Application Note 1686 Authors: Jian Wang and Tamara Schmitz Development of a Spice Op-Amp Macro Model for Current-Feedback Amplifiers Current-feedback amplifiers (CFAs) are the high-speed relatives of more common voltage-feedback amplifiers (VFAs). CFAs have wider bandwidths and faster slew rates. Applications such as DSL rely on their fast and strong output drives. Models are important because they allow engineers to test designs before they go through the time-intensive and costly process of building a working prototype. In this application note, we introduce you to a circuit model for a currentfeedback amplifier. Since it would take far too long to simulate every nuance of a complete design, this macro model simulates the most common effects, such as transient response, frequency response, voltage noise and output slew rate limiting. Detailed descriptions of each stage in the model will be presented with examples of model performance and correlation to actual device behavior. Figure 1 shows the conceptual approach of the current feedback amplifier. CFAs have a unity gain buffer to force the inverting input (Vin-) to follow the non-inverting input (Vin+). This topology is very different from the high impedance inputs of the voltage-feedback amplifier. In the CFA case, the inverting input node shows the low input impedance (Zin-) from the output of the buffer. The error signal is the current (I) flowing into or out of the inverting node. The error current is converted by a large transimpedance, Z, to the output voltage. Rf is used to control the feedback current. This is another major deviation from the voltage-feedback topology. In VFAs, the feedback network (Rf and Rg) primarily set the voltage gain. In CFAs, the feedback network does set the gain, but the value of Rf also controls the bandwidth of the amplifier. Vin+ G=1 Vout Zin- +- INPUT STAGE GAIN STAGE FREQUENCY SHAPING STAGES OUTPUT STAGE NOISE MODULE FIGURE 2. THE BLOCK DIAGRAM OF A CFA MACRO MODEL The Input Stage As stated, the biggest difference between a CFA and a VFA is the input stage. Figure 3 shows an example, the input stage of a CFA model. Four bipolar transistors are included, Q1 to Q4. The effective output of this stage is at nodes 11 and 12, which are coupled into the gain stage by using voltage controlled current source. The bias current I1-I2 of Q1-Q2 should be set by: I1 = I 2 = kT 2q • Z in − i.e. Z in − = 1 2gm (EQ. 1) The impedance at the inverting input, Zin-, can be measured. In this example, I1 = I2 = 85µA. R1, R2, C1, C2 are used to fit the frequency response and to control the input stage slew rate. There are many other components used in the input block. Cs1 and Cs2 are the inverting input capacitance. Cin1 is the noninverting input capacitance. The input bias current is modeled by current sources Ib1 and Ib2. The input offset voltage is modeled by the voltage source Vos. The current source Gb1 is used to model the input current common mode rejection at the inverting input. Z•I I Vin- Rf Rg FIGURE 1. THE BLOCK DIAGRAM OF A CURRENT FEEDBACK AMPLIFIER A five-stage model represents the actual circuit and the block diagram is shown in Figure 2. These five basic blocks are the input stage, the gain stage, the frequency-shaping stage, the output stage and the noise module. December 21, 2011 AN1686.0 1 FIGURE 3. INPUT STAGE OF A CFA (EL5165) CAUTION: These devices are sensitive to electrostatic discharge; follow proper IC Handling Procedures. Copyright Intersil Americas Inc. 2011. All Rights Reserved. 1-888-INTERSIL or 1-888-468-3774 | Intersil (and design) is a trademark owned by Intersil Corporation or one of its subsidiaries. All other trademarks mentioned are the property of their respective owners. Application Note 1686 The Gain Stage This stage is similar to the gain stage of a VFA. It performs many important functions. 1. This stage sets the open loop trans-impedance of the part. 2. It provides output slew rate limiting. 3. It contributes the dominant pole to the AC characteristic. 4. It level shifts the signal from two voltages referred to the supplies to a single voltage referred to the mid-point. 5. It limits the output voltage swing. Taking a closer look at the components of the gain stage in Figure 4, slew rate limiting is set by limiting the current to C3 and C4 in Figure 4. The current limiting is set in the input stage by clamping the voltage across R3 and R4 shown in Figure 3. This voltage is decided by voltage sources V1 and V2 and diodes D1 and D2. R7-C3 and R8-C4 decide the dominant pole of this model. D3-D4 and V5-V6 are used to control the output clamping voltage. FIGURE 5. HIGHER ORDER POLE STAGES Noise Module The input current noise of the current feedback op amp model can't be neglected. It is described as: pA / Hz Voltage noise is also important, so both a voltage noise module and a current noise module are needed for our CFA model. For the right noise analysis, one trick called "noiseless resistors" can be used at the input stage. All the resistors in the input stage should be substituted by voltage controlled current sources (Device G in SPICE) where input and output terminals are connected together and the gm is set to the reciprocal of the required resistance. We will use two pieces to construct the total noise model. The noise module of Figure 6 generates 1/f and white noise by using a 1.5V voltage source biasing a diode-resistor series combination. White noise is generated by the thermal noise-current generated in the material of the resistors. in2 = FIGURE 4. GAIN STAGE Frequency-Shaping Stages The frequency-shaping stages of a CFA model are very similar to that of a VFA. Each frequency-shaping block provides unity gain, so it is easy to add more poles and zeros. For more information of the frequency-shaping stages, see [Reference 2]. For our example op amp, only three pole stages are used, which is shown in Figure 5. E3 is used to set the reference level at the middle of the supplies, V+ and V-. where k is the Boltzmann’s constant 4kT R (EQ. 2) So the required value of the resistor for a given noise-voltage spectral density is: en2 R= 2 × 4kT where en is the spectral density of the white noise voltage (EQ. 3) Flicker noise, also called 1/f noise, refers to the noise exhibiting power spectral density inversely proportional to the frequency. More generally, this noise has a spectral density of: SN ∝ where B > 0 1 f (EQ. 4) β As a data point, the frequency where the flicker noise curve crosses the white noise curve is defined as the corner frequency. The small amount of flicker noise that remains is modeled within the SPICE diode model. Referring to Figure 6, I dAF i = 2qI d + KF • frequency (EQ. 5) 2 n 2 AN1686.0 December 21, 2011 Application Note 1686 where Id is the DC diode current. AF and KF are the model parameters of the SPICE diode and q is the charge of the electron. The flicker noise exponent (AF) is set to 1 and the flicker noise coefficient (KF) is set: KF = Ea2 2R 2 • I d where Ea is the noise-voltage spectral density at 1Hz. (EQ. 6) The simulated voltage noise will show the 1/f noise-voltage spectral density with the correct corner frequency. The noise module in Figure 7 only simulates the white noise portion of the current noise by utilizing thermal noise of two parallel resistors. FIGURE 8. OUTPUT STAGE Simulation Results The simulation results of the macro model should be compared with the real world device to verify its functionality. The example op amp used in this application note is the EL5165, a high speed current feedback amplifier. Some SPICE simulation results are compared with the measured results. The gain plots in Figure 9 shows a remarkable similarity between our model on the left at a gain of two and measurements from the EL5165 on the right. The measurements from the device are taken at both unity gain and a gain of two. There is less than 1% error between the -3dB frequency of the two gain-of-two measurements. FIGURE 6. NOISE VOLTAGE MODULE FIGURE 9. AC RESPONSE FOR GAIN = +2 OF THE MACRO MODEL (LEFT) AND EL5165 (RIGHT) (Rf = 499Ω) FIGURE 7. NOISE CURRENT MODULE At a gain of three, the discrepancy increases a bit, as shown in Figure 10. Now the error in cut-off frequency is within 5%. Output Stage After the frequency shaping-stages, the signal appears at Node VV5, which is referenced to the midpoint of the two supply rails. Each controlled source can generate enough current to support the desired voltage drop across its parallel resistor. R13 and R14 are equal to twice the open loop output resistance, so their parallel combination gives the correct Zout. D5-D8 and G9-10 are used to force a current from the positive rail to the negative rail to correct the real current sink or source. G11-12 drive the output. 1 G9 = G10 = G11 = G12 = 2 Z out (EQ. 7) (EQ. 8) R13 = R14 = 2Z out 3 FIGURE 10. AC RESPONSE FOR GAIN = +3 OF THE MACRO MODEL (LEFT) AND EL5165 DEVICE (RIGHT) (Rf = 249Ω) Since the feedback resistor controls the bandwidth, we offer a comparison of frequency response for a range of resistors in Figure 11. The values of resistance are listed on the plot on the right that comes from the data sheet. The left plot is from our macro model. Again, the matching of -3dB frequency is within 5%. AN1686.0 December 21, 2011 Application Note 1686 Several of Intersil's current feedback amplifiers use the same model topology with different internal model values. EL5165 Macro Model Netlist .subckt EL5165 3 2 7 4 6 * FIGURE 11. FREQUENCY RESPONSE WITH DIFFERENT RF FOR THE MODEL (LEFT) AND THE DEVICE (RIGHT) After examining the small-signal response versus frequency, the next logical step is to inspect the large-signal response, as shown in Figure 12. Here we look at the slew rate of the step response with a gain of two driving a load of 150Ω from a ±5V supply. While the time base (x-axis) is at different resolutions in the two plots, the slew rate of the macro model shown on the left is 5000V/µs. This agrees to 5% from the slew rate on the output signal of the EL5165 on the right. FIGURE 12. LARGE-SIGNAL STEP RESPONSE FOR THE MACRO MODEL (LEFT) AND FROM THE DEVICE DATA SHEET (RIGHT) After frequency response and slew rate, our next goal is to match the noise performance. We have chosen to demonstrate the voltage noise, so the model in Figure 6 is included, but not the current noise model from Figure 7. The voltage noise model exhibits the spectrum shown in the plot on the left and has a value of 2.15 nV / Hz at 1MHz The datasheet curve for voltage noise is on the right and has a value of: 2.1 nV / Hz at 1MHz. *Input Stage C_Cin1 V_Vos 0 3 1.8p 1 N4150175 1.5mVdc I_I1 8 4 DC 85uAdc I_I2 7 5 DC 85uAdc I_Ib1 7 1 DC 2uAdc I_Ib2 7 2 DC 2uAdc G_Gi1 8 10 8 10 0.001 G_Gi2 5 9 5 9 0.001 G_Gi3 12 4 12 4 0.01 G_Gi4 7 11 7 11 0.01 G_Gb1 7 2 1 16 0.0000001 Q_Q1 7 1 8 Inpn Q_Q2 11 9 2 Inpn Q_Q3 4 1 5 Ipnp Q_Q4 12 10 2 Ipnp D_D9 19 11 DX D_D10 12 18 DX V_V7 7 19 1.5dc V_V8 18 4 1.5Vdc C_C1 9 7 0.03p C_C2 4 10 0.03p C_Cs1 2 7 0.25p C_Cs2 4 2 0.25p I_I3 7 4 DC 3mAdc *Gain stage FIGURE 13. INPUT NOISE VOLTAGE VERSUS FREQUENCY FOR MACRO MODEL (LEFT) AND DEVICE DATA SHEET (RIGHT) Conclusion A more comprehensive SPICE macro model for a current feedback amplifier is developed. This macro model includes effects such as transfer response, accurate AC response, DC offset and voltage noise. It is convenient to use such a model and change the parameters to fit other current feedback amplifiers. 4 C_C3 VV2 7 0.405p C_C4 4 VV2 0.405p D_D3 13 7 DX D_D4 4 14 DX G_G1 7 VV2 7 11 0.01 G_G2 VV2 4 12 4 0.01 R_R7 VV2 7 10meg R_R8 4 VV2 10meg R_R17 13 0 1G AN1686.0 December 21, 2011 Application Note 1686 R_R18 14 0 1G R_R21 0 101 55 0 103 55 V_V5 13 VV2 1.6Vdc R_R22 V_V6 VV2 14 1.6Vdc *Current noise *High-order poles R_R23 0 NI2 0.022 E_E3 16 4 7 4 0.5 R_R24 0 NI2 0.022 C_C5 VV3 7 0.0032p R_R25 0 NI1 0.022 C_C6 4 VV3 0.0032p R_R26 0 NI1 0.022 C_C7 4 VV4 0.0009p G_Gn1 3 0 NI1 0 1 C_C8 VV4 7 0.0009p G_Gn2 2 0 NI2 0 1 C_C9 4 VV5 0.0001p * C_C10 R_R9 VV5 7 0.0001p * Models VV3 7 100k * R_R10 4 VV3 100k .model Ipnp pnp(is=1e-15 bf=1E9 VAF=65) R_R11 VV4 7 100k .model Inpn npn(is=1e-15 bf=1E9 VAF=65) R_R12 4 VV4 100k .model Iden d(kf=100e-14 af=1) R_R15 VV5 7 100k .MODEL DY D(IS=1E-20 BV=50 Rs=1) R_R16 4 VV5 100k .MODEL DX D(IS=1E-15 Rs=1) G_G3 4 VV3 VV2 16 0.00001 G_G4 7 VV3 VV2 16 0.00001 G_G9 7 VV4 VV3 16 0.00001 G_G10 4 VV4 VV3 16 0.00001 G_G11 7 VV5 VV4 16 0.00001 G_G12 4 VV5 VV4 16 0.00001 .ends EL5165 *Output stage G_G5 15 4 6 VV5 0.0001 G_G6 17 4 VV5 6 0.0001 G_G7 7 6 7 VV5 -0.04 G_G8 6 4 VV5 4 -0.04 R_R13 6 7 25 R_R14 4 6 25 D_D5 7 15 DX D_D6 7 17 DX D_D7 4 15 DY D_D8 4 17 DY *Voltage noise E_EN N4150175 3 101 103 1 V_V15 102 0 0.5Vdc V_V16 104 0 0.5Vdc D_DN1 102 101 Iden D_DN2 104 103 Iden 5 AN1686.0 December 21, 2011 Application Note 1686 References [1] Derek Bowers, Mark Alexander, Joe Buxton, "A Comprehensive Simulation Macromodels for 'Current Feedback' Operational Amplifiers,” IEEE Proceedings, Vol. 137, April 1990 pp.137-145. [2] Mark Alexander, Derek Bowers, "AN-138 SPICE-Compatible Op Amp Macro-Models", Analog Devices Inc., Application Note 138. [3] "AN-840 Development of an Extensive SPICE Macromodel for 'Current-Feedback' Amplifiers”, National Semiconductor Corp., Application Note 840. [4] “Current Feedback Amplifier Theory and Applications”, Intersil Corp., Application Note 9420. Intersil Corporation reserves the right to make changes in circuit design, software and/or specifications at any time without notice. Accordingly, the reader is cautioned to verify that the Application Note or Technical Brief is current before proceeding. For information regarding Intersil Corporation and its products, see www.intersil.com 6 AN1686.0 December 21, 2011