When selecting a step-down converter for ripple-sensitive circuits, picking solely by rated current and efficiency is a recipe for trouble. Converters that engage Eco-mode or pulse skipping at light loads can see output ripple soar to tens of mV, severely degrading the performance of analog sensors, RF front-ends, and PLL/VCO supply rails. Texas Instruments’ TPS54202 and TPS54308 share the same 6-pin SOT-23 package and identical pinout, yet they differ decisively in light-load operating mode — a difference that completely changes the choice for ripple-sensitive designs.
목차
The Ripple Problem in Sensitive Circuits
When generating a 5V or 3.3V local rail from a 24V industrial bus, the most common choice is a synchronous step-down converter. But in systems where the load is not constant — think IoT gateways where current consumption swings from a few mA during sensor polling idle periods to hundreds of mA during active cycles, or wireless modules that draw single-digit mA in standby but surge past 1A during transmission — the output ripple during light-load intervals becomes a real problem. As the load lightens, the converter enters pulse-skipping mode, the switching frequency becomes irregular, and the output capacitor repeatedly charges and discharges, dramatically increasing ripple amplitude.
The trouble with this ripple goes beyond mere voltage variation. When the ripple frequency components overlap with an ADC’s sampling bandwidth, effective resolution (ENOB) takes a hit. It couples directly into PLL phase noise, and survives as system noise that even the CMRR of an analog signal chain cannot reject. Switching to an LDO eliminates the ripple, but in a 24V-to-5V step-down, an LDO manages only about 21% efficiency — at 1A load, that means 19W of heat. You’re trading ripple for a heatsink and extra board area.
Why Eco-Mode Ripple Grows at Light Load
The TPS54202 is a 2A converter employing fixed 500kHz peak current-mode control with Advanced Eco-mode™. Eco-mode works by skipping switching cycles when the load drops below the level that the high-side FET’s minimum on-time can sustain — the converter only turns on the FET when the output voltage falls below the setpoint, then returns to idle once the target is reached. The impressively low 45μA no-load quiescent current (Iq) is a direct consequence of this Eco-mode operation.
This efficiency optimization comes at a cost, however. During pulse-skipping intervals, the effective switching frequency can drop to just a few hundred Hz. A single burst of energy overcharges the output capacitor, and the load slowly bleeds it off — a pattern that repeats. As a result, output ripple amplitude climbs from 10–15mV at heavy load to 40–80mV at light load, and can exceed 100mV under certain conditions. The ripple frequency also becomes irregular, making it difficult to filter out any specific band with a notch filter.
How FCCM Solves the Ripple Problem
The TPS54308 comes in the same 6-pin SOT-23 package with an identical pinout to the TPS54202, but there is one decisive difference: it operates in FCCM (Forced Continuous Conduction Mode), never stopping switching even at light load. No matter how low the load current drops, the 350kHz switching frequency remains constant, and the inductor current is allowed to flow in the negative direction to maintain continuous conduction. This design philosophy trades away some light-load efficiency (Iq = 300μA, versus 45μA for the TPS54202) in exchange for keeping output ripple predictably low across the entire load range.
In FCCM, output ripple is fundamentally determined by \( \Delta V_{OUT} pprox \Delta I_L imes ESR + \Delta I_L / (8 imes f_{SW} imes C_{OUT}) \), where \( \Delta I_L \) itself is fixed by the constant switching frequency and duty cycle. There is no structural reason for ripple to increase at light load. In practice, with a 12V-to-5V, \( L = 10\mu H \), \( C_{OUT} = 2 imes 22\mu F \) MLCC configuration, the TPS54202 exhibits over 60mVpp ripple at light load (10mA), while the same board fitted with a TPS54308 keeps ripple within 15–20mVpp. With ripple frequency locked at 350kHz, an additional notch filter targeting that single band can further suppress it.
Why Not Just Use an LDO?
The first workaround an engineer facing a ripple problem reaches for is an LDO. Output ripple is virtually nonexistent at the μV level, and the external BOM is just two capacitors. But this choice inflicts a severe efficiency penalty, especially when the input-to-output voltage differential is large. For example, an LDO supplying 5V at 0.5A from a 24V input dissipates \( (24V – 5V) imes 0.5A = 9.5W \) — nearly four times the power delivered to the load (2.5W). A SOT-223 package cannot handle this thermally, and even a TO-220 with a heatsink will cook nearby components.
The TPS54308 FCCM converter, under the same conditions (24V to 5V, 0.5A), delivers roughly 88–90% efficiency, meaning the converter itself dissipates only about 0.28W. That’s a 97% reduction in loss compared to the LDO — translating directly to no heatsink required, a tiny SOT-23 6-pin package, and layout freedom on tight boards. Moreover, the FCCM keeps ripple at a sufficiently low 15–20mVpp — not quite LDO levels, but adding a single LC filter stage after the converter can bring it below 5mVpp for ultra-low-noise rails.
| Approach | Output Ripple | Efficiency (24V→5V, 0.5A) | Power Loss | Heatsink |
|---|---|---|---|---|
| LDO (e.g., LM7805) | ~μVpp | 21% | 9.5W | Required (TO-220 + heatsink) |
| TPS54202 (Eco-mode) | 60–80mVpp (light load) | ~90% | 0.28W | Not required |
| TPS54308 (FCCM) | 15–20mVpp | ~89% | 0.31W | Not required |
Real-World Swap: TPS54202 to TPS54308
On a sensor interface board operating from 24V, producing 5V for a variable 10mA–500mA load, the initial design using the TPS54202 performed well at medium-to-heavy loads (200mA and above). But during the idle periods between sensor polling cycles (approximately 10–15mA), output ripple shot up to 70mVpp. This ripple jittered the LSB of a 16-bit ADC by 4–5 bits, dragging effective resolution down to roughly 11 bits — and software averaging could not fully eliminate it.
After swapping to the TPS54308 on the same board, with the same \( L = 10\mu H \) and \( C_{OUT} = 2 imes 22\mu F \) components, light-load ripple dropped to 18mVpp — a 74% reduction. Thanks to the fixed 350kHz switching frequency, the ripple FFT spectrum also concentrated into a single peak. ADC effective resolution recovered to 14.2 bits, and with software oversampling, better than 15 bits was achievable. Power dissipation increased by roughly 30mW (from the Iq difference of 45μA to 300μA), but overall efficiency remained at 89% with no thermal issues whatsoever.
Notably, the TPS54202 and TPS54308 share the exact same pinout (1: GND, 2: SW, 3: VIN, 4: FB, 5: EN, 6: BOOT). All these improvements were obtained by simply desoldering the TPS54202 and soldering in a TPS54308 — zero PCB changes required. One minor consideration: the TPS54308 switches at 350kHz versus the TPS54202’s 500kHz, so adjusting the inductor value can further optimize for the same ripple target if desired.
Spec Comparison and Selection Criteria
| Parameter | TPS54202 | TPS54308 |
|---|---|---|
| Input Voltage | 4.5–28V | 4.5–28V |
| Output Current | 2A | 3A |
| Integrated FETs (HS+LS) | 148 + 78mΩ | 85 + 40mΩ |
| Switching Frequency | 500kHz (spread spectrum) | 350kHz (fixed) |
| Light-Load Mode | Advanced Eco-mode™ (pulse skip) | FCCM (forced continuous conduction) |
| Quiescent Current (Iq) | 45μA | 300μA |
| Soft Start | 5ms (internal) | 5ms (internal) |
| Protection | OCP, OVP, TSD | OCP, OVP, TSD |
| Package | SOT-23 (6) | SOT-23 (6) |
| Pinout | GND-SW-VIN-FB-EN-BOOT | GND-SW-VIN-FB-EN-BOOT (identical!) |
For ripple-sensitive analog circuits, precision sensor supplies, or local rails in RF blocks, the predictable low ripple that the TPS54308’s FCCM delivers is far more valuable than the TPS54202’s light-load efficiency advantage. Conversely, in battery-powered devices where light-load efficiency directly governs runtime and the load consists of ripple-tolerant digital circuits, the TPS54202’s 45μA Iq and Eco-mode make it the better choice. Since both parts share an identical pinout, the most practical approach during prototyping is to test both, measure ripple and efficiency, and make the final decision based on real data.