Frequency Response of the Follower, the Common Gate and the Cascode
Measure why a follower can ring, which stages escape the Miller effect, and where the cascode hides a second pole.
You Will Be Able To
- Explain why a follower's output impedance rises with frequency when its source resistance is high.
- Measure the overshoot of a follower that drives a capacitor.
- Find the input capacitance of a stage from the gain across each of its capacitors.
- Choose the width of a cascode device from its internal pole.
- Lesson
- E9.6
- Time
- 35 min
- Runs on
- ngspice
- Check
- 5 questions
A source follower buffered a 1.2 V reference that had 100 kΩ of output resistance. On a 10 mV step into 1 pF the output overshot by 14.1 % and rang.
- from 1 kΩ
- from 100 kΩ
- Time 20 nsfrom 1 kΩ = 401.1 mV · from 100 kΩ = 398.7 mV
From 1 kΩ the output rises smoothly and never passes its final value.
- Time 33.92 nsfrom 1 kΩ = 401.7 mV · from 100 kΩ = 402.9 mV
From 100 kΩ the output peaks 1.17 mV above its final value. The step is 8.33 mV, so that is 14.1 %.
- Time 100 nsfrom 1 kΩ = 401.7 mV · from 100 kΩ = 401.7 mV
The ringing has a period of 48.4 ns, about 20.7 MHz, and then dies away.
Read the plot as a table
| Time | from 1 kΩ | from 100 kΩ |
|---|---|---|
| 0 s | 393.4 mV | 393.4 mV |
| 20 ns | 401.1 mV | 398.7 mV |
| 33.92 ns | 401.7 mV | 402.9 mV |
| 50 ns | 401.7 mV | 401.8 mV |
| 100 ns | 401.7 mV | 401.7 mV |
| 150 ns | 401.7 mV | 401.7 mV |
| 200 ns | 401.7 mV | 401.7 mV |
| 250 ns | 401.7 mV | 401.7 mV |
| 300 ns | 401.7 mV | 401.7 mV |
| 350 ns | 401.7 mV | 401.7 mV |
| 400 ns | 401.7 mV | 401.7 mV |
Run 3 of the lab. The transistor and the load are the same in both followers.
The same follower driven from 1 kΩ did not overshoot at all. The source resistance changed what the follower's output looks like.
The Follower Output Looks Like an Inductor
- Low frequency
Cgs passes no current, so the gate stays still. Push the output up and M1 cuts its current: 3.75 kΩ.
- High frequency, R_S = 1 kΩ
Cgs now conducts, but 1 kΩ still holds the gate. M1 answers as before.
- High frequency, R_S = 100 kΩ
Cgs drags the gate along with the output. Vgs barely changes, so M1 barely answers. The impedance rises.
Read the circuit as text
- VIN voltage source (1.2 V): + →
n1, − →0 - R_S resistor: pin 1 →
g, pin 2 →n1 - C_gs capacitor (96.5 fF): pin 1 →
g, pin 2 →out - M1 NMOS transistor (W/L = 10/2): gate →
g, drain →VDD, source →out - I_B current source (20 µA): + →
out, − →0 - C_L capacitor (1 pF): pin 1 →
out, pin 2 →0 - out port: terminal →
out
out= 0.393 V
The lab removes CL for the impedance run and puts it back for the step.
A follower has a low output impedance because M1 fights any change at its source. It can fight only while its gate holds still.
- RS = 1 kΩ
- RS = 100 kΩ
- Frequency 1 kHzRS = 1 kΩ = 3.752 kΩ · RS = 100 kΩ = 3.752 kΩ
At low frequency both read 3.75 kΩ. The formula 1/(gm + gmb + gds) gives 3.745 kΩ.
- Frequency 10 MHzRS = 1 kΩ = 3.751 kΩ · RS = 100 kΩ = 4.506 kΩ
At 10 MHz the 100 kΩ curve is rising, and its phase is +23.6°. Voltage leads current, as in an inductor.
- Frequency 100 MHzRS = 1 kΩ = 3.707 kΩ · RS = 100 kΩ = 12.69 kΩ
At 100 MHz the 100 kΩ follower shows 12.7 kΩ. The 1 kΩ follower still shows 3.71 kΩ.
Read the plot as a table
| Frequency | RS = 1 kΩ | RS = 100 kΩ |
|---|---|---|
| 1 kHz | 3.752 kΩ | 3.752 kΩ |
| 10 kHz | 3.752 kΩ | 3.752 kΩ |
| 100 kHz | 3.752 kΩ | 3.752 kΩ |
| 1 MHz | 3.752 kΩ | 3.76 kΩ |
| 10 MHz | 3.751 kΩ | 4.506 kΩ |
| 100 MHz | 3.707 kΩ | 12.69 kΩ |
| 1 GHz | 2.268 kΩ | 6.165 kΩ |
| 10 GHz | 621.7 Ω | 680.3 Ω |
| 100 GHz | 76.33 Ω | 77.45 Ω |
Run 2 of the lab. The test current is 1 A (AC), so volts read as ohms.
An impedance that rises with frequency is inductive. The follower contains no coil. RS and Cgs together delay the gate, and that delay acts like one.
The rise stops at 13.2 kΩ, well below RS. The body sits at ground, so gmb still answers a change at the source, whatever the gate does.
An Inductor and a Capacitor Ring
Put a capacitor on an inductive output and the two exchange energy. The gain peaks at one frequency, and a step rings near it.
- from 1 kΩ
- from 100 kΩ
- Frequency 1 kHzfrom 1 kΩ = -1.587 dB · from 100 kΩ = -1.587 dB
Both start at a gain of 0.833, which is −1.59 dB. The body effect keeps it below 1 (E3.5).
- Frequency 15.85 MHzfrom 1 kΩ = -2.209 dB · from 100 kΩ = -0.6593 dB
From 100 kΩ the gain climbs to 0.927 at 15.8 MHz: 0.93 dB of peaking. From 1 kΩ there is none.
- Frequency 40.45 MHzfrom 1 kΩ = -4.6 dB · from 100 kΩ = -9.491 dB
The 1 kΩ follower is 3 dB down at 40.5 MHz. The 100 kΩ follower got there at 30.2 MHz.
Read the plot as a table
| Frequency | from 1 kΩ | from 100 kΩ |
|---|---|---|
| 1 kHz | -1.587 dB | -1.587 dB |
| 10 kHz | -1.587 dB | -1.587 dB |
| 100 kHz | -1.587 dB | -1.587 dB |
| 1 MHz | -1.59 dB | -1.581 dB |
| 10 MHz | -1.846 dB | -1.034 dB |
| 15.85 MHz | -2.209 dB | -0.6593 dB |
| 40.45 MHz | -4.6 dB | -9.491 dB |
| 100 MHz | -10.04 dB | -25.8 dB |
| 1 GHz | -25.6 dB | -60.58 dB |
| 10 GHz | -42.1 dB | -81.98 dB |
Run 1 of the lab.
Peaking in the frequency response and overshoot in the step are one effect seen twice. Here 0.93 dB of peaking comes with 14.1 % of overshoot.
Four things decide the ringing together: RS, Cgs, gm and CL. Change one and check the step again.
Which Stages Escape the Miller Effect
E6.1 showed the Miller effect: a capacitor across a gain of −A looks (1 + A) times larger from the input. The rule is general. Look at the gain between the two plates of each capacitor.
| Stage (20 µA, W/L = 4/0.5) | Gain | Input capacitance | What the capacitors see |
|---|---|---|---|
| Common source | 9.707 | 21.88 fF | Cgd spans the gate and the drain: × (1 + 9.71) |
| Cascode | 9.695 | 13.43 fF | M1's drain is node x, which moves only 0.89 times |
| Common gate | 10.95 | 12.24 fF | The gate is fixed: no capacitor joins input and output |
| Follower | 0.8414 | 3.554 fF | Cgs sees 1 − 0.84 of the input: it shrinks |
From the netlist of the next figure, at 1 MHz. Loads are 40 kΩ; the follower has a 20 µA sink.
The two gate-input stages give Cgd = 0.96 fF and 11.6 fF that is not multiplied. In the common source, 0.96 fF acts like 10.3 fF.
The common gate has no capacitor from input to output. Its 12.2 fF sits beside a 3.65 kΩ input resistance, so its input pole is at 3.56 GHz.
- common source
- cascode (Miller-free)
- Frequency 100 kHzcommon source = 19.74 dB · cascode (Miller-free) = 19.73 dB
Both stages have a gain of 9.7, which is 19.7 dB.
- Frequency 71.01 MHzcommon source = 16.73 dB · cascode (Miller-free) = 18.38 dB
The common source is 3 dB down at 71.0 MHz. Its 21.9 fF input loads the 100 kΩ source.
- Frequency 117.2 MHzcommon source = 14.02 dB · cascode (Miller-free) = 16.71 dB
The cascode holds on to 117 MHz. Its input is only 13.4 fF.
Read the plot as a table
| Frequency | common source | cascode (Miller-free) |
|---|---|---|
| 100 kHz | 19.74 dB | 19.73 dB |
| 1 MHz | 19.74 dB | 19.73 dB |
| 10 MHz | 19.66 dB | 19.7 dB |
| 71.01 MHz | 16.73 dB | 18.38 dB |
| 100 MHz | 14.99 dB | 17.36 dB |
| 117.2 MHz | 14.02 dB | 16.71 dB |
| 1 GHz | -3.62 dB | -0.8607 dB |
| 10 GHz | -30.47 dB | -37.97 dB |
| 100 GHz | -52.85 dB | -63.33 dB |
The cascode gives 1.65 times the bandwidth from the same devices.
The Cascode Hides a Pole at Node x
The cascode pays for its small input capacitance with an extra node. Node x has a resistance near 1/(gm2 + gmb2) and a capacitance Cx.
Its pole is near (gm2 + gmb2)/(2πCx). A wider M2 raises gm2. It also adds its own Cgs and source junction to Cx.
- M2 2 µm
- M2 4 µm
- M2 16 µm
- M2 64 µm
- Frequency 1 MHzM2 2 µm = 4.463 kΩ · M2 4 µm = 3.285 kΩ · M2 16 µm = 2.185 kΩ · M2 64 µm = 1.825 kΩ
A wider M2 gives a lower resistance at x: 4.46 kΩ at 2 µm, 1.82 kΩ at 64 µm.
- Frequency 686.6 MHzM2 2 µm = 4.408 kΩ · M2 4 µm = 3.235 kΩ · M2 16 µm = 2.062 kΩ · M2 64 µm = 1.29 kΩ
The 64 µm node is already 3 dB down at 0.687 GHz. Its Cx is 127 fF.
- Frequency 4.328 GHzM2 2 µm = 3.155 kΩ · M2 4 µm = 2.191 kΩ · M2 16 µm = 902.3 Ω · M2 64 µm = 286 Ω
The 2 µm node holds until 4.33 GHz, with only 8.24 fF.
Read the plot as a table
| Frequency | M2 2 µm | M2 4 µm | M2 16 µm | M2 64 µm |
|---|---|---|---|---|
| 1 kHz | 4.463 kΩ | 3.285 kΩ | 2.185 kΩ | 1.825 kΩ |
| 10 kHz | 4.463 kΩ | 3.285 kΩ | 2.185 kΩ | 1.825 kΩ |
| 100 kHz | 4.463 kΩ | 3.285 kΩ | 2.185 kΩ | 1.825 kΩ |
| 1 MHz | 4.463 kΩ | 3.285 kΩ | 2.185 kΩ | 1.825 kΩ |
| 10 MHz | 4.463 kΩ | 3.285 kΩ | 2.185 kΩ | 1.824 kΩ |
| 100 MHz | 4.462 kΩ | 3.284 kΩ | 2.182 kΩ | 1.806 kΩ |
| 686.6 MHz | 4.408 kΩ | 3.235 kΩ | 2.062 kΩ | 1.29 kΩ |
| 1 GHz | 4.348 kΩ | 3.181 kΩ | 1.946 kΩ | 1.033 kΩ |
| 4.328 GHz | 3.155 kΩ | 2.191 kΩ | 902.3 Ω | 286 Ω |
| 10 GHz | 1.772 kΩ | 1.187 kΩ | 420.8 Ω | 125.2 Ω |
| 100 GHz | 194 Ω | 128 Ω | 43.62 Ω | 14.07 Ω |
Run 2 of the lab, with each output held by a voltage source.
| Width of M2 | Resistance at x | Cx | Pole |
|---|---|---|---|
| 1 µm | 6.389 kΩ | 6.357 fF | 3.919 GHz |
| 2 µm | 4.463 kΩ | 8.240 fF | 4.328 GHz |
| 4 µm | 3.285 kΩ | 12.50 fF | 3.875 GHz |
| 16 µm | 2.185 kΩ | 37.12 fF | 1.963 GHz |
| 64 µm | 1.825 kΩ | 127.0 fF | 686.6 MHz |
From the lab. Cx is 1/(2π · pole · resistance).
The pole has a best width, near 2 µm here. From 2 µm to 64 µm the resistance falls 2.4 times, but Cx grows 15 times.
M2 moves toward weak inversion as it widens, and there gm stops growing with width (E2.5). Its capacitance keeps growing.
- Device
- Measure
- Analysis
- Follower
Cgs ties the gate to the output. A high RS lets the gate move, and the output rings.
- Common source
Cgd sits across the full gain. The enlarged input capacitance loads the source resistance.
- Common gate and cascode
Neither multiplies a capacitor by its gain. The cascode pays with a pole at node x.
Read the diagram as text
- Follower: behind 100 kΩ
- Common source: gain 9.71
- Common gate: gate fixed
- Cascode: input only 13.4 fF
- Gate and output: peak at 15.8 MHz into 1 pF
- Input, the gate: 21.9 fF, 71.0 MHz behind 100 kΩ
- Input, the source: pole at 3.56 GHz
- Node x: pole at 4.33 GHz at best
- Follower → Gate and output: Cgs joins them
- Common source → Input, the gate: Cgd × (1 + 9.71)
- Common gate → Input, the source: 12.2 fF, 3.65 kΩ
- Cascode → Node x: Cx
Find the node first, then the resistance and the capacitance on it.
Asked in interviews
Why can a source follower ring?
Its output impedance is 1/gm only while the gate is held. With a high source resistance, Cgs lets the gate follow the output at high frequency. The impedance then rises with frequency, like an inductor. With a load capacitor that makes a resonance. Here 100 kΩ and 1 pF gave 14.1 % of overshoot.
Which single stages suffer from Miller multiplication?
Only a stage with a capacitor between its input and an inverting, high-gain output: the common source. The common gate has its gate at AC ground. The cascode keeps the drain of its input device nearly still. In a follower, Cgs sees 1 minus the gain, so it shrinks.
Where is the second pole of a cascode, and how do you size the cascode device?
At the node between the two devices, near (gm2 + gmb2)/Cx. A wider cascode device has more gm but adds its own capacitance to that node. Sweep the width: here the pole peaked at 4.33 GHz for 2 µm and fell to 0.687 GHz for 64 µm.
Try It
Run 1 drives both followers and reads the gain into 1 pF. peak_100k is the top of the peaking.
Run 2 turns the input off and removes both loads with alter. It pushes 1 A (AC) into each follower output and into node x of each cascode. Every voltage then reads as an impedance.
Run 3 puts the loads back and applies a 10 mV step. The overshoot is the peak above the final value, as a percentage of the output step.
Predict before each change. Set RSB to 10k: does the follower still ring? Set CLB to 0.2p, then to 5p. Set IBB to 5u.
Run the netlist to draw its plots here.
cgs = 9.646508e-14 gain_lf = 8.329780e-01 peak_100k = 9.269041e-01 at= 1.584893e+07 f3_1k = 4.045048e+07 f3_100k = 3.015312e+07 zout_hand = 3.745223e+03 zout_lf = 3.751582e+03 z1k_100meg = 3.706894e+03 z100k_100meg = 1.268673e+04 z100k_max = 1.319306e+04 at= 1.584893e+08 ph_10meg = 2.362964e+01 rx_1 = 6.388606e+03 fx_1 = 3.918758e+09 rx_2 = 4.462876e+03 fx_2 = 4.327899e+09 rx_4 = 3.285095e+03 fx_4 = 3.875468e+09 rx_16 = 2.184605e+03 fx_16 = 1.962754e+09 rx_64 = 1.824564e+03 fx_64 = 6.866445e+08 cx_1 = 6.357196e-15 cx_2 = 8.240019e-15 cx_4 = 1.250109e-14 cx_16 = 3.711772e-14 cx_64 = 1.270367e-13 pk_1k = 4.017341e-01 at= 2.424250e-07 end_1k = 4.017341e-01 pk_100k = 4.029060e-01 at= 3.392500e-08 end_100k = 4.017341e-01 os_1k = 0.000000e+00 os_100k = 1.406497e+01 t_ring = 4.841619e-08 targ= 7.454910e-08 trig= 2.613291e-08Check Yourself
Common Mistakes
Wrong Treat a source follower as a buffer that is safe behind any source.
Behind a high resistance its output turns inductive, and a load capacitor then makes it ring.
Fix Simulate a step with the real source resistance and the real load. This follower was clean from 1 kΩ and overshot 14.1 % from 100 kΩ.
Wrong Multiply every capacitor of every stage by 1 + A.
The factor depends on the gain between the two plates of that capacitor, and that gain differs from stage to stage.
Fix Find both plate voltages first. A common gate multiplies nothing, and a follower shrinks its Cgs.
Wrong Make the cascode device as wide as the layout allows, to push its pole out.
Its gm stops growing with width while its capacitance keeps growing, so the pole comes down.
Fix Sweep the width and read the pole. Here 2 µm gave 4.33 GHz and 64 µm gave 0.687 GHz.
Wrong Measure the follower's output impedance with the input source still driving.
The reading then mixes the gain of the stage into the impedance.
Fix Turn the input's AC value off and push a test current into the output.
alter ITB ac=1
ac dec 40 1k 100galter VIN ac=0
alter ITB ac=1
ac dec 40 1k 100gRemember
- A follower's output impedance is low only while its gate is held. Behind 100 kΩ it rose from 3.75 kΩ to 13.2 kΩ.
- A rising impedance is inductive. With a load capacitor it rings: 14.1 % overshoot into 1 pF here.
- The Miller factor comes from the gain across each capacitor. Only the common source multiplies Cgd by its full gain.
- The cascode's input stays small (13.4 fF against 21.9 fF), which gave 1.65 times the bandwidth behind 100 kΩ.
- The cascode adds a pole at node x. It has a best cascode width: 2 µm here.