in the example) for vgs_eff = -50mV. The midrange gm/ids is around
10-15 while we get gm/ids of 5 for very high vgs_eff. In my experience those values
are quite stable whether if you're working with 180nm or 45nm (if in doubt, simulate
gm/ids curve and and pick the -50,50-100,200,500 op points yourself) If you're having
trouble obtaining gm/ids in your simulator you can generate d(ids)/(ids) with your
waveform viewer. The result is quite accurate. This is also shown in the attachment
Just compare the lower two graphs.

Best Regards

Andi
 

baenisch said:
There's no magic about gm/ids.
... Usually this is used for differential input pairs to get lowest offset voltage and lowest noise.
Click to expand...
Thank you for your excellent clarification, Andi! I just would like to amend that, indeed, the input noise density is minimum in weak inversion, however the output noise current is maximum (for a given Id). Similar relations are valid for input offset voltage and current mismatch, s. e.g. the workshop paper from Eric Vittoz below, p. 11 .
 

oermens, Can you attach pdf files to here???

I tried to download but, first of all, can;t read "the WORD", and...I think I need to register on the site.but since I can't read it, I couldn;t do it..

thanks.
 

i can not post the link according to the forum rules, the rules changed some time ago:


Do not post e-books or requests for e-books or ieee papers here.

you can use google translate to view the site in english...
 

oermens said:
you can use google translate to view the site in english...
Click to expand...
Sure, but it seems you need points (or cells ;-) )to download:
"Sorry, your assets are insufficient to conduct this operation (less than 0 cell)."
 

make a post anywhere in the forum, you will get 5 points. it seems a lot of people make junk threads that you can reply to in order to increase your points (although since I dont read chinese, i can't distinguish between what is and isn't junk)
 

Have to correct myself. The book on gigapedia is not the gm/Id one that is discussed here.
 

I just noticed that all chapters are now available on springerlink for my academic account. anyone who has one should check.
 

Status
Not open for further replies.

Similar threads

Menu
Log in
Register
Cookies are required to use this site. You must accept them to continue using the site. Learn more…