Class AB Voltage Follower and Low-Voltage Current Mirror with Very High Figures of Merit Based on the Flipped Voltage Follower

Abstract

The application of the flipped voltage follower to implement two high-performance circuits is presented: (1) The first is a class AB cascode flipped voltage follower that shows an improved slew rate and an improved bandwidth by very large factors and that has a higher output range than the conventional flipped voltage follower. It has a small signal figure of merit FOM_SS = 46 MHz pF/µW and a current efficiency figure of merit FOM_CE = 118. This is achieved by just introducing an additional output current sourcing PMOS transistor (P-channel Metal Oxide Semiconductor Field Effect Transistor) that provides dynamic output current enhancement and increases the quiescent power dissipation by less than 10%. (2) The other is a high-performance low-voltage current mirror with a nominal gain accuracy better than 0.01%, 0.212 Ω input resistance, 112 GΩ output resistance, 1 V supply voltage requirements, 0.15 V input, and 0.2 V output compliance voltages. These characteristics are achieved by utilizing two auxiliary amplifiers and a level shifter that increase the power dissipation just moderately. Post-layout simulations verify the performance of the circuits in a commercial 180 nm CMOS (Complementary Metal Oxide Semiconductor) technology.

1. Introduction

Two of the basic building blocks of analog integrated circuits are the voltage follower and the current mirror. The conventional common-drain amplifier or voltage follower (denoted here as CONV_VF)) [1,2,3] of Figure 1a has been used for many years as a buffer due to its infinite input resistance, medium–low output impedance R_out = 1/g_m (in the order of tens of kΩs) close to the unity voltage gain, and relatively high bandwidth BW = g_m/(2π C_L). The flipped voltage follower of Figure 1b [4] is an improved voltage follower that uses local negative feedback to provide lower output resistance R_out = 1/[g_m(g_mr_o)] = 1/(g_mA) (hundreds of Ω), where C_L is the load capacitance, g_m and r_o are the transconductance gain and output resistance, and A = g_mr_o is the intrinsic gain of the MOS transistor. The basic Flipped Voltage Follower(FVF) version of Figure 1b (denoted here as CONV_FVF) suffers the serious limitation that it has a very low peak-to-peak output swing V_oswingpp, which is independent of the supply voltage and given by V_oswingpp = V_TH − V_DSsat (where V_TH is the threshold voltage and V_DSsat is the drain-source saturation voltage of M_FVF). It operates in class A with a peak positive output current and positive slew rate limited by the bias current I_bias to a value SR⁺ = I_bias/C_L. Several versions of the flipped voltage follower have been reported with improved output range and lower output resistance. For example, the cascode FVF (denoted here as CONV_CSCFVF) shown in Figure 1c and reported in [5] uses an additional branch with a cascode transistor M_CAS that increases the local feedback loop gain and provides even lower output resistance by a factor A = g_mr_o so that, in this circuit, R_out = 1/[g_m(g_mr_o)²] = 1/[g_mA²] (on the order of tens of Ωs). It also has an increased output swing, which is dependent on the supply voltage and given by V_outswingpp = V_DD − (V_GS + V_DSsat) = V_DD − V_TH − 2V_DSsat. The cascode FVF of Figure 1c (denoted here as CONV_CSCFVF) is a class A circuit with a positive slew rate seriously limited by the bias current to a value: SR⁺ = I_bias/C_L. Class AB versions of the FVF have also been reported [6,7,8,9] to overcome this limitation to a certain degree.

The low-voltage cascode current mirror of Figure 2a (denoted here as CN_CS_CM) has been used for many years as a high-bandwidth linear current amplifier. It has a moderately low input resistance R_in = 1/g_m (on the order of tens of kΩs), moderately high output resistance R_out = r_o(g_mr_o) = r_oA (on the order of tens of MΩs), high linearity, low gain error, low output compliance voltage (V_outmin = 2V_DSSat), and moderate input voltage requirements V_in = V_GS = V_TH + V_Dssat. A simple rearrangement of the circuit of Figure 2a is shown in Figure 2b. It injects the input current source I_in at the source of the cascode transistor M_1C (node V_x) instead of at its drain (node V_y). This reduces the input voltage requirements from V_GS to V_DSsat and leads to a reduction in the input resistance by a factor A from R_in = 1/g_m to R_in = 1/[g_m(g_mr_o)] = 1/[g_mA]. Notice that, in this circuit, the input transistors M₁ and M_1C form a flipped voltage follower with a constant input voltage V_cn at the gate of M_1C and the current input signal I_in injected at the output terminal of the FVF (node V_x). In spite of the improvement in the mirror characteristics, this modification suffers from a non-linear current mirror gain resulting from lambda effect and unequal drain source voltages in the input and output transistors M₁ and M₂ of the current mirror. This is due to the fact that the cascode input and output transistors M_1C and M_2C have unequal drain currents which cause their gate-source voltages to be different. The gate source voltages of M_1C and M_2C determine the drain-source voltages of M₁ and M₂ and the linearity of the mirror. This effect can be greatly mitigated in a BiCMOS process by replacing the cascode transistors by bipolar transistors.

In this paper, the authors introduce two improved circuits based on the flipped voltage follower: (1) a power-efficient class AB cascode FVF (denoted here as HP_CSCFVF) with high swing, very low output resistance, and essentially higher small signal and large signal figures of merit than previously reported AB FVF versions and (2) a low-voltage high-performance cascoded current mirror (denoted here as HP_CS_CM) with much lower input resistance and much higher output resistance than the conventional current mirror and highly linear gain. The proposed circuits are described in Section 2. Section 3 shows the post-layout simulation results that verify the high-performance characteristics of the proposed circuits, and Section 4 provides the conclusions.

2. Proposed Circuits

2.1. High-Performance Class AB Voltage Follower HP_CSCFVF

2.1.1. Description

Figure 3 shows the scheme of the proposed class AB high-performance cascode FVF (HP_CSCFVF). It is a modification of the class A cascode FVF reported in [5] and is shown in Figure 1c. It includes an additional branch with a PMOS transistor M_source that provides efficient class AB operation. M_source has a small quiescent current I_Qsource, but it can inject positive output currents I_out into the load C_L, which are much larger than I_Qsource. On the other hand, transistor M_sink can sink very large negative load currents (also much larger than the quiescent current of M_sink given by I_Qsink = I_Qsource + I_bias), as is discussed below. The biasing branch has two diode-connected PMOS transistors M_b and M_bC. Based on replica biasing, this branch sets the voltage V_y to a value V_y = V_DD − V_SGb and the gate voltage of M_source to a value V_g = V_y + V_bat, where V_bat = I_batR_bat. The values of I_bat and R_bat are selected so that V_bat has an approximate value V_bat = V_SDsat = 0.1 V. In this case, the quiescent source-gate voltage of M_source is given by V^Q_SGsource = V_SGb − I_batR_bat. This leads to a quiescent voltage V^Q_SGsource = V_SGb − V_SDsat ≈ V_TH close to the threshold voltage of M_source. This quiescent source-gate voltage sets a relatively small quiescent current I_Qsource in M_source, which is independent of the supply voltage. A capacitor C_bat forms a high-pass filter with R_bat and C_bat. This is used to transfer fast transient variations from V_x to the gate V_g of M_source.

2.1.2. Class AB Operation of Proposed Voltage Follower

For positive input voltage variations in V_in, the output voltage V_oHPCSC increases, and the gate voltage V_x of M_sink decreases. The current in M_sink decreases (and eventually turns off) and the dynamic changes in V_x are transferred to V_g by C_bat to the gate of M_source. The decrease in V_g causes the current I_source of M_source to increase, providing a positive output current I_out that can be significantly larger than the quiescent current I_Qsource of M_source (in the design described in Section 3, I_Qsource has a value I_Qsource = 7 µA). For negative input voltages, the output voltage decreases, and V_x increases. This increases the current I_sink and provides a large negative output current I_out that can be much larger than the quiescent current I_Qsink of M_sink. The FVFs in Figure 1b,c and Figure 3 use an RC compensation network formed by C_c and R_c for the local feedback loop. These elements provide a dominant pole and a high-frequency zero at V_x that approximately match the output pole of the open loop gain ω_pout = g_mFVF/C_L at the output node V_oHPCSC. This allows FVF circuits to significantly improve their bandwidth with respect to the conventional voltage follower of Figure 1a. This simple but effective FVF compensation and bandwidth extension technique was reported in [10].

2.2. High-Performance Low-Voltage Current Mirror (HP_CS_CM)

The scheme of the proposed low-voltage high-performance current mirror (HP_CS_CM) is shown in Figure 4a. It is a modified version of the FVF current mirror of Figure 2b that includes two auxiliary amplifiers A_SEinvfcamp with gain A_aux and a level shifter FVF_levelshifter. The transistor level implementation of the auxiliary amplifiers and the level shifter is shown in Figure 4b. Each of the amplifiers form local negative feedback loops with the cascode transistors M_1C and M_2C. They boost their effective gain by the gain A_aux = (g_mr_o)² = A² of the auxiliary amplifiers. They are implemented in Figure 4b using single-ended folded cascode inverting amplifiers formed by M_FCA1 and M_CA1 and by M_FCA2 and M_CA2. A modified flipped voltage follower is used to generate a very-low-impedance node V_G that operates as the signal ground (or reference node) for the input common source transistors M_FCA1 and M_FCA2 of the auxiliary amplifiers. Transistors M_BAT, M_FCA1, and M_FCA2 have the same W/L dimensions, the same quiescent current, and quiescent gate-source voltages. For this reason, the negative feedback loops of the auxiliary amplifiers shown in Figure 4a,b cause the gate voltages V_refX, V_X, and V_XP to have the same value. This results in equal drain-source voltages of the input and output mirror transistors and leads to a highly linear and accurate current mirror gain. On the other hand, the large gain boosting of the cascode transistors M_1C and M_2C provided by the folded cascode auxiliary amplifiers leads to an extremely low input resistance R_in = 1/(g_mA³)/2, which is lower by a factor of A_aux = A²/2 than the input resistance of the FVF mirror of Figure 2b and to an extremely high output resistance R_out = r_oA³ that is higher by the same factor A_aux than the output resistance of the mirrors of Figure 2. The value of V_refX (selected by the designer) sets the input voltage requirement V_in of the mirror. It must be higher than V_DSsat in order to keep the input and output mirror transistors in saturation. In the proposed design, V_refX was selected to have a value V_refX = 0.15 V, but it can also have been lower since input and output transistors had a value V_DSat = 0.06 V in the design discussed in Section 3. Remarks: (1) The FVF level shifter is a modified version of the basic FVF (or CONV_FVF). It has a resistor R in series with transistor M_BFVF. This resistor R in Figure 4a is used to generate a voltage drop that pulls down the voltage at node V_z and allows transistor M_BAT to have enough drain-source (V_DS > V_DSSat) voltage to operate in saturation. (2) The implementation of the auxiliary amplifiers using folded cascode amplifiers with a floating virtual ground node V_G in which the nominal voltage is set by the designer has the purpose of reducing the supply requirements of the circuit. (3) A distinctive characteristic of the proposed mirror is that the modified FVF with resistor R allows the quiescent value of V_G to be set to a value that is convenient to minimize the supply voltage and the input voltage requirements of the circuit. It also allows M_BAT to be maintained in saturation. Previous implementations of mirrors with auxiliary amplifiers (i.e., the regulated cascode mirrors discussed in [1]) required the source of the auxiliary amplifier’s input transistors to be connected to one of the supply rails and does not allow the supply requirements to be minimized or V_in to be set. (4) If required, the gain A_aux of the auxiliary amplifiers can be further boosted from a value A_aux = A²/2 in the circuit of Figure 4b to a value A_aux = A³/2 by utilizing double-cascoded auxiliary amplifiers. This would also boost the output impedance by an additional factor A/2 and decrease the input impedance of the mirror by the same factor. (5) The local negative feedback loops formed by the auxiliary amplifiers have only one high-impedance node at V_Y and V_YP. Compensation elements R_c and C_c are used to generate a dominant pole (and a zero) at these nodes. This is in order to compensate these loops and to prevent instability. (6) In order to reduce power dissipation, the auxiliary amplifiers and the FVF level shifters are biased with currents I_bias/k, which is a factor k times smaller than the bias current I_bias of the input and output mirror transistors M₁ and M₂. In the proposed design, a value k = 10 was used. This lead to a total quiescent current and power dissipation of the proposed mirror that is only 25% higher than the power dissipation of the mirrors of Figure 2. (7) The proposed current mirror can be easily transformed into a class AB mirror using the techniques reported in [11].

3. Simulation Results

3.1. Simulations of the High-Perfromance Class AB Follower HP_CSCFVF

The CONV_VF, CONV_FVF, CONV_CSCFVF, and proposed HP_CSCFVF circuits of Figure 1a–c and Figure 3 were simulated in a commercial 180 nm CMOS technology with dual rail voltages V_DD = 0.75 V, V_SS = −0.75 V (or V_supply = V_DD − V_SS = 1.5 V), I_bias = I_bCAS = 5 µA, R_bat = 55 kΩ, C_bat = 2 pF, I_bat = 2 µA, C_L = 100 pF, and W/L = 5/0.2 (µm) for all PMOS, and NMOS transistors, except the PMOS and NMOS transistors, implementing biasing sources that had dimensions W/L = 5/0.4 (µm), values Cc = 0.6 pF, and Rc = 75 kΩ were used. In order to save on silicon area, R_c was implemented with an NMOS transistor with W/L = 0.75/15 and with the gate connected to the positive rail V_DD. R_c and C_c were selected to provide a dominant pole ω_pdom = 1/R_xC_c at node V_X: and a high-frequency zero ω_z = 1/R_cC_c at V_x that approximately matches at the output pole ω_pout = g_mFVF/C_L at the output node V_oHPCSC of the FVF, as suggested by the design guidelines in [10]. Transistors M_source and M_sink were scaled up by factors 10 and 3, respectively. This was performed in order to equalize their dynamic output currents and to achieve symmetrical slew rates (SR+ and SR−). The total quiescent current and power dissipation of the proposed circuit were I_TotQ = 21 µA and P_dissQ = 31.5 µW, respectively. The small signal transconductance g_m and output conductance g_ds of the NMOS and PMOS unit transistors had the following values: g_mN = 148 µA/V, g_dsN = 2.94 µA/V, g_mP = 160 µA/V, and g_dsP = 2.2 µA/V.

Figure 5 shows the frequency responses of the CONV_VF, CONV_FVF, CONV_CSCFVF, and proposed HP_CSCFVF. The bandwidth of the HP_CSCFVF is 14.6 MHz, that of the CONV_CSCFVF is 3.47 MHz, that of the CONV_FVF is 2.5 MHz, and that of the CONV_VF is 0.347 MHz. Notice that the bandwidth of the proposed circuit is a factor almost 42 times larger than the bandwidth of the CONV_VF and 4.2 times larger than the CONV_CSCFVF. Figure 6a shows the pulse response of the proposed HP_CSCFVF and of the CONV_CSCFVF to a 1 MHz 0.9 V_pp pulse input. Figure 6b shows the corresponding load capacitor currents. It can be seen that the proposed circuit has close to symmetrical positive and negative peak output currents (and consequently slew rate) with the values I_outpk⁺ = 2.6 mA and I_outpk⁻ = 2.47 mA, respectively. Notice that the proposed circuit has peak output currents, which are a factor 118 times larger than the total quiescent current of the circuit. This corresponds to a very large current efficiency factor CE = I_outpk/I_TotQ = 118. The peak currents (and slew rates) of the conventional circuits is much lower due to their class A operation.

The positive and negative slew rates of the proposed circuit are SR+ = 34.47 V/µs and SR− = 34.03 V/µs (for C_L = 100 pF). Figure 7 shows the pulse response for various C_L values of 10 pF, 32 pF, 55 pF, 80 pF, and 100 pF. It can be seen that, in all cases, the pulse response has a only a small overshoot. Figure 8 shows the AC response of the output resistance of the proposed circuit. It has a very low value R_out = 2.11 Ω at low frequencies. The layout of the proposed design is shown in Figure 9. It occupies a 114 µm × 47 µm Si area.

Table 1. (A) Corner analysis at 27 °C; (B) corner analysis at 120 °C; (C) corner analysis at −20 °C.

(A)
Corner	tt	ff	fs	sf	ss	SD
ITotQ (µA)	21	22	21	22	21	0.49
f3dB (MHZ)	14.6	18	15.2	14.5	12.5	1.77
SR (V/µs)	24.3	28	21.5	25.6	21.7	2.4
Ioutpk (mA)	2.68	3.04	2.5	2.7	2.42	0.2
(B)
Corner	tt	ff	fs	sf	ss	SD
ITotQ (µA)	26	29	26	27	25	1.35
f3dB (MHZ)	15	18.5	15.2	14.8	12.7	1.86
SR (V/µs)	22.5	25.5	20.18	22.34	20.13	1.9
Ioutpk (mA)	2.45	2.73	2.38	2.42	2.22	0.16
(C)
Corner	tt	ff	fs	sf	ss	SD
ITotQ (µA)	20	20	20	19	19	0.49
f3dB (MHZ)	14.4	17.6	15.8	14.8	12.3	1.74
SR (V/µs)	26.6	29.3	22.9	25.8	22.8	2.4
Ioutpk (mA)	2.79	3.06	2.67	2.8	2.5	0.18

A–C show corner analysis of the proposed HP_CSCFVF at three different temperatures (27 °C, 120 °C, and −20 °C). It can be said that the proposed VF’s characteristic is very stable against process and temperature variations. The standard deviation (SD) of each parameter for variation in the process has been given in Table 1 for the considered temperatures (27 °C, 120 °C, and −20 °C).

The THD of the proposed circuit is 0.2% for a 0.5 V_pp 100 kHz input signal and 1% for a 0.5 V_pp 1 MHz sinusoidal input signal. The equivalent input noise power spectral density and RMS noise are 29 nV/√Hz and 130 µV_RMS. The small signal figure of merit is FOM_SS = 46 MHz·pF/µW, and the large signal current efficiency figure of merit is FOM_CE = I_outpk/I_TotQ = MIN{I_outpk+, I_outpk−}/I_TotQ = 118. The global figure of merit is $\mathrm{F}\mathrm{O}{\mathrm{M}}_{\mathrm{Global}}=\sqrt{\mathrm{F}\mathrm{O}{\mathrm{M}}_{\mathrm{S}\mathrm{S}}\mathrm{F}\mathrm{O}{\mathrm{M}}_{\mathrm{C}\mathrm{E}}}=73$.

Table 2. Comparison of the proposed HP_CSCVF with state-of-the-art work.

Parameter	Ref./Year [6]/2012		Ref./Year [12]/2016	Ref./Year [13]/2018	Ref./Year [14]/2016	Ref./Year [15]/2021	Ref./Year [16]/2018	CONV_VF Figure 1a	This Work Figure 3
Process technology (µm)	0.35	0.18	0.18	0.5	0.045	0.5	0.18	0.18	0.18
	Exp	Sim by Auth.	Sim	Exp	Sim	Exp	Sim	Sim	Sim
Supply (V)	3	±0.9	1.2	1.5	1.2	2	1.2	±0.75	±0.75
ITotQ (µA)	81	243	20.8	80	8.3	69	20	9	21
Load Cap. (pF)	20	50	10/100	50	1	47	1	100	100
BW (MHz)	5.8	3.65	15@100 pF	10	170	32	670.2	0.347	14.6
Ioutpk+ (mA)	1.62	3.16	0.32	1.8	0.17	1.59	0.116	0.085	2.6
Ioutpk− (mA)	1.67	3.16	NA	1.8	0.08	1.42	0.120	0.034	2.47
SR+ (V/µS)	79.4	63.2	32@10 pF	36	42	33.8	116.6	2.5	34.47
SR− (V/µS)	83.6	63.2	NA	36	50	30.3	120.5	12	34.03
Output resistance (Ω)	NA	NA	56	NA	1.15k	NA	144	1.2k	2.11
Quiescent power PdissQ (µWatt)	243	437	25	120	10	138	24	13.5	31.5
FOMCE = Ioutpk/ITotQ	20	12	15	22.5	10	20	5.8	3.7	118
FOMSS = BWxCL/PdissQ [(MHz)pF]/µW	0.47	0.42	60	4.16	17	10.9	28	2.5	46
FOMGlobal	3.06	2.24	30	9.7	13	15	12.7	3.04	73

shows a comparison of the performance characteristics of the proposed HP_CSCFVF with other voltage followers reported recently in the literature. It can be seen that the proposed HP_CSCFVF has the lowest output impedance, the highest small signal (FOM_SS), a large signal and current efficiency (FOM_CE), and global (FOM_Global) figures of merit in the table.

Figure 10 shows the Monte Carlo (MC) analysis of the proposed HP_CSCFVF power dissipation over 200 sample Monte Carlo simulation for process variation and mismatch. The mean quiescent power is 35.38 µW, and the standard deviation is 0.605 µW. From the corner analysis and Monte Carlo analysis, it can be ascertained that the proposed VF is robust against process variation, temperature, and mismatch effect.

3.2. Simulation Results for Low-Voltage High-Performance Current Mirror HP_CS_CM

The CN_CS_CM and proposed HP_CS_CM current-mirror circuits of Figure 2a and Figure 4 were simulated in a commercial 180 nm CMOS technology with a supply voltage of 1V and I_bias = 2 µA. The resistor used to implement R_bat in the FVF level shifter has a value 75 kΩ. It is implemented using a PMOS transistor. The W/L ratio of the PMOS and NMOS transistors used in the input and output stages of both current mirrors are W/L = 5 µm/0.4 µm. Transistors used in the auxiliary amplifiers and the FVF level shifter of the proposed current mirror are scaled down by factor k = 10. This was performed in order to reduce the quiescent power dissipation. The compensation elements had values C_c = 1.5 pF and R_c = 4 kΩ.

Figure 11 shows the frequency response of the proposed high-performance cascoded current mirror (HP_CS_CM). It has a 144 MHz bandwidth. Figure 12 shows the frequency response of the input impedance of the conventional cascode mirror CN_CS_CM and of the proposed HP_CS_CM. The CN_CS_CM has constant input impedance of 22 kΩ, whereas the input impedance of the proposed current mirror is 0.212 Ω up to 800 Hz and 8.9 kΩ at 118 MHz. The HP_CS_CM has an input impedance, at low frequencies, that is close to 10⁵ times lower than the CN_CS_CM.

Figure 13 shows the frequency response of the output impedance of the CN_CS_CM and of the HP_CS_CM. The proposed HP_CS_CM has an output impedance of 112 GΩ until 100 Hz, and the lowest output impedance is 18 MΩ through its bandwidth. On the contrary, the output impedance for the CN_CS_CM is 355 MΩ. Thus, the proposed high-performance current mirror has an output impedance that is 315 times higher than the CN_CS_CM at low frequencies.

Figure 14 shows the transient response to a triangular input current waveform. It can be seen that the proposed HP_CS_CM closely follows the linear input triangular current waveform. Figure 15 shows the transient pulse response of the proposed HP_CS_CM to a 10 MHz, 10 µA input pulse. It can be seen that the proposed HP_CS_CM has no peaking in the output pulse response.

An ideal current source is a circuit element that maintains a constant current flow independent of the voltage developed across its terminals as this voltage is determined by other circuit elements. To verify this property, Figure 16 shows the DC output transfer characteristics for dc voltage variation of 0 to 1 V by stepping the input current up from 0 to 10 µA in steps of 2.5 µA. It can be seen that the mirror compliance (minimum output) voltage for performance as a current source is approximately 0.2 V.

Figure 17 shows the input voltage of the CN_CS_CM and of the proposed HP_CS_CM by sweeping the input current I_in from o to 10 µA. It can be seen that the proposed HP_CS_CM has a constant 0.15 V input voltage, while the input voltage of the conventional mirror CN_CS_CM varies from 450 mV to 550 mV.

Figure 18 shows the gain error (in percent) e = 100[(I_out − I_in)/I_in] in the current transfer characteristic of the CN_CS_CM and the proposed HP_CS_CM as a function of the input current. It can be seen that, as expected, errors are similar since, in both mirrors, the drain-source voltages of input and output transistors are very similar. The total harmonic distortion of the proposed HP_CS_CM is given in

Table 3. Variation in THD with frequency for HP_CS_CM at different frequencies with 200 µA amplitude sinusoidal current.

Frequency (Hz)	THD (dB)
500	−60
10 k	−62
1 M	−60
100 M	−40

for a 200 µA amplitude sine wave at frequencies 500 Hz, 10 KHz, 1 MHz, and 100 MHz.

A Monte Carlo analysis with 200 samples was executed for some important parameters of the proposed HP_CS_CM for a 2 µA bias current.

Table 4. Summary of results of 200 samples Monte Carlo analysis of proposed current mirror’s parameter.

Parameter Name	Mean Value	Standard Deviation
Bandwidth (MHz)	144	0.789
Input resistance (dBΩ)	−13.7	0.728
Output Resistance(dBΩ)	221	0.505
Quiescent Power (µW)	5.33	0.066
Gain (A/A)	0.999	17.8 µ

gives the mean value and standard deviation of the bandwidth, input, output resistance, quiescent power, and gain for 200 runs. It can be see that the proposed current mirror is robust against process variation and mismatch effects. A noise analysis was also performed for the HP_CS_CM. The input referred noise was 14.5 pA/√Hz.

Table 5. Comparison of the proposed HP_CS_CM with state-of-the-art work.

Parameter	[17]	[18]	[19]	[20]	This Work
Input Compliance Voltage	520 m	39.6 m	-	-	150 m
Current Transfer error (%)	1.71	0.6	0.16	0.22	0.1
Input resistance (Ω)	21.43	496	68.3	130	0.212
Output Resistance (Ω)	1.14 G	1 M	10.5 G	9.5 G	112 G
Bandwidth (Hz)	6.17 G	181 M	402 M	2.7 G	144 M
Noise (pA/√Hz)	-	-	7.8	-	-
Supply (V)	1	0.9	1	1	1
Power (µW)	916.65	154	110	142.9	5
FOMCM (MHZ/µW)	6.73	1.17	3.6	18.89	28.8
Technology (µm)	0.18	0.18	0.18	0.18	0.18

compares the performance characteristics of the proposed HP_CS_CM to other low-voltage mirrors in the literature [17,18,19,20]. The proposed mirror has the highest output resistance and the lowest input resistance of all mirrors. Bandwidth is dependent on power dissipation. A mirror figure of merit FOM_CM = BW/P_diss is used to compare the circuits. Notice that the proposed mirror has the highest FOM_CM in the table (the input compliance voltage of the resistance based mirror in [18] is lower but it has the serious shortcoming that, with the reported 39.6 mV input voltage, it is subject in practice to very large random gain/linearity errors caused by mismatch in V_DS due to random offset of input and output transistors in the control circuit).

4. Conclusions

Two high-performance circuits based on the flipped voltage follower were introduced: (1) One was a class AB high-performance cascode flipped voltage follower that uses an additional output branch with a PMOS current-sourcing transistor. Replica biasing techniques are used to bias the sourcing transistor with a small quiescent current independent of the supply voltage. Under dynamic conditions, the sourcing transistor can provide very large positive output currents, which are over a factor 100 larger than the total quiescent current of the proposed circuit. Simulations in a commercial 0.18 µm CMOS technology have shown that it has low supply voltage requirements, greatly enhanced bandwidth, approximately symmetrical and large slew rates, and the largest small signal and large signal figures of merit of all class AB voltage followers. (2) The other was a low-voltage high-performance current mirror with 0.15 V input and 0.2 V output compliance voltages, 1 V supply voltage, extremely high output resistance (112 GΩ), extremely low input resistance (0.212 Ω), and the highest figure of merit.

This high-performance current mirror is implemented by utilizing two auxiliary amplifiers and a level shifter that boost the gain of the mirror cascode transistors and that equalize the drain source voltages of input and output mirror transistors. The auxiliary circuit increases the power dissipation of the mirror by only 25%. These characteristics were also verified with simulations in a commercial 0.18 µm CMOS technology.