4-A 100-MHz 100-dB Operational Amplifier

I709 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL 21, NO 12. DECEMBER 1992 A 100-MHz 100-dB Operational Amplifier with Multipath Nested Miller Compensation Structure Ruud G. H. Eschauzier, Leo P. T. Kerklaan, and Johan H . Huijsing, Senior Member, ZEEE Abstract-A 100-MHz bipolar operational amplifier has a gain of 100 dB. The op amp owes its high unity-gain bandwidth and high gain to an all-n-p-n signal path and multipath nested Miller compensation (MNMC). The phase margin with a 100-pF load is 40" at 100 MHz and the amplifier settles in 60 ns to 0.1% on a 1-V step. For comparison, a similar op amp without the mul- tipath technique has been realized. The unity-gain bandwidth of this nested Miller compensation (NMC) op amp is 60 MHz and the settling time is 70 ns. Theory and measurements con- firm that the multipath technique almost doubles the band- width of nested Miller compensated amplifiers. I. INTRODUCTION PEED or bandwidth demands are generally in conflict S with demands on low-frequency accuracy or gain in an operational amplifier (op amp). Op amps consisting of three gain stages to obtain an acceptable dc gain cannot be frequency compensated by conventional means. A widespread compensation method like simple pole split- ting is only capable of handling the two dominant poles that occur in two-stage amplifiers. For this reason, high-frequency bypass techniques are extensively used in high-frequency high-gain amplifiers [l] . In these amplifiers one gain stage is bypassed by a capacitor, short circuiting the stage for high frequencies. This compensation method greatly worsens the settling time of the operational amplifier because of the strong and inevitable pole-zero doublet the feedforward capacitor in- troduces [2]. A more effective way to compensate an amplifier con- taining three gain stages or more is by nested Miller com- pensation (NMC) [3]. This compensation technique nests Miller feedback loops, as shown in Fig. 1. The structure starts off with an output device with a Miller capacitor connected across it. For every gain stage added to the cir- cuit, an additional Miller capacitor is introduced, closing a wider feedback loop. Manuscript received April 27, 1992; revised July 10, 1992. This work was supported by the Technology Foundation (STW). R.G. H. Eschauzier and J . H. Huijsing are with the Department of Elec- trical Engineering, Delft University of Technology, 2628 CD Delft, The Netherlands. L. P. T. Kerklaan was with the Department of Electrical Engineering, Delft University of Technology, 2628 CD Delft, The Netherlands. He is now with Philips Industrial Electronics, 5600 MO Eindhoven, The Neth- erlands. IEEE Log Number 9203617. r I -L Fig. 1. Principle of nested Miller compensation. Unfortunately, NMC causes a bandwidth reduction compared to simple pole splitting. The bandwidth halves, for example, when stepping from a two-stage simple Miller compensated amplifier to a three-stage amplifier with nested Miller compensation. The multipath nested Miller compensation (MNMC) structure proposed in this paper overcomes the bandwidth reduction typical of conventional NMC by introducing an independent path for high frequencies. The MNMC struc- ture consists of NMC with a multipath input stage con- nected in parallel with the regular input stage of the op amp. The multipath stage directly drives the output tran- sistor, bypassing the intermediate stage for high frequen- cies. In the ideal case no pole-zero doublets occur, because the multipath input stage can be independently configured from the remainder of the amplifier. In practice, the matching depends on the ratio of transconductances and the ratio of capacitors, both being among the best con- trolled parameters in a standard IC process. To take full benefit of the MNMC, the operational am- plifier presented is based on an all-n-p-n topology, allow- ing an extremely high unity-gain bandwidth. Combined with an effective class-AB control, the result represents the state of the art in high-bandwidth precision opera- tional amplifiers. For comparison, two operational amplifiers have been realized. The first is compensated using NMC and dis- plays a unity-gain bandwidth of 60 MHz with a 100-pF load. The second op amp has an additional multipath in- put stage, which raises the bandwidth to 100 MHz under the same conditions. The organization of the paper is as follows. The next section addresses the principle of operation of the nested and the multipath nested Miller compensation structures. 0018-9200/92$03.00 0 1992 IEEE 1710 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 27, NO. 12, DECEMBER 1992 In Section I11 the two realized op amps are discussed, in- cluding the all-n-p-n topology and class-AB control. Sec- tion IV gives the experimental results. The paper finishes with the conclusions and references. 11. PRINCIPLE OF OPERATION A. Nested Miller Compensation Fig. 2 shows the simplified schematic of a three-stage amplifier with NMC. Fig. 3 is the corresponding Bode plot. It shows the frequency characteristic of the output stage, resulting from Cml and Cm2. The output stage has two dominant poles pI and p2, represented in Fig. 2 by the corresponding resistors and capacitors. Inserting C,, splits the poles, such that p1 shifts to a higher frequency pi and p2 to a lower frequency p;. This is the effect of normal pole splitting, resulting in a well-behaved combination of the intermediate and out- put stage. Miller capacitor Cm2, which closes the second loop, acts on the newly placed pole pi and the additional pole p3, the latter being introduced by the input stage. Splitting these two poles results in a straight 20-dB/decade rolloff from the dominating pole frequency pi up to the unity- gain frequency wo. The Miller capacitors also help reduce distortion by applying all internal gain across the output stage. The process of pole splitting is further clarified by Fig. 4(a) and (b). These figures show the root loci for C,, and Cm2, respectively. Pole p i , positioned by the inner Miller capacitor Cml. limits the bandwidth of the op amp. To first order pi depends on the ratio of the transconductance of the output stage and the load capacitor C,: The design criteria for the NMC follow from requiring a Butterworth frequency response from the amplifier with unity-gain feedback: ( 3 ) where gm2 is the transconductance of the intermediate stage and gm3 is the transconductance of the first stage. Expression (2) gives a hint about the dimensioning of a NMC amplifier. The unity-gain frequency of the inner loop, set by the transconductance of the intermediate stage gm2 and the inner Miller capacitor Cml, has to equal half the limiting pole frequency pi . As (3) reveals, the unity-gain bandwidth of the NMC op amp wo is one-fourth of the limiting pole frequency pi. This is half the value that could be obtained in a two- stage amplifier with simple Miller compensation. The bandwidth reduction is due to the downward shift of pole pi when the outer Miller loop with capacitor C,,,, CI P 3 P 2 P I Fig. 2. Three-stage NMC. Fig. 3 . Bode plot of the NMC structure. +& p3 p2" Fig. 4 . Root locus of the NMC structure with effects of (a) C,,,, and (b) C,.Z. is closed. At approximately half its frequency polepi col- lides with p3. Fig. 4(b) illustrates this effect. Pole fre- quency Ip;' I obeys B. Multipath Nested Miller Compensation Fig. 5 demonstrates how adding an independent paral- lel input stage transforms the NMC structure into the MNMC structure. Transistors Q, through Q3 together with the two capacitors Cml and Cm2 build up the conventional NMC structure. The multipath input stage is transistor pair Q4. This differential pair directly drives the output tran- sistor, overruling Q2 for high frequencies. In Fig. 6 the Bode plot of the MNMC amplifier is shown. Drawn as a dashed line is the high-gain low-fre- quency part established by the three-stage NMC ampli- fier. The solid line represents the high-frequency part cre- ated by the two transistors Q4 and Q , and Miller capacitor Cml. Since this HF part is solely determined by a two- stage amplifier with simple Miller compensation, no P3 P2 P1 Fig. 5 . Multipath NMC. Fig. 6. Bode plot of the MNMC structure bandwidth reduction takes place. Matching of the high- and low-frequency parts is easy, as the following analysis confirms. It is important to note that the multipath input stage adds a zero to the transfer function. The positions of the poles do not change compared to NMC. This makes clear that the dimensioning of the MNMC circuit should be dif- ferent from the NMC circuit, otherwise the second pole p;' will stay at its place and no bandwidth improvement is to be expected. The position of pole p;' with respect to its original position pi before closing the second Miller loop is given by ,, - P i Pi 4gm2 2 2 Pi Cm, P I - - + - I - - . From ( 5 ) it follows that the greater ratio gm2/Cm1 (the unity-gain frequency of the inner Miller loop) compared to the limiting pole frequency p i , the lower the bandwidth of the circuit. Setting gm2 to zero leads to p;' = p i . Ob- viously, in this case there is no bandwidth reduction, since only two stages are active. A better choice is With this ratio, the bandwidth reduction is only about 10% and still three stages contribute to the low-frequency gain. Putting the multipath zero on top of pole p j requires (7) The condition in (7) is satisfied by (6). As (7) reveals, the matching of the pole and the zero only depends on the matching of transconductances and capacitors. The position of the doublet is Or, with (6) since for a phase margin of 60" the unity-gain frequency wo of the op amp should be half p i (=pi ' ). The root locus in Fig. 7 shows the movement of the poles in the MNMC structure. In contrast to NMC, clos- ing the outer Miller loop only moves the poles a fraction, because of the low value g m 2 / C m I . Pole p j is eliminated by the multipath zero. 111. CIRCUIT DESCRIPTION A. All-n-p-n Topology 1711 ESCHAUZIER er a1 OP AMP WITH MULTIPATH NESTED MILLER COMPENSATION STRUCTURE ~ -~ ~-~ - ~ - - Fig. 8 is a simplified schematic of the op amp with NMC. To assure a high bandwidth, only n-p-n transistors are present in the signal path. As a consequence, in the push-pull output stage an emitter follower has to be used for the push and an inverting amplifier for the pull tran- sistor. The emitter following Qdo0 has a capacitor Cpl con- nected from its base to ground and the inverting amplifier Qsoo a Miller capacitor from base to collector. Capacitors Cml and Cpl have equal values. Surprisingly, when driven by a current signal both tran- sistor configurations behave symmetrically [4]. Not only do they have the same transimpedance z,, but also their output impedances zout are equal. Because of the differential second stage in Fig. 8, the circuit has a capacitor Cp2 added to it to balance out Miller capacitor C,, . The level-shift circuit, depicted as a voltage source in Fig. 8, has the characteristics of an all-pass current net- work [4]. In Fig. 9 the circuit is shown. For input currents there are two separate routes from input to output. For low frequencies the signal goes through the resistor and the p-n-p transistor. For high frequencies the path is through the capacitor and the n-p-n current follower. The crossover frequency fnp is set by the RC product. As long as& is lower than thef, of the p-n-p transistor, no pole- zero doublets occur, because no current is lost in the all- pass network. The bandwidth of the level-shift circuit equals the n-p-n'sf,. The location of the level shift in the circuit is dictated by noise considerations. Situating the level shift directly following the input stage would have increased the noise, because this would mean abandoning the passive collector loads. The MNMC op amp is largely similar to the NMC op 1712 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 27, NO. 12, DECEMBER 1992 multipath zero I o p2' 3- pl' pl" p3' p3 p2" Fig. 7 . Root locus of the MNMC structure. I I I input stage ' intermediate' level- output stage stage shift Fig. 8. Simplified schematic of the NMC op amp. Fig. 9 . All-pass level-shift circuit. amp in Fig. 8. Added is the multipath input stage, as Fig. 10 points out. B. Class-AB Control The feedback class-AB circuit controls the quiescent current of the output stage and prevents cutoff of the out- put devices by assuring a minimum quiescent current [5]. The control circuitry is independent from the signal path, as demonstrated in Fig. 11. The signal is applied to the output transistors as two differential currents, while the class-AB circuit controls the biasing by two common currents. Class-AB operation does not interfere with the output signal, because the common currents cancel in the output stage (i.e., when one of the transistors is controlled at its quiescent current due to a high output current, the I I I I I input stage ' intermediate' level- ' output stage stage shift Fig. 10. Simplified schematic of the MNMC op amp Fig. 11. Class-AB control - lout Fig. 12. Class-AB characteristic. class-AB control doubles the driving of the other transis- tor). The class-AB circuit incorporates a combined error am- plifier and decision gate. The decision gate, comprising e611 and & I , selects the smaller of the two transistor currents in the output stage. Controlling this current keeps the output transistors from shutting off. The transistors e611 and function as two emitter followers, but only the device corresponding to the lowest output current be- comes properly biased. This active emitter follower trans- fers its input voltage to the common-emitter node of the error amplifier. The error amplifier consists of the decision gate to- gether with e601 and Qm2. The input voltages of the de- cision gate are derived from the push and pull transistor currents by diodes Q750 and Q7m. The reference of the error amplifier is current If& across diode Q710. Fig. 12 shows the class-AB characteristic. The quiescent current is set by Zfef and the emitter ratios of the transistors. The minimum value is limited to half the quiescent current. Fig. 13. Total schematic of the NMC op amp 1713 VCr ZJQO V n Fig. 14. Total schematic of the MNMC amplifier. C. Total Schematic Fig. 13 shows the total schematic of the NMC opera- tional amplifier. This circuit diagram includes the biasing and level-shift elements. The actual circuit uses Darling- ton transistors in the output stage to improve the gain. The bias current is generated in the PTAT current source con- sisting of Q80()-&0 [ 6 ] . Resistor R,,, initiates a small current in the right-hand branch. Because of the cross- couplea structure, the magnitude of the current is of no consequence to the PTAT output current and start-up is guaranteed. The PTAT current is 100 pA at room tem- perature. The quiescent current of the output transistors is set to 4.5 mA. The MNMC op amp (Fig. 14) is, apart from the addi- tional input stage with Qlos and e,,,, largely comparable to the operational amplifier without the multipath tech- nique. To limit the bandwidth reduction, indicated by ( 5 ) , the transconductance of the intermediate stage is reduced 1714 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 21. NO. 12, DECEMBER 1992 Lbond s Lbond I C'ood Fig. 15. Separate voltage and current terminals at the output (a) (b) Fig. 16. Photomicrographs of the (a) NMC op amp and (b) MNMC op amp. by lowering the tail current of Q200 and and inserting degeneration resistors R,, and R210. The doublet fre- quency, according to (8), is 15 MHz. The lower tail current of the intermediate stage ensures that, despite the extra input stage of the MNMC op amp, the total supply currents of the two amplifiers are equal. IV. REALIZATIONS AND EXPERIMENTAL RESULTS The chips were fabricated in a 3-GHz 5 n-p-n bipolar IC process. To be able to drive a 100-pF load with a unity-gain bandwidth of 100 MHz, load and feedback are separated by two output pins and corresponding bond wires (Fig. 15). The pins act as current and voltage ter- minals and isolate the driving of the load from the feed- back path. Without this measure the load capacitance and inductance of the bonding wires would introduce a pair of complex poles in the feedback loop, resulting in instabil- ity of the circuit. Clearly the two output bonding wires can be seen in Fig. 16. Fig. 16(a) is a photomicrograph of the NMC and Fig. 16(b) of the MNMC op amp. The die area of both amplifiers is equal. The extra area needed on the MNMC chip for the multipath input stage is used in the NMC amplifier to accommodate the Miller capacitors. These capacitors are larger due to the lower bandwidth of the op amp. The die area of the chips is 1.2 mm x 1.5 mm. In Fig. 17 the Bode plots of the op amps are shown. The NMC op amp has a unity-gain bandwidth of 60 MHz with a phase margin of 40°C. The unity-gain bandwidth of the MNMC op amp is 100 MHz, with a phase margin of slightly less than 40". Both op amps are loaded by a 100-pF capacitor in parallel with a 1-kQ resistor, as is the case in the following measurements. Fig. 18 gives the slew response of the op amps to an input step of 1 V. Since the input stages are not degen- erated by emitter resistors, the slew rate is determined by the unity-gain bandwidth of the amplifiers. The slew rate of the NMC op amp (Fig. 18(a)) is 20 V/ps, and that of the MNMC op amp (Fig. 18(b)) is 35 V/ps. Fig. 19 gives an impression of the small-signal settling of the amplifiers. The input step is 100 mV. The 0.1% settling time corresponding to the NMC (Fig. 19(a)) is 40 ns. The step response very much resembles the designed for Butterworth curve. As Fig. 19(b) indicates a slow set- tling component is detectable in the step response of the MNMC amplifier. The doublet spacing corresponding to the slow settling component is approximately 5 % . The 0.1 % settling time is 50 ns. The contribution of the slow settling component to the total settling time becomes relatively less important for large input steps. Because most of the large-signal step response is governed by slewing of the op amp, the MNMC settles faster to 0.1 % after a 1-V input step than its NMC counterpart. This is confirmed by the plots in Fig. 20. Settling times are 70 and 60 ns, respectively. The last plot concerning the two op amps is shown in Fig. 21(a) and (b), which represents the input-referred voltage noise of the NMC and MNMC op amps, respec- tively. The voltage noise of the op amps is 2 nV/&. ESCHAUZIER et al.: OP AMP WITH MULTIPATH NESTED MILLER COMPENSATION STRUCTURE 0 MKR 59 274 131.098 I I Z A: REF 0: R T / H 403.095m d U 70.00 18 D I V D I V START 100 000.000 Hz OIV D I V START 100 000.000 H Z 10.00 3 6 . 0 0 STOP 200 000 000.000 HZ 10 .00 3 6 . 0 0 STOP 200 000 000.000 HZ (a) (b) Fig. 17. Bode plots of the (a) NMC op amp and (b) MNMC op amp. -20 GOO Fig. 18. Slew response of the (a) NMC op amp and (b) MNMC op amp (500 mV/div) 77 so0 ne , ns (a) (b) Fig. 19. Small-signal settling of the (a) NMC op amp and (b) MNMC op amp ( 5 mV/div) 1715 For frequencies above 15 MHz (the crossover frequency of the multipath input stage) the noise of the MNMC op amps goes up slightly. Because the intermediate stage is not active in this frequency region, noise of the level-shift circuits contributes to the input noise through the multi- pa