Designer’s Guide Community

grosser Community Member Offline Posts: 57	gm boosting Jun 28^th, 2006, 12:54am hello can you suggest me how to increase transconductance gmx of Mx transistor without increasing bias current? what circuit solution i should use? regards
Back to top	comp.PNG IP Logged

aamar

Community Member

Offline

Posts: 57
Germany

Re: gm boosting
Reply #1 - Jun 29^th, 2006, 6:20am

Hello,

I think you should try to plot the gm vs. vds (at your (vgs-vth) bias value) for your transistor for different channel lengths while keeping the W/L ratio constant (so that the current will not be changed).
i.e. W=10*L and use the L as a parameteric sweep variable

I tried it and I think the larger the channel length (while W/L is constant) you can increase your gm.

Best regards,

aamar

Back to top

IP Logged

Croaker Senior Member Offline Posts: 235	Re: gm boosting Reply #2 - Jun 29^th, 2006, 8:23am Aamar, what is the reason for that? These don't predict a higher gm under the conditions W/L=constant and I=constant. gm=kW/L(Vgs-Vth) gm=sqrt(2IkW/L) gm=2I/(Vgs-Vth)
Back to top	IP Logged

grosser Community Member Offline Posts: 57	Re: gm boosting Reply #3 - Jun 29^th, 2006, 10:31am Croaker is right It's not so easy, but i really need a good solution, so if You have any ideas please public it here regards
Back to top	IP Logged

Croaker Senior Member Offline Posts: 235	Re: gm boosting Reply #4 - Jun 29^th, 2006, 11:35am grosser wrote on Jun 29^th, 2006, 10:31am: Croaker is right It's not so easy, but i really need a good solution, so if You have any ideas please public it here regards I wasn't saying he wasn't right, only that I didn't get how that worked based on the simple square-law eqns.
Back to top	IP Logged

vivkr

Community Fellow

Offline

Posts: 780

Re: gm boosting
Reply #5 - Jun 29^th, 2006, 11:24pm

grosser wrote on Jun 28^th, 2006, 12:54am:

hello

can you suggest me how to increase transconductance gmx of Mx transistor without increasing bias current?

what circuit solution i should use?

regards

Hi,

You can certainly increase the Gm of Mx by using larger W/L ratio. Your bias current is set by the lower transistors and so this can be kept constant. In a differential pair, one can easily apply some small amount of +ve feedback to boost the Gm. You can look up the paper by Castello et al. (JSSC, June '90) and see if you can adapt his ideas somehow. However, be aware that circuits with +ve feedback need to be designed very carefully or else there will be instability. If nothing else works and you have access to good NPN BJTs, use them. MOS devices can never beat BJTs in terms of Gm/I. You will obviously need some way to compensate for the base current there.

The change in Gm vs. L is not apparent from the square-law model, and you cannot derive it thus. Short-channel devices have reduced Gm owing to velocity saturation which tends to shift the I-V curve from square-law towards a linear one. Look up any good device physics book (Tsividis for example).

Regards
Vivek

Back to top

IP Logged

grosser

Community Member

Offline

Posts: 57

Re: gm boosting
Reply #6 - Jun 30^th, 2006, 1:14am

vivkr wrote on Jun 29^th, 2006, 11:24pm:

You can certainly increase the Gm of Mx by using larger W/L ratio. Your bias current is set by the lower transistors and so this can be kept constant.

The change in Gm vs. L is not apparent from the square-law model, and you cannot derive it thus. Short-channel devices have reduced Gm owing to velocity saturation which tends to shift the I-V curve from square-law towards a linear one.

Regards
Vivek

Certainly larger W/L gives better transconductance, but i've already used the largerst possible W/L. Using longer channel and constant W/L increases gmx, but the increase is very small. I need about 5 times gmx

regards

Back to top

IP Logged

mikki33 Community Member Offline Analog/Mixed Signal/High Speed Posts: 57 Israel	Re: gm boosting Reply #7 - Jun 30^th, 2006, 1:20am You can make a replica of M_x and drive with that Nmos transistors at the output (25 pF), instead of cascoded current source. In this case you will have class A/B output stage with higher gain (and probably with lower current consumption)...
Back to top	If you don't have time to do it good you will find time to do it again IP Logged

aamar

Community Member

Offline

Posts: 57
Germany

Re: gm boosting
Reply #8 - Jun 30^th, 2006, 1:25am

Hello Croaker,

As Vivkr mentioned it cannot be derived from the square law model, or may be we cannot at the moment and it needs a look in a physics book, but from the practical point of view, I use the L to optimize the gm/gds of a transistor and in the same way I tried it only for the gm and the results were true, this means that gm can be improved with increasing the L till a limit where the change will not be noticable.

So I am sorry for not giving a clear answer now, but I will try to find it.

Best regards,

aamar

Back to top

IP Logged

mikki33 Community Member Offline Analog/Mixed Signal/High Speed Posts: 57 Israel	Re: gm boosting Reply #9 - Jun 30^th, 2006, 2:04am see the pic
Back to top	1_003.GIF If you don't have time to do it good you will find time to do it again IP Logged

grosser Community Member Offline Posts: 57	Re: gm boosting Reply #10 - Jun 30^th, 2006, 2:26am thank you for the pict you mean it's a fully diferential opamp to the left? i have found this soulution in attached picture but it doesn't work. it happens cause vdd=1.3V and vout to the left is 0.8V. this implies very low voltages at the opamps inputs which are cut-off
Back to top	with_opamp.PNG IP Logged

mikki33 Community Member Offline Analog/Mixed Signal/High Speed Posts: 57 Israel	Re: gm boosting Reply #11 - Jun 30^th, 2006, 3:00am So, don't use sourse follower at the input. You may make OTA to be differential to differential instaed of differential to single ended and convert to singlr ended only at the output.
Back to top	If you don't have time to do it good you will find time to do it again IP Logged

grosser

Community Member

Offline

Posts: 57

Re: gm boosting
Reply #12 - Jun 30^th, 2006, 3:37am

mikki33 wrote on Jun 30^th, 2006, 3:00am:

So, don't use sourse follower at the input. You may make OTA to be differential to differential instaed of differential to single ended and convert to singlr ended only at the output.

if i don't use source follower, i have 0.8V at the non inverting input and about 160mV at the inverting one. One input transistor is cutt-off and introduces low frequency parasitic poles.

Back to top

IP Logged