Feedforward Operational Amplifier
Feedforward Operational Amplifier
1/7/2005
Research Interest
Vin
Next Stage
SUBADC
DAC
The Classical Pipeline Stage has many different performance limitations. Amplifier Design is the main challenge when designing for low power, low voltage, high speed and resolution
Amplifier Settling Time (Bandwidth) Amplifier Settling Accuracy (Gain) Slew Rate
Performance Limitations
In phase signals %
Smaller feature si4e is actually good for this architecture /Better +B#56ower ratio3 while maintaining high () gain and going well )eyond the speed of internal parasitic poles
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005
7pen ;oop poles and 4ero /ignoring RsAIdeal Buffer &dded3C GB A Cf/(C' B Cf)*gm?5CII Z A gm'5Cf P' A go'5CI (CI=C1+Cf) P? A go?5CII (CII=C2+CL) Basem Soufi -Feedforward Operational 4ant P'56
Amplifier- ISU Copyright (c) 2003-2005
Pole-Zero cancellation
Two stage model ac simulation when no parasitics poles are included. Used ideal vccs resistors caps and a buffer to simulate the model.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 8
Two Stage model simulation when no parasitics poles included. Used ideal vccs resistors caps feedbac> networ> and a buffer to simulate the model.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 9
(uning the PolePole-6ero *ismatch with process .ariations /ase with no parasitic poles
"DE phase shift occurs at the pole. BDE phase shift occurs at the 4ero. Idea7
45
(etect the respective fre$uencies of the phase shifts of the 4ero and pole and tune one of them to overlap the other. The principle of pole"4ero mismatch correction is 8O( new. 1'2
#hen we have the closed loop poles widely separated then the best settling time is achieved when we have '112 pole"4ero cancellation. For widely separated poles the maGimum allowable overlap mismatch that results in a system that settles at least as fast as one without a mismatch is in the order of the settling accuracy re$uirement. So for 1.11'2 settling accuracy we need 1.11'2 tuning accuracy 192 this fact ma>es tuning very difficult if one desires very fast settling Basem Soufi -Feedforward Operational re$uirement.
Amplifier- ISU Copyright (c) 2003-2005 10
/losed-loop dominant pole mo.ement with different feed)ac: /losedfactors when perfect ;ero;ero-pole cancellation occurs *odel simulation with no parasitic pole in the first stage
Accurate second stage pole and zero cancellation
(ominant pole rolls to higher fre$uencies as the feedbac> factor increases identical to a first"order closed loop response. This occurs with accurate pole"4ero cancellation.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 11
Ideal two stage model with no parasitics in the first stage Poles mo.ement with different feed)ac: factors when the ;ero is faster than the second stage pole
Second Stage Dominant Pole Zooming in
First stage and second stage pole come together and form a comple< con=ugate pair at low feed)ac: factor .alues >close to #%>second stage open loop gain??, then, the pole of the second stage come )ac: close to the ;ero and the first stage tra.els to higher fre,uencies
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 12
Ideal two stage model with no parasitics Poles mo.ement with different feed)ac: factors when the ;ero is slower than the second stage pole
Zooming in
First stage pole come closer to the ;ero and gets partially cancelled and forms a dipole 4hile the second stage tra.els to higher fre,uencies
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 13
So, what does this tell us@ The relative position of the 4ero in respect to open loop poles has drastic effects on the behavior of the closed loop poles. It is an interesting case" case"study to see how fast the system settles relative to a two" two"pole system and how sensitive to process variations the system is when we have small closed loop factor or more specifically when we have compleG conHugate pair and a 4ero.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005
14
15
*odeling the two stage amplifier with a parasitic pole in the first stage
Feed-Forward Cap
Ideal Buffer
Vin
(eriving the transfer function Iout5Iin we see a three pole two 4ero system. If there are more parasitic capacitances in the first stage getting by"passed by the capacitor then an additional 4ero will appear. The case of one parasitic pole in the first stage is chosen to simplify the analysis.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 16
17
18
*o.ement of the poles in the closed loop configuration when the model has a parasitic pole in the first stage
In all the following figures +reen is the dominant pole of the first stage red is the dominant pole of the second stage and blue is the parasitic pole of the first stage. #e can see the additional 4ero in the transfer function due to by"passing the parasitic pole. #e can also observe that the parasitic pole always travels to much higher fre$uencies.
#hen the dominant =ero is after the pole of the second stage.
#hen the dominant =ero accurately cancels the second stage pole.
19
(uning Facts
Tuning poles and 4eros in FF opamps is 8O( new J?K. #hen having a parasitic pole the pole location of the second stage and the dominant 4ero are not located at "DE and BDE phase shifts respectively as the case when we had an ideal first order first stage amplifier. This ma>es the tuning the amplifier in the open loop phase domain harder if not impossible. <owever sweeping the FF capacitor in simulation over a certain range will guarantee a pole 4ero cancellation. Such tuning is illustrated neGt. #hen changing the FF capacitor will also change the +B of the amplifier however fine tuning change in the feedforward capacitor changes the +B negligibly.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005
20
10
6ero mo.ement with FF capacitance sweeping, and fine tuning the dominant ;ero location )y sweeping a .aractor at the parasitic node will )e shown ne<t
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 21
22
11
Fine tuning the dominant ;ero )y sweeping a .aractor added at the parasitic node
Fine tuning the dominant 4ero by sweeping a varactor at the parasitic node.
;arge sensitivity of @on" (ominant 4ero M its movement while incrementing the parasitic capacitance
23
(uning Ideas
#e can have a spectrum based tuning. Since it is a differential circuit the amplifier can be configured in a closed loop S) amplifier. The third" third "order harmonic distortion can be detected and the capacitor can be swept to minimi4e it. In &() design calibration algorithms can be used to tune the amplifier for less integral non" non" linearity. Since in two stage &mplifier there is only one 4ero" 4ero"pole cancellation ta>ing place the search algorithm for the optimal tune can be done with small compleGity.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 24
12
Buffered forward path adds a pole to the system with speed limitation and added power consumption. <owever the positive feedbac> is solved.
"'
)ancellation of the positive feedbac> with a negative feedbac> eliminates the pole in the feedforward path and provides a much more attractive speed5power tradeoff.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 26
13
The final circuit contains a feedforward path whose positive feedbac> path is cancelled with a buffered negative feedbac>. The canceling feedbac> paths should match well. This means the buffer should be as close to "' as possible while having the capacitors matched as well. Since the opamp &' already has a finite input capacitance there is a room for mismatch in the cancellation method The bandwidths of the buffer of the canceling negative feedbac> needs not to be any larger than the bandwidth of &'.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005
27
/omparison - #
The circuit in JEK provides and to a certain degree pole" pole "4ero trac>ing with process variations. <owever the feedforward path consumes a lot of power for a certain bandwidth as the designer in JEK saysC 9The 9 main restriction here is that the nondominant pole of the feedforward and second stage must be placed after the overall unity"gain bandwidth of the amplifier in order to minimi4e phase degradation.: The feedforward path of JEK consumes L'2 of total power...
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005
28
14
/omparison/omparison -'
The circuits presented in JDK and JEK reduce the output impedance of amplifier ma>ing it very difficult to maintain the () characteristic of the amplifier. To re" re"iterate the active feedforward scheme consumes a lot of power degrades the phase and reduces the output impedance.
29
/omparison/omparison -9
The presented circuit provides <igh +ain <igh Bandwidth without sacrificing much power in the feedforward bandwidth. If step response is very critical pole" pole"4ero calibration should ta>e place. If the amplifier is used in pipeline &() tuning can be done via detecting the lowest linearity errors of the outputs during the calibration process. & hybrid between this FF techni$ue and the regular miller compensation can be developed to provide a highest possible efficiency for an operational amplifier.
Basem Soufi -Feedforward Operational Amplifier- ISU Copyright (c) 2003-2005 30
15
31
32
16
References
J'K 9# ' F mi<ed analog%digital circuits using $ 9 Cm /*OS !SI technologyG Matsuura T.N Oano !.N <ira>i M.N Sasa>i O.N Miyamoto M.N Ishii T.N @agai R.N @ishida T.N Se>i !.N Imai4umi 8.N &nbo T.N Sumi @.N Ri>ino !.N Solid" Solid"State )ircuits )onference '00D. (igest of Technical 6apers. D'st ISS)). '00D I888 International 'P" 'P"'Q Feb. '00D 6age/s3C?E1 " ?E' J?K 9(echni,ue to eliminate slowslow-settling components that appear due to dipolesG Schlarmann M.8.N +eiger R.;.N )ircuits and Systems ?11'. M#S)&S ?11'. 6roceedings of the DDth I888 ?11' Midwest Symposium on Iolume ' 'D 'D" "'L &ug. ?11' 6age/s3CLD " LL vol.' J*K 9Relationship )etween amplifier settling time and polepole-;ero placements for second second-order systemsG Schlarmann M.8.N +eiger R.;.N )ircuits and Systems ?111. 6roceedings of the D*rd I888 Midwest Symposium on Iolume ' Q"'' &ug. ?111 6age/s3CED " E0 vol.' JDK 9A new multipath amplifier design techni,ue for enhancing gain without sacrificing )andwidthG Schlarmann M.8.N ;ee 8.!.F.N +eiger R.;.N )ircuits and Systems '000. IS)&S R00. 6roceedings of the '000 I888 International Symposium on Iolume ? *1 May" May"? Sune '000 6age/s3CP'1 " P'E vol.? JEK 9A ro)ust feedforward compensation scheme for multistage operational transconductance amplifiers with no *iller capacitorsG Thandri B.!.N Silva" Silva"Martine4 S.N Solid" Solid"State )ircuits I888 Sournal of Iolume *Q Issue ? Feb. ?11* 6age/s3C?*L T ?D*
33
17