Electronics Design AUTPS63020 output drooping 200 mV and ringing as battery drains through 3.3 V — mode transition issue?
Asked by stale_biscuit_03 ·
Building a battery-powered sensor logger — single Li-ion cell (3.7V nominal, 2.8–4.2V range during use). Using a TPS63020 to regulate down to 3.3V. Layout is pretty closely based on the TI evaluation board: 4.7µH Würth inductor, two 22µF 0402 X5R caps on the output, 10µF + 1µF on the input.
Works fine when the battery is fresh off the charger (4.0–4.2V) and also works fine when it's mostly depleted (under about 3.0V). The problem is the middle. As the battery drains through roughly 3.3–3.6V, the 3.3V output sags about 200mV and I can see ringing on the scope. Worse under higher load (~200mA), and it's resetting the MCU occasionally, which is how I caught it.
Things I've tried so far:
- Added a 100µF bulk electrolytic on the output — helped a little but didn't fix it
- Tried different loads (80mA to 250mA): worst around 200mA in that voltage range
- Verified the BST cap (100nF per datasheet) and checked layout against the eval board
- Switching node looks clean above and below the problem region on the scope
Is the instability around Vin ≈ Vout a known limitation of this part, or is there something specific in my design causing it? I chose the TPS63020 specifically because it's supposed to handle this voltage crossing gracefully.
3 Replies
The mode transition region is genuinely harder to regulate than pure buck or pure boost, but what you're describing — 200mV sag and ringing, load-dependent, specifically in the 3.3–3.6V input range — points to output capacitor effective value more than a controller problem.
What's happening in the transition region
In pure buck mode, the TPS63020 runs the buck-side switches and keeps the boost-side high switch on permanently. In pure boost mode, it's the reverse. In the crossover region, all four switches transition to active regulation simultaneously. Duty cycle on both sides approaches its limit — buck side near 100%, boost side approaching 0% — and the headroom for the feedback loop to correct transients narrows significantly. Any sudden load demand in that window hits the output capacitance directly.
Your output caps are probably not 44µF in practice
Two 22µF 0402 X5R caps looks right from the datasheet recommendation. The problem is that 22µF in 0402 nearly always comes with a 4V or 6.3V voltage rating — the only voltage ratings achievable at 22µF in that package. At 3.3V DC bias, a 4V-rated X5R 0402 is at roughly 80% of rated voltage. X5R ceramics derate significantly with DC bias; at that bias level you'd typically see 35–55% of nominal capacitance remaining, depending on manufacturer and specific part. Your two "22µF" caps might be delivering somewhere around 15–20µF total in circuit — well under what's needed for stable operation in the crossover region.
Check your specific parts using Murata's SimSurfing or TDK's SEAT tool. Plug in the part number and 3.3V applied voltage and you'll see the actual capacitance.
Fix 1: Upgrade the output caps
Use 22µF X5R in 0805 with a 10V voltage rating. At 3.3V applied, a 10V-rated 0805 X5R typically retains around 65–75% of nominal — still derated, but two parts give you roughly 28–33µF actual, which is a meaningful improvement. Run the actual part numbers through the manufacturer tools to confirm before committing to the BOM.
Fix 2: Check the PS/SYNC pin
If PS/SYNC is floating or pulled low, the TPS63020 operates in power-save mode: it drops to PFM at light loads and switches back to PWM as load rises. That PWM/PFM transition can overlap with the buck/boost mode transition in the 3.3–3.6V battery range, stacking two sources of instability at the same operating point. Tie PS/SYNC high (to 3.3V output) to force PWM-only operation and see whether the sag improves. It costs you the light-load efficiency improvement from power-save mode, but it isolates the variable cleanly.
The DC-DC topology selection guide explains why this crossing point is the hard case for Li-ion → 3.3V conversion and what the alternatives look like if you need tighter regulation than a 4-switch converter can reliably deliver through transition.
Worth being specific about the derating numbers because this catches people every time.
X5R ceramics have a DC bias characteristic that drops steeply above roughly 50% of rated voltage. The exact shape varies by manufacturer, case size, and dielectric lot, but the general range for 0402 X5R at near-rated voltage is 30–55% of nominal. Murata, TDK, Yageo, and KEMET all publish capacitance-vs-DC-bias curves for their parts — Murata's SimSurfing and TDK's SEAT tool give you the in-circuit value for a specific part number at a specific applied voltage.
For 3.3V output rails I won't spec anything with a voltage rating below 10V for the output ceramics, and I'll go to 16V-rated if the package allows. At 3.3V on a 16V-rated X5R 0805, you're at about 20% of rated voltage — derating is minimal, typically 85–95% of nominal retained. At 3.3V on a 4V-rated part you're at 82% of rated voltage, which is where the curve gets steep and you lose more than half your capacitance.
The 100µF electrolytic you added helping is consistent with this diagnosis. Electrolytics don't derate with DC bias, so that bulk capacitance was genuinely stable. It wasn't enough to fully close the gap in the crossover region, but the direction of improvement confirms the problem is effective output capacitance, not the controller.
Practical rule: wherever you have a fixed-voltage DC output in a switching converter, run the output cap part numbers through the manufacturer derating tool before signing off the BOM. Nominal value is almost irrelevant for stability analysis; the in-circuit derated value is what the converter actually sees.
Tied PS/SYNC high first since it was a zero-part fix. The ringing is noticeably reduced but the sag is still there, so the power-save mode was contributing but isn't the whole story.
Ran the 0402 cap part numbers through SimSurfing — both rated 4V, and at 3.3V applied I'm getting 9.8µF and 10.4µF out of the two parts. About 20µF total instead of the 44µF nominal I thought I had. Exactly what you described.
Ordered 22µF 10V 0805 X5R parts to replace them, and also grabbing a 47µF polymer electrolytic in case I want belt-and-suspenders. Will report back once I've reflowed the new parts.
I'd seen DC bias derating mentioned before but didn't realise how steep it gets at that voltage ratio. Good thing I caught it at bring-up and not in the field.