Voltage Buffer Compensation using Flipped Voltage Follower in a Two-Stage CMOS Op-amp Sri Harsh Pakala, Mahender Manda, Punith R. Surkanti, Annajirao Garimella and Paul M. Furth VLSI Laboratory, Klipsch School of Electrical and Computer Engineering New Mexico State University, Las Cruces, NM 88003, USA. Email:
[email protected] Abstract—In Miller and current buffer compensation techniques, the compensation capacitor often loads the output node. If a voltage buffer is used in feedback, the compensation capacitor obviates the loading on the output node. In this paper, we introduce an implementation of a voltage buffer compensation using a Flipped Voltage Follower (FVF) for stabilizing a two-stage CMOS op-amp. The op-amps are implemented in a 180-nm CMOS process with a power supply of 1.8V while operating with a quiescent current of 110µA. Results indicate that the proposed voltage buffer compensation using FVF improves the Unity Gain Frequency from 5.5MHz to 12.2MHz compared to Miller compensation. Also, the proposed technique enhances the transient response while lowering the compensation capacitance by 47% and 17.7% compared to Miller and common-drain compensation topologies. Utilization of FVF or its variants as a voltage buffer in a feedback compensation network has wide potential applications in the analog design space. Index Terms—Flipped voltage follower, frequency compensation, CMOS op-amps, voltage buffers, current buffers.
I.
INTRODUCTION
Miller [1], cascode [2], nested Miller (NM) and reversenested Miller (RNM) compensation schemes are the most widely used techniques to stabilize multi-stage amplifiers [3][21]. These compensation techniques are generally robust and offer advantages such as: (i) pole splitting, (ii) Left-Half-Plane (LHP) zero creation and its accurate placement through a nulling resistor and (iii), in the case of cascode compensation, eliminating the feed-forward path due to the presence of a current buffer, which is usually implemented with a commongate transistor [5]-[14]. These techniques typically exhibit one inherent disadvantage, in which the compensation capacitor loads the output node. For example, Miller compensation between input node X and output node Y of the gain stage shown in Fig. 1(a) loads the output with the compensation capacitor, as node X is approximately an AC ground compared to node Y. For stabilizing an op-amp designed specifically for high-speed applications, it is immensely important to reduce the loading effect on the output in order to achieve a fast and stable transient response. Buffers introduced in the compensation path do not affect the gain, but assist in reducing the loading effect on the output node. These buffers in compensation path eliminate the feed-forward path and generate LHP zeros. A current buffer (CB) when introduced in feedback compensation network as shown in Fig. 1(b) can be very effective due to its low input impedance, represented by ri_CB. While the utilization of the low-impedance node as a compen-
Fig. 1. Figure illustrating the loading of compensating capacitance on output node Y in case of (a) Miller compensation with nulling resistor and (b) current buffer. In case of (c) voltage buffer, compensating capacitance doesn’t load the output node Y.
-sating node in compensation schemes using CBs helps in reducing the size of compensating capacitor CCB that is required to ensure a stable system, the output node Y is still loaded by CCB. Fig. 1(c) shows a voltage buffer (VB) introduced in the feedback compensation path between node Y and X in series with capacitance CVB. The elimination of a direct connection of CVB to the output improves the transient response. The low impedance node at the output of the VB provides an additional advantage of establishing effective feedback [11], [22]-[23]. Thus a VB with very low output impedance r0_VB is preferred. Practical VBs can be implemented using a common-drain (CD) transistor. This paper introduces a VB compensation using a variant of the flipped voltage follower (FVF). In many applications, FVFs exhibit enhanced performance compared to a regular CD topology because of their very low output impedance [24]. In Section II, VB circuit topologies are analyzed and compared. Section III outlines the implementation details of the proposed compensation technique. Section IV presents the simulation results of Miller, CD and the proposed FVF compensation schemes. Section V illustrates the benefits of the FVF configuration over Miller and CD compensation schemes. Conclusions are presented in Section VI. II.
VOLTAGE BUFFERS
This section describes the implementation of different types of VBs suitable for compensation networks. A. Common-Drain Topology: The PMOS version of CD amplifier is shown in Fig. 2(a). Input VIN is applied to the PMOS CD transistor MCD, while the
VDD
VDD
I CD
VDD M2
VOUT VIN
MCD gmCD
2IFVF
VOUT VIN
MFVF gmFVF
VIN
M FFVF gmFFVF
VOUT
M2 I FVF VSS
(a)
VSS
(b)
IFVF VSS
(c)
Fig. 2. Voltage buffer topologies (a) common-drain, (b) fllipped voltage follower and (c) folded flipped voltage follower stages.
output VOUT is taken at its source terminal. Since the bias current ICD is always constant, the sourcee-to-gate voltage VSG,CD of MCD is also fixed. Therefore, aany small-signal change in the input will be directly followed by the output in order to keep VSG,CD constant. Since the output is always one VSG higheer than the input, the range of the input is {VSS, VDD – VSAT,ICDD – VSG,CD}. The input impedance looking into gate of CD is vvery high and the output impedance is 1/gmCD. The CD VB hhas high current sinking capability, whereas its sourcing capaability is limited by bias current ICD. B. Flipped Voltage Follower: The schematic of a PMOS version of FVF [24] is shown in Fig. 2(b). FVF consists of PMOS input transistor MFVF, transistor M2 with shunt feedback and bias ccurrent IFVF. FVF has high current sourcing but limited current sinking capability. The internal feedback loop helps in reducing the output impedance 1/gmFVF of the control transsistor by the loop gain gm2r0FVF. Therefore, the FVF output impeedance is 1 (1) . ROUT =
(g m 2 r0 FVF ) g mFVF
Although FVF exhibits ultra-low output impeedance, it suffers from limited operating voltage range. The inpput voltage range is limited by feedback transistor M2 and is given by VDD − VSG , M 2 + VSD , MFVF − VSG , MFVF ≤ VIN , FVF ≤ VDD − VSD , M 2 − VSG , MFVF . (2) Because of its low input range less than |VTHHP|, FVF may not be suitable for VB compensation. C. Folded Flipped Voltage Follower: The folded flipped voltage follower (F FFVF) [24] has ultra-low output impedance and doesn’t sufffer from limited operating voltage range. The PMOS version of FFVF, shown in Fig. 2(c), consists of PMOS input transsistor MFFVF and NMOS feedback transistor M2. In order tto eliminate the voltage clamping issue as in FVF, the feedbback transistor is folded. This doubles the bias current requirred to 2IFVF, but helps in obtaining a voltage range similar to a CD stage. The input range of FFVF is given by VSS + VGS , M 2 + VSD , MFFVF − VSG , MFFVF ≤ VIN , FFVF ≤ VDD − VSAT , 2 IFVF − VSG , MFFVF . (3) From Fig. 3, we note that the FFVF has sim milar input-output ranges compared to CD. On the other hhand, the output resistance of the FFVF is approximately 25x loower than that of CD, as seen in Fig. 4. Hence, the FFVF is an ideal candidate to implement in VB compensation.
Fig. 3. Input-output characteristic of CD and d FFVF topologies.
Fig. 4. Output impedance characteristic of CD and FFVF topologies.
III.
PROPOSED VOLTAGE BUFFER COMPENSATION USING FOLDED FLIPPED VOLTA AGE FOLLOWER Fig. 5 depicts the schematic off a two-stage op-amp with Miller, CD and FFVF compensation n techniques. The first stage is a PMOS differential pair with an NMOS current mirror load (M1- M5), while the second stage is a common-source amplifier (M6- M7). The output stage is sized K times larger in width than the unit sized tail PMOS transistor in the first stage and hence biased with K IBIAS. p-amp is given by The DC gain of the two-stage op (4) Av = g m1 R1 g m 2 ROUT , O where gm1 and gm2 are the effectivee transconductances of the first and second stages, respectively. R1 and ROUT are defined as the lumped resistances at the outp put of the first stage V1 and op-amp output VOUT, respectively. Three T variants of two-stage op-amps are implemented using the following feedback compensation networks: (i) Miller (ii) CD (iii) FFVF. Miller compensation network, formed fo by capacitor CM and nulling resistor RM, is placed betweeen nodes V1 and VOUT. CD compensation network is implemen nted through an additional branch biased with current ICD forrmed by transistors M8-M9, compensation capacitor CCD and d resistor RCD connected between node V1 and source of CD transistor M9 at node VX,CD. The proposed FFVF compensation n network is implemented through the addition of a FFVF brranch M10-M13, with a bias current of 2IFVF. Compensation caapacitor CFFVF and resistor RFFVF are connected between nodes V1 and VX,FFVF. Resistors RCD and RFFVF can be optionally used to move LHP zero formed by the VBs to desired domin nant frequencies [21], [25]. Both CD and FFVF compen nsation schemes require a nominal Miller capacitor CM and nu ulling resistor RM connected to node V1 in parallel, in order to sttabilize the op-amps over a wide range of load resistor RL and capacitor CL. Though an
VDD
VBIAS
M5
M11
M8
IBIAS
ICD V X,FFVF
M1
Vin-
gm1
M2
C CD
R CD
M10
M9
RL
gm10
IFVF M4
M13 Mil ler Compensation
Input Stage
CL
CM R M
V1
V SS
VOUT
VX,CD gmCD
M3
K IBIAS
g mFFVF
M12
C FFVF R FFVF
Vin+
2IFVF
M6
CD Compensation
FFVF Compensation
gm2
M7 Outtput Stage
Fig. 5. Figure illustrating Millerr, CD and proposed FFVF compensation schemes in a CMOS two-stage op-amp. o
additional Miller compensation network is reqquired in parallel with FFVF compensation, the compensatioon path created through low impedance node of transistor M12 allows the use of a very low-valued Miller capacitor CM. Bothh CD and FFVF compensation schemes require an extra biass current in their respective buffers when compared to Miller. Careful placement of LHP zeros help m maintain adequate phase margin over varying load condition. The LHP zeros generated by CD and FFVF compensation schhemes are: 1 (5) ω Z ,CD = ⎛ 1 ⎞ ⎜ + RCD ⎟ CCD ⎜ g m,CD ⎟ ⎝ ⎠
ω Z , FFVF =
1 ⎛ 1 ⎜ ⎜ (g ⎝ m10 r0 , FFVF g mFFVF
) + R FFVF
⎞ ⎟ C FFVF ⎟ ⎠
(6)
where, gmCD, gmFFVF and gm10 are the transconductances of CD and FFVF input transistors and FFVF feeddback transistor, respectively. r0,FFVF is the output resistance of FFVF input transistor M12. FFVF technique needs lower qquiescent current than CD compensation, such that 2IFVF