A source follower buffers a reference behind 100 kΩ and overshoots 13.9 % into 2 pF. Find the cause and fix the follower without making it slow.
Circuit context: A source follower that buffers a 1.2 V reference behind a 100 kΩ filter resistor and drives the 2 pF input of a converter. Repair a reference buffer that rings each time its input steps.
A sensor chip filters its 1.2 V reference with a 100 kΩ resistor. A source follower then drives the 2 pF input of a converter. Each time the reference steps, the follower output overshoots by 13.9 % and rings.
Two fixes were tried on the bench. Extra capacitance on the output left 9.2 % of overshoot and made the buffer slow. A larger series resistor at the gate cut the overshoot to 3.3 % and the bandwidth to 8.3 MHz.
Fix the subcircuit buf. The grader keeps the source, the 100 kΩ and the 2 pF. A 10 mV step must overshoot by at most 5 % at tt, ss and ff. At tt the buffer must also keep 30 MHz of bandwidth and a gain of 0.8. Its supply current must stay at 45 µA or less.
Start with the step response and the size of the transistor.
What the Grader Measures
Your subcircuit buf (ports in, out, vdd) under the grader's own bench, one run per corner (tt, ss, ff):
| Check | Corner | Rule |
|---|---|---|
| Overshoot | tt | ≤ 5% |
| −3 dB bandwidth | tt | ≥ 30 MHz |
| Gain at 1 kHz | tt | ≥ 0.8 |
| Supply current | tt | ≤ 45 µA |
| Overshoot | ss | ≤ 5% |
| Overshoot | ff | ≤ 5% |
Full specification
.subckt buf in out vdd
* SKY130 1.8 V FETs (sky130_fd_pr__nfet_01v8, sky130_fd_pr__pfet_01v8), resistors, capacitors, DC current sources
.ends
| Check | Corner | Rule | How the grader measures it |
|---|---|---|---|
| Overshoot | tt, ss, ff | ≤ 5 % | tran: peak above the final value, as a share of the output step, for a 10 mV input step |
| −3 dB bandwidth | tt | ≥ 30 MHz | ac: where the gain falls to 0.7071 of its 1 kHz value |
| Gain at 1 kHz | tt | ≥ 0.8 | ac: |v(out)| with AC 1 on the source |
| Supply current | tt | ≤ 45 µA | op: the current out of the 1.8 V source |
The 100 kΩ source resistance and the 2 pF load belong to the bench. You cannot change them, only what stands inside buf.
The grader runs its bench once per corner, at 27 °C.
Examples
Example 1
The Starter: W/L = 20/2 at 40 µA- Design
.subckt buf in out vdd * a large device was chosen "for low noise": W/L = 20/2 at 40 uA XM1 vdd in out 0 sky130_fd_pr__nfet_01v8 W=20 L=2 IB out 0 40u .ends
- Measured by the grader
Overshoot: 13.91% (needs ≤ 5%) Overshoot at ss: 13.91% (needs ≤ 5%) Overshoot at ff: 13.91% (needs ≤ 5%) −3 dB bandwidth: 21.33 MHz (needs ≥ 30 MHz) Gain at 1 kHz: 0.8329 (needs ≥ 0.8) Supply current: 40 µA (needs ≤ 45 µA)
The output peaks 13.9 % above its final value and the bandwidth is 21.3 MHz. The corners change almost nothing: the ringing comes from the circuit around the device.
Example 2
More Load: 8 pF Added at the Output- Design
.subckt buf in out vdd CX out 0 8p XM1 vdd in out 0 sky130_fd_pr__nfet_01v8 W=20 L=2 IB out 0 40u .ends
- Measured by the grader
Overshoot: 9.188% (needs ≤ 5%) −3 dB bandwidth: 8.871 MHz (needs ≥ 30 MHz) Gain at 1 kHz: 0.8329 (needs ≥ 0.8) Supply current: 40 µA (needs ≤ 45 µA)
The resonance moves down in frequency and still rings, with 9.2 % of overshoot. The bandwidth falls to 8.9 MHz. Much more capacitance would stop the ringing and leave only 3 MHz.
Example 3
A Narrower Device: W/L = 4/2- Design
.subckt buf in out vdd XM1 vdd in out 0 sky130_fd_pr__nfet_01v8 W=4 L=2 IB out 0 40u .ends
- Measured by the grader
Overshoot: 1.854% (needs ≤ 5%) −3 dB bandwidth: 21.51 MHz (needs ≥ 30 MHz) Gain at 1 kHz: 0.8185 (needs ≥ 0.8) Supply current: 40 µA (needs ≤ 45 µA)
A fifth of the width has about a fifth of the gate capacitance, and the overshoot falls to 1.9 %. But the narrow, long device has less than half the g_{m}, so the bandwidth stays at 21.5 MHz.
Constraints and Contract
- The netlist defines .subckt buf in out vdd ... .ends exactly once, with these ports in this order.
- Inside buf: X instances of sky130_fd_pr__nfet_01v8 or sky130_fd_pr__pfet_01v8, resistors, capacitors, DC current sources, comments and .param lines. No voltage sources.
- The grader writes the .lib line for each corner itself. Everything outside buf is replaced by its bench.
Hints
Hint 1
A follower fights changes at its output only while its gate is held. Which capacitor lets the output move the gate, and what stops the 100 kΩ source from holding it?
Hint 2
Read the gate capacitance of the device from the operating point (@m.x1.xm1.msky130_fd_pr__nfet_01v8[cgg]). Compare it with the 2 pF load and with what a smaller device would have.
Hint 3
Width and length do not act alike. Both change Cgs, but they move gm in opposite directions, and the bandwidth into 2 pF needs gm.
What the grader checks: The grader places buf under its own bench: 1.8 V, a 1.2 V source behind 100 kΩ, and 2 pF. Hidden measurements read the overshoot of a 10 mV step, the bandwidth, the gain and the supply current. The corners are tt, ss and ff at 27 °C. Engine: ngspice 42 (libngspice0 42+ds-3build1) with the pinned SKY130 models, measured 2026-11-12 with tools/spice-judge.mjs. The grader does not check large steps, noise, the input range or other loads.
Run simulates your netlist as written, with your own analyses and measurements. Submit sends it to the grader, which keeps only the subcircuit buf and makes 3 ngspice runs. The numbers come from ngspice and the SKY130 models the site pins; they prove the measured cases, not every condition a chip meets.
See all 1 steps and the final capstone