1994 huijsing

8/8/2019 1994 huijsing

1/9

IEEE J O U R N A LOF SOLID-STATE CIRCUITS. VOL. 29. NO I? . DECEMBER 1994 1 SO 5

A Compact Power-Efficient 3 V CMOSRail-to-Rail Input/Output Operational

Amplifier for VLSI Cell LibrariesRon Hogervorst , JohnP. Tero, Ruud G. H. Eschauzier, and Johan H. Hui j s ing

Abstruct- This paper presents a two-stage, compact, power-efficient 3 V CMOS operational amplifier with rail-to-rail inputand output ranges. Because of it s small die area of 0.04 mm', it isvery suitable as a VLSI library cell. The floating class-AB controlis shifted into the summing circuit, which results in a noise andoffset of the amplifier which are comparable to that of a threestage amplifier. A floating current source biases the combinedsumming circuit and the class-AB control. This current sourcehas the same structure as the class-AB control which provides apower-supply-independent quiescent current. Using the compactarchitecture, a 2.6 MHz amplifier with Miller compensation anda 6 .4 MHz amplifier with cascoded-M iller compensation has been

realized. The opamp s have, respectively, a bandwidth-to-supply-power ratio of 4 MHdmW and 11 MHdmW for a capacitiveload of 10 pF.

I. INTRODUCTION

HE design of low-cost mixed/mode VLSI systems re-T uires compact, power-efficient library cells. Digital li-brary cells fully benefit from the continuing down-scalingof CMOS processes as well as from the ongoing reductionof the supply voltage. The down-scaling of processes resultsin smaller digital cells, because these cells can be designedusing minimum-length components. The power consumptionof digital cells is reduced by using a lower supply voltage[ I ] .In contrast to digital cells, analog library cells, such as the op-erational amplifier, cannot be designed using minimum-lengthcomponents, for reasons of gain, offset etc. Furthermore,a lower supply voltage doesnot necessarily decrease thedissipation of the cell because it often leads to more complexdesigns, resulting in a larger quiescent current.To obtaincompact, low-voltage, power-efficient analog cells, simplelibrary cells with good performance need to be developed.

A two-stage opamp is very suitableto obtain a compact de-sign. Published two-stage rail-to-rail opamps, howeve r, containcomplex class-AB output stages, which use large die area[2],[ 3 ] . Moreover, the class-AB control contributes significantlyto the noise and offset of the amplifier. Very recently, anoperational amplifier which overcomes the aforementionedproblems has been describedin [4]. This amplifier, however,uses a complex floating current source to bias the summing

Manuscript received June 6. 1994; revised August 22, 1994.R. Hogervorst, R. G.H. Eschauzier, and J. H. Huijsing are with Delft

University of Technology, Facultyof Electrical Engineering, LaboratoryforElectronic Instrumentation, 2628 CD Delft, The Netherlands.

J. P. Tero is with Philips Sem icond uctors , USA Application SpecificBusiness Group, Sunnyvale, CA 94088 USA.

IEEE Log Number 9406269.

circuit and the class-AB control. The transconductance ofthe input stage varies strongly with the common-mode inputvoltage, which impedes an optimal frequency compensation.Moreover, the quiescent current in the output transistors de-pends on supply voltage variations.

In this paper a compact, two-stage,3 V CM OS opamp withrail-to-rail input and output ranges will be presented. Becauseof its small die area, it is very suitable as a VLSI library cell.The opamp contain s a constant-g,, rail-to-rail input stage and asimple class-AB output stage. To save die area, the class-AB

driver circuit has been incorporated in the folded-cascodedsumming circuit of the rail-to-rail input stage. The floatingarchitecture of the class-AB driver prevents that it contributesto the noise and offset of the amplifier. This makes the inputoffset and noise of the amplifier comp arable to those of a three-stage amplifier. The combined summing circuit and class-ABcontrol is biased by a simple floating current source whichhas the same structure as the class-AB control, resultingin aquiescent current which is independent of the supply voltage.

Using the compact opamp, two designs with different typesof frequency compensation have been realized. The first opampis compensated using the well-known Miller splitting tech-nique [ 5 ] , esulting in a unity-gain frequency of 2.6 MHz. Thesecond opamp is compensated by inserting the folded-cascodeof the summing circuitin the Miller loop, which increases theunity-gain frequency up to 6.4MHz. The compact design ofthe opamps provides a low-power consumption of0.5 mW ata 3 V supply voltage. Both opamps occupy 0.04 mm2.

The constant-g, rail-to-rail input stage and class-AB drivercircuit are described in SectionI1 and Section111, respectively.Section IV describes the topology of the compact opamp. Theoverall design and the frequency compensation are describedin Section V. The realizations and measurement results arediscussed in Section VI. Finally, some conclusions are drawnin Section VII.

11. CONSTANT-GMAIL-TO-RAILNPUTSTAGEIn order to obtain a reasonable signal-to-noise ratio in low-

voltage design, the input stage should be able to deal withcommon-mode input voltages from rail-to-rail. This can beachieved by placing an N-channel and a P-channel differentialinput pair in parallel, as is shown inFig. 1 [ 5 ] .The N-channelinput pair, M I - M 2 , is able to reach the positive supply railwhile the P-channel input pair,M3 - M4, is able to reach

00 8-9200/94$04.00 0 994 IEEE

Authorized licensed use limited to: ULAKBIM UASL - ISTANBUL TEKNIK UNIVERSITESI. Downloaded on November 19, 2008 at 03:25 from IEEE Xplore. Restrictions apply.

8/8/2019 1994 huijsing

2/9

1506 IEEE JOURN A L OF SOLID-STATE CIRCUITS,VOL. 29, NO. 12, DECEMBER 1994

VDD

vss

Fig. 1.voltage is larger thant ibp Si,,,+ 21 is,, .

Common-mode input rangeof the rail-to-rail input stage. The supply

M D

vss

Fig. 2.voltage is smaller than\ is&, l i s n + 21i1.,,t.the negative supply rail. This input stage requires a supplyvoltage of at least:

Common-mode input rangeof the rail-to-rail input stage. The supply

where V,,, and Vgsp are the gate-sou rce voltage of anN -channel and the gate-source voltage of a P-channel transistor,respectively. Vdsat s the voltage across a current source whichis necessary to ensure that it operates as a current source.Ifthe supply voltage is belowKup,mi,,he input stage ceases tooperate in the middle of the common-mode input range, as isshown in Fig. 2 .

A drawback of the rail-to-rail input stage is that itsyvaries by a factor of two over the common-mode input range,as shown in Fig. 3. This large variation of theg, impedesan optimal frequency compensation[ 5 ] .An optimal frequency

compensation requires a constantyl n of the input stage. Inorder to obtain a constanty over the common-mode inputrange, the gm at the lower and upper part of the common-mode input range has to be increased by a factor of two. Sincethe g, of a MO S transistor operatingin strong inversion isproportional to the square-root of its drain current, the tail-current of the actual active input pair could be increased bya factor of four.

This principle is realized in the circuit as shown in Fig. 4[6]. The gm-control is implemented by means of two currentswitches,M5 and Mg, and two current mirrors,M G- M7 andMg - M I O , ach with a gain of three. T he principles of theyTn-control can be best understood by dividing the common-modeinput range into three parts.

If low common-mode input voltages are applied, i.e., volt-ages between VSS andVSS+ 1 V, only the P-channel inputpair operates. The N-channel current switch conducts whilethe P-channel one is off. The N-channel current switch takesaway the current Irefl nd directs i t to the current mirror,MG- M 7, where it is multiplied by a factor three and addedto ire^. Since I r e f l an d Ire= re equal, the tail-current of theP-channel input equals41r,f.

2.4

2.2

2

1.8

1.6

1.4

1.2

1

0.80 0 5 1 1.5 2 2.5 3

Vcommon

Fig 3. . . . . ail-to-rail input stage,- ormalizedg1,2 versu? the comm on-mode inputvoltageforRail-to-rail input stage with three-times current mirrors

M6 M7 M11 M1230/3 90/3 90/2 San

30/3 90 0 20w 2ow

Fig. 4. Rail-to-rail input stage withg? , , control by three-times currentmirrors.

If intermediate common-mode input voltages are applied,i.e., voltages betweenVS S + 1.3 V and VDD- 1.3 V, theP-channel as well as the N-channel input pair operate. Now,both current sw itches are off. The result is that the tail currents

of the N-channel input pair and thatof the P-channel inputpair are equal to I r e f .If high common-mode input voltages are applied, i.e.,

voltages between VDD- 1 V and VDD, only the N -channe linput pair operates. The P-channel current switch conductswhile the N-channel current switch is off. The P-channelcurrent switch takes away the currentIres and feeds it into thecurrent mirror,M S - M I O , here it is multiplied by a factor3and added to the currentI ref l . he result is that the tail-currentof the N-channel input pair equals41,,f.

It can be calculated that for each part of the common-modeinput range theg, is given by:

where p is the mobility of the charge carriers,CO , s thenormalized oxide capacitance,W and L are the width andthe length of a transistor, respectively. The subscriptsNand P refer to an N-channel or P-channel input transistor,respectivel y .

A h i d li d li i d ULAKBIM UASL ISTANBUL TEKNIK UNIVERSITESI D l d d N b 19 2008 03 25 f IEEE X l R i i l

8/8/2019 1994 huijsing

3/9

HOGERVORST PI d .. A COMPACT POWER-EFFICIENT AMPLIFIER FOR VLSl LIBRARIES 1507

From (2) it can be observed that for a constantgm the Wover L ratios of the P-channel and the N-channel input pairhave to obey the following relation:

If the ratio ~ ~ $ 7ve r p p differs from its nominal value because

of process variations, theQ~~ will have an additional variation.For example ifp , ~ ver ~ i p hanges about1S%,the additionalvariation will be approximately7.5%.

It can be concluded that theg m is approximately constantover the common-mode input range except for two take-overranges, i.e., common-mode input voltages betweenVS S+ 1V and Ls ~ + 1 . 3 V and common-mode input voltagesbetween VDD- 1 .3 V and VDD- 1 V, where the g m variesonly IS%, as is shown in Fig. 3. In the take-over rangesthe current through one of the current switches changesfrom 0 to Iref. r vice versa.

The offset of the rail-to-rail input stage changes over thecommon-mode input range because the N-channel input pairand the P-channel input pair have, in general, a differentoffset. This change in offset limits the CMRR of the inputstage, since it is defined as the change of offset relative tothe change in common-mode input voltage. To maximize theCMRR the change of offset should be spread out over a largepart of the common-mode input range. In this circuit, thechange of offset is spread out over the two take-over rangesof the current-switches. This allows a relatively large CMMRfor these types of input stages.

At supply voltages below 2.9V both current switches mightconduct at the same common-mode input voltage. To preventthe positive feedback loop,A45 - M l o , from becoming active.Mzg-hl31 are added to the circuit. Each side of the differentialpair, -4429 - h f 3 0 , is connected via a voltage source,vb 5 orvb6 , to either one of the supply-rails. The differential pair,M29 - M 3 0 measures the supply voltage. If the supply voltage

is larger than 2.9V the gate-voltage of the current switch,M , .is biased by M31. At supply voltages lower than 2.9V, thedifferential pair gradually tums offM 8 .Thus, the gate-voltageof the N-channel current switch moves towards the positivesupply rail. This means that the current switch is always off atsupply voltages below 2.9V. In this way the positive feedbackloop can never become active.

The current mirror,A411 - M14, together with the foldedcascodes, iWlj - M 1 6 , form a summing circuit. This summingcircuit adds the signals coming from the complementary rail-to-rail input stage.

111. RAIL-TO-RAILLASS-ABOUTPUT TAGE

To make efficient use of the supply voltage and supply-current, an opamp requires class-AB biased output transistorsconnected in a common-source configuration. Moreover, theclass-AB control should be compact to efficiently use die area.

The compact class-AB output stage is shown in Fig.S [7 ] . tconsists of two common-source connected output transistors,M25 and Adz,, which are directly driven by two in-phase signalcurrents, Ilnl nd I l n 2 . The floating class-AB con trol is formedby M l g and M z o . The stacked diode-connected transistors,

Fig. 5. Rail-to-rail output stage with floatingclass-AB control

1 IbZ I I . sslb3 IM V M lb6 lb5Fig. 6. Two-stage cascaded operational amplifier.

M23 - M 2 4 an d M21 - M22, bias the gates of the class-AB transistorsM 1 9 and M 2 0 , respectively. As was shown inthe previous section the minimum required supply voltage islimited by the demand for a fully rail-to-rail common-mode

input range. Therefore, two stacked gate-source voltages areallowed in the class-AB output stage.The floating class-AB control transistors, the stacked diode-

connected transistors and the output transistors set up twotranslinear loopsM20. M21, MZ2, and h l l g , M23, M24,M26, which determine the quiescent current in the outputtransistors. The class-AB action is performed by keeping thevoltage between the gates of the output transistors constant.Suppose the in-phase signal current sources,I i n l and I i n2 , arepushed into the class-AB output stage. As a result, the currentof the P-channel class-AB transistor,M20, increases while thecurrent in the N-channel class-AB transistors,M 1 9 , decreasesby the same amount. Consequently, the gate-voltages of boththe output transistors move up. Thus the output stage pullsa current from the output node. This action continues untilthe current through the P-channel class-AB transistor is equalto 1t,7. Now, the current of the P-channel output transistoris kept at a minimum value, which can be set byW over Lratios of the class-AB control transistors. Note that the currentthrough the N-c han nel outpu t transistor is still able to increase.A similar discussion can be held when input signals are pulledfrom the class-AB output stage.

Authorized licensed use limited to: ULAKBIM UASL - ISTANBUL TEKNIK UNIVERSITESI Downloaded on November 19 2008 at 03:25 from IEEE Xplore Restrictions apply

8/8/2019 1994 huijsing

4/9

1508 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 29, NO. 12 , DECEMBER 1994

Fig. 7. Three-stage casca ded operational am plifier

M I2 Vbl M21 1x3 VDD

V-

MI 7 M18 Vb4 lb5 M24

Fig. 9.mirrors which are biasedby two separate current sources.

Compact two-stage opamp. The sum ming circuit contains two current12 Vbl M21 Ib8

Both drawbacks can be overcome by adding an intermediatecircuit to the two stage cascaded opamp, as is shownin Fig.

gain between the input transistors,M I - M a ,and the biascurrents of the class-AB control,I b 6 and I b 7 , so the noiseand offset contribution ofI b 6 and Ib7 can be neglected. Theresult is that the noise and offset of the operational amplifier

lb3 lM V M l b5 is only determined by the input transistors,A 4 1 - M 4 , andby and by I b 3 - bbJ. The price to pay is moredie area and a lower unity-gain frequency. In general, morestages result in a lower unity-gain frequency of the opamp. Ifnested Miller compensation is used,th e unity-gain frequencydecreases by a factorO f two [ 5 ] . The die area is increasednot only because an additional stage is needed, but alsobecause additional capacitors,Cnr3 and C,zfi7 re needed tocompensate the oPamP. These capacitors, which are normall!,of the order of several PFs, OCCUPY considerable die area.

An alternative way to reduce the noise and offset contribu-tion of the class-AB control, without the cost of die area anda loss of unity-gain frequency, isto shift the floating class-AB

V- 7. The intermediate stage,M 3 1 and M 3 2 , increases the current

-Fig. 8. Compact two-stage opamp. T he floating class-AB controlis biasedby the cascodes of the summing circuit.

A drawback of the Class-AB control is that the quiescentcurrent of the output transistors depends on supply voltagevariations. The supply voltage variations are directly put. bythe gate-source voltages of the output transistors, across thefinite output impedances of the floating class-AB transistors.The result is a power-supply dependent variation of the qui-escent current.

Iv. TOPOLOGYF THE OPAMP

In the previous sections the rail-to-rail input stage and therail-to-rail class-AB output stage have been described. In thissection the overall topologyof the operational amplifier willbe described.

The conventional way to design a two-stage opamp is toplace the input stage,M1 - &I , and A411 - M l c , and theclass-AB output stage.M l g - 4426 , in cascade, as is showninFig. 6. The capacitorsCnll and c1,112 can be used to frequencycompensate the opamp. For reasons of clarity, the g,-controlof the input stage has been omitted. Although a comp act designis obtained, this approach has important draw backs. Firstly, thegain of the amplifier decreases because the bias current sourcesof the class-AB control,I,, and I b : , are in parallel with thecascodes, Idla and A 1 1 6 , of the summing circuit. Secondly,apart from the input transistors, - h14, nd A 4 1 1 - A 1 1 2and Ib 3 - b 4 of the summing circuit, also the bias currentsources of the class-AB control, andIb7 , contribute to thenoise and offsetof the amplifier. These bias sources contributesignificantly to the input offset and noise, because the currentgain between the bias sources of the class-AB control and thedrain currents of the input transistors is equal to one.

control, M l g an d M ~ o ,nto the summing circuit,AI11 - M 1 6 ,

as is shown inFig. 8. The floating class-AB control is biasedby the cascodes of the summing circuit. Now, the noise andoffset of the amplifier are mainly determined bythe inputtransistors and the summing circuit. This is also the case inthe three stage amplifier as is shownin Fig. 7. The opamp iseven smaller than the two-stage cascaded ope rational amplifier,as is shown in Fig.6, because the bias sources of the clawAB control have been eliminated. Note that for reasons ofsimplicity, the C,lrl and C,bl2 have been omitted.

A drawback of shifting the class-AB control into the sum-ming circuit is that the quiescent current of the output transis-tors depends on the common-mode input voltage. When thecommon-mode input voltage varies, the tail currents of theinput pairs, and therefore the currents through the cascodes,change. The result is that the bias current of the class-ABcontrol, and consequently the quiescent currentof the outputtransistors, depends on the common-mode input voltage. Thisproblem can be overcome by using a summing circuit withtwo current mirrors,M 1 1 - d 1 4 and M 1 5 - M 1 8 , which arebiased by two separate current sources,I b 3 and I b 4 . as is shownin Fig. 9 . Both sides of each current mirror are loaded byequal common-mode currents, coming from the input stage,


8/8/2019 1994 huijsing

5/9

HOGERVORSTer a/. A COMPACT POWER-EFFICLENT A MPLIFIERFOR VLSI LIBRARIES 1509

z VDD12 Vbl M21 Ib8

# I I I I I

M18 Vb4 lb5

Fig. IO . Compact two-stage opamp. The summing circuit containstw ocurrent mirrors which are biased by one floating current source.

M1 - M 4 . Since I,, and I,, have the same value, the outputcurrent of both current mirrors is equalto I b 3 .

A drawback of the separate biased current mirrors is thatthe bias current sources of the current mirrors contribute tothe noise of the amplifier because the current gain between thecurrent sources and the drain currentsof the input transistorsis equal to one. Mismatch in the bias current sources will alsocontribute to the offset of the amplifier. To overcome theseproblems the current mirrors are biased by a floating currentsource, I63, as is shown in Fig. 10. Because of the floatingarchitecture of the current source,i t does not contribute to thenoise and offset of the amplifier[8], [9].

As described in Section111 the class-AB control, andtherefore the quiescent current in the output transistors, suffersfrom supply voltage variations. To make the quiescent currentof the output transistors insensitive to supply voltage varia-tions, the floating current source should have the same supplyvoltage dependency as the class-AB control. Fig.11 showsthe amplifier with a practical realizationof such a floatingcurrent source, - ~1428. he floating current source hasthe same structure as the floating class-AB control. The value

of the current source is set by two translinear loops,M11,MZ1, M 2 2 , 1142~ and ~2117, 1Vf23, 1,124, hl27. The mirrors,~W11 - M14 and M15 - 5118, are loaded by the drain currentsof the input pairsM1 - and AI, - M a , respectively. Thesedrain currents, and consequently the gate-source voltages ofM ll and k l 1 7 , change with the common-mode input voltage.If, for example, the common-mode input voltage approachesthe positive supply rail, the SI,-control circuit increases thecurrent of and decreases the current ofI b 2 . As a result, thegate-source voltage ofMI 1 decreases while the gate-sourcevoltage of hf17 increases. However, this hardly effects thevalue of the floating current source,Ad27 - M ~ s ,ecause anincrease of the gate-source voltage of one mirror compensatesfor a decrease in the other mirror. The result is that the currentof the floating-current source vanes only5% #over the fullcommon-mode input range, as is shownin Fig. 12.- M 2 8 , has the samestructure as the class-AB driver circuit,hflg - bf20, he supplyvoltage dependency of the current mirror compensates for thesupply voltage dependencyof the class-AB driver. The resultis a quiescent current which is insensitive to supply voltagevariations.

Since the floating current source,

M I 2 V b l M21 lb8

V

MI 8 Vb4 lb5

Fig 1 1 Compact two-stage opampThe floating current sourceI \ imple-mented by meansof -1127 and 1 I z q

Floating current source

o0 85

0 0 5 i 1 5 2 2 5 3

Vcommon

Simulation resultof the normalized kalue of the flodting currentig. 12source vcrws the common-mode input voltage

The resulting two-stage topology is compact which makesit very suitable as a VLSI library cell. It has an offset andnoiye which are comparable to those of a three gtage amplifier.

Moreover, the quiescent current of the operational amplifier isinsensitive to supply voltage variations.

v. OVERALL DESIGNND FREQUENCY OMPENSATIONUsing the topology as described in the previous section, a

compact opamp with M iller compensation has been designed,and is shown Fig. 13. The opamp consistsof the rail-to-railinput stage, M -M 4,g m- con tro l, h.l;-n/!10 andM2,-Af31, asumming circuit,M 1 1 M 1 8 , and a rail-to-rail class-AB outputstage, MlY - M26. The floating current source,M 2 7 - M 2 8 ,biases the summing circuit and the floating class-AB control.The opamp is compensated using the conventional Millertechnique. The capacitorsCjlrl and C, , I~ round the outputtransistors, M25 and hf26, plit apart the poles ensuring a20

dB per decade roll off of the amplitude characteristic. Theconventional Miller splitting shifts the output pole up to afrequency of approximately

Sm oWout =

-L

(4)

where gTROs the transconductance ofthe output transistors andCr. is the load capacitor. InFig. 14 , a second design of the


8/8/2019 1994 huijsing

6/9

1510 IEEE JOURNA LOF SOLID-ST.4TE CIRCUITS, VOL. 29, NO 12. D ECE M B ER 1994

Fig. 13.Miller compensation.

Overall designof the compact rail-to-rail operational amplifier with

Fig. 14.cascoded-Miller compensation.

compact rail-to-rail opamp is shown. This opamp is basicallythe same as the opamp shown inFig. 13, except for thefrequency compensation scheme. This opamp is compensatedby including the cascode stages,MI^ and M 1 6 , in the Millerloops. This compensation technique shifts the output pole toa frequency of approximately

Overall designof- he compact rail-to-rail operational amplifier with

( 5 )C,\I gmo

J,,t = ~-

GS.out CL

where C,tf and C G S . , ~ ~re the total Miller capacitor andthe total gate-source capacitance of the output transistor,respectively.

The output pole in ( 3 , nd therefore the unity-gain fre-quency of the cascoded Miller compensation, is a factorC , ~ ~ / C G S , , ~ ~arger than the output pole in (4 ). In the opam p asshown inFig. 14, this ratio is chosen to be about2.5, resultingin a unity-gain frequency of the compact opamp with cascodedMiller compensation which is about2.5 times higher comparedto the opamp with conventional Miller compensation.

VI . REALIZATIONS ND MEASUREMENTRESULTS

The opamps have been realized in a1 pm BiCMOS process.The N-channel transistors and the P-channel transistors have

threshold voltages of0.64 V and -0.75 V, respectively. Themicrograph of the compact opamp with Miller compensationand the micrograph of the opamp with cascoded-Miller com-pensation are shown inFig. 15. A Bode plot of the compactopamp with Miller compensation is shown inFig. 16 . Theopamp has a unity-gain frequency of2.6 MHz, with a loadcapacitor of 10 pF. The unity-gain phase margin is66'. Fig.

(b )

Fig. 15.Miller compensation.(b ) cascoded-Miller compensation.

Micrograph of the compactrail-to-rail operational amplifier with (a)

16 shows the Bode plot of the compact opamp with cascodedMiller compensation. The opamp has a unity-gain frequency of6.4

MHz and a phase margin of530, with the same capacitiveload of 10 pF. To compare the amplifiers, the following figureof merit is used:

F=-1psup L = l O p F (6)

where B is the unity-gain frequency of the opamp,Psup sthe quiescent dissipation andCL is the load capacitor[ lo] .The figure of Merits of the opamp with Miller compensationand that of the opamp with cascoded-Miller compensation are4 and 11 MHz/mW, respectively, for a capacitive loadof 10pF. It can be concluded that the opamp with cascoded-Millercompensation uses the power about2.5 times more efficiently,with respect to bandwidth than the Miller compensated opamp.

Fig. 17 shows the small signal step responseof the opampwith Miller and cascoded-Miller compensation, respectively.The opamp with Miller compensation responds to small-signals within 1% of the final value in 220 ns, for a loadof 10 pF and a step of 100 mV. The opamp with cascodedMiller compensation has a small signal-settling time180 ns,for the same accurracy of1 % , capacitive load and step. Thelarge signal step response of the opampsis shown in Fig. 18.

B


8/8/2019 1994 huijsing

7/9

HOCERVORST er al.: A COMPACT POWER-EFFICIENT AMPLIFIER FOR VLSl LIBRARIES 1511

a

NETWORKA: RE F 8: RE F 0 MKR 2 60 0 9 5 8 . 6 4 7 Hz

60.00 2 5 0 . 0 T / R - 2 4 8 . 1 9 8 m d 8[ dE 1 d e g 1 8 - 1 1 4 . 2 9 1 d e g

I I I l l I I I l l I

.a, rU6 a 1

D I V OV S TA R T l o o o o o . o o o n z D I V D I V S TA RT 10 0 0 0 0 . 0 0 0 HZ1 0 . 0 0 5 0 . 0 0 STOP 20 00 0 0 0 0 . 0 0 0 HZ 1 0 . 0 0 50 0 0 STOP 20 00 0 0 0 0 . 0 0 0 H z

RBW: 100 H Z ST: 4 3 . 1 se c R A N G E : R--10. T--1OdEm RBW: 30 0 HZ S T: 1 4 . 3 s e c R A N G E : R- -10 . T- O d E m

( a ) (b)

Fig. 16. Bode plot of the compact rail-to-rail operational amplifier with(a ) Miller compensation, and ( bj caacoded-Miller compensation

100 n s / d l v

(a) (b)

Fig. 17.Miller compensation. (y-axis scale= 10 mVldiv).

Small-signal step response(lLtv,, = 100 m V) of the compact rail-to-rail operational amplifier with (a) Miller compensation and(b ) cascoded

The 1% large-signal settling time for the compact opamp withMiller compensation is 440 ns, for a capacative loadof 10pF and a step of 1 V. The 1% large-signal settling time forthe compact opamp with cascoded Miller compensation is275ns, for the same capacitive load and step. It can be concludedthat the compact opamp with cascoded Miller compensationis faster than that with the Miller compensation.

From Fig. 18it can be observed that the slew-rate of theopamp s change by a factor of two. As was explained in Section11, in the intermediate common-mode, input voltage range inboth input pairs are active. At the upper and lower part of thecommon-mode input range, the tail current of the actual activeinput pair is increased by a factor of four. Thus, in the outerparts of the common-mode input range, there is two times asmuch current to charge the compensation capacitor as thereis in the intermediate partof the common-mode input range.Therefore, slew-rate changes by a factor of two. The slew-rateof the compact opamp with Miller compensation is2 V/ps ,when the common-voltage is in the range ofVSS+ 1. 3 V andVD D 1.3 V. It is 4 V/ps when the common-voltage isinthe range ofVS S nd VS S+ 1 V or in the range of VD D - 1V and VDD. The slew-rate of the compact opamp with Millercompensation is 4 V/ps. when the common-voltage is in therange of VS S+ 1.3 V and VDD- 1.3 V. It is 8 V/pm whenthe common-voltage is in the range ofVS , and Ifss + 1 V orin the range ofVDD- 1 V and VDD.

At high output currents, the cascoded-Miller compensationcould give rise to peaking[ 111. However, at the maximumoutput current of this opamp, which has a value of3 mA ,the peaking is negligible. If the operational amplifier has todrive output currents much larger than3 mA, the amplifierwith Miller compensation should be used.

A list of specifications is given in TableI. The minimumsupply voltage is 2.5 V. At this voltage both opamps dissipateonly 0.45 mW. At supply voltages between2.5 and 2.9 Vthe opamp is able to deal with common-mode voltages in therange from L& - 0.4 V to VDD- 1. 4 V. At supply voltagesabove 2.9 V, the opamps are able to deal with common-modeinput voltages from rail-to-rail, or even beyond the rails. Thecommon-mode input range is fromVSS 0.4 V to ID , +0.5V. The maximum supply voltage is 6 V and is determinedby the process. The gain of both opamps is approximately85 dB. The gain can be increased by applying gain boostingtechniques to the cascodes M 14 and M I6 of the amplifiers asshown in Figs. 13 and 14 [12]. The offset of both opampsis about 5 mV, which is comparable to thatof three-stageoperational amplifiers. The offset can be reduced, to valuesof about 2 a 3 mV, by increasing the area of the inputtransistors and by using common-centroid layout structures.The CMRR of the opamps is determined by the change ofoffset relative to the change in common-mode input voltage.The offset changes gradually during the take-over ranges, i.e.,


8/8/2019 1994 huijsing

8/9

1512 IEEE JOURNALOF SOLID-STATE CIRCUITS, VOL. 29, NO. 12, DECEMBER 1994

840.000 ns200 n S / d l v

(a) (b)

Fig. 18.compensation (y-axis scale= 200 mV/div).

Large-signal step response of = 1 V) the compact rail-to-rail operational amplifier with (a) Miller compensation; ( b) cascoded-Miller

TABLE IMEASUREMENTESULTS supply= 3.3 V, Rload = I O kl2,

Cload = IO pf, TA = 27C. UNLESS THER WIS ETATEDl s b l e I: Measurement rrsults Vsupply=JdV,Rload=lOkQ Cload=lO pF, TA=27C,

unless otherwise stated

Parameft1 IIO W P l IOpamP2 IDk mea II0.04 10.04 I mm2Supply voltage range ll2.5-6 I2.5.6 I V

Quiesant ~ m ~ t 11180 I180 IPA

Offsetvoltage

Input mise v o l a g e 0 IO kHz

CMRRVc h m s,y 4V IO V s e l Vandfrom V&l 3 V mV D D-I3Vanb rom VDD-lV U, VD& 5

V

8/8/2019 1994 huijsing

9/9

HOGERVORSTCf al.: A COMPACT POWER-EFFICIENT AMPLIFIER FORVLSl LIBRARIES 1513

Ron Hogervorst, for a photograph and biography, see this issue, p.1504 Ruud G. H. Eschauzier, for a photograph and biography, see this issue, p.1504.

John Tero was born in Northampton, England in1944 and received the B.Sc. degreein physics fromthe University of Leicesterin 1966, and the M.Sc.degree in electronic devicesfrom the UniversityofLondon in 1970.

He joined Hawker Siddeley Dynamicsin 1966and workedon satellite attitude control and propul-sion systems. He joined Plesset Research in1972an d designed high frequency integrated circuitsformobile radio and radar applications in bothsil-icon and gallium arsenide technologies.In 1979

Johan H. Huijsing, for a photograph and biography, see this issue,p. 1504.

he moved to Ferranti Semiconductors to lead a group manufacturing anddeveloping RF bipolar power transistors, and qince1984, he has been withPhilips Semiconductorsin Sunnyvale CA, where he is currently DesignEngineering Managerfor LAN products.As such, he is involved in thedevelopment of w ired and wirelessLA N circuits in bipolar. CMO S, BiCMOStechnologies and his currentareas of interests are in analog signal processing,high speed data recoveryand generation and low voltage circuit techniqu e\.

1994 huijsing

Documents