Here's a conversation that happens at least once a week in Shenzhen design reviews.
Engineer: "I need a 12V rail from mains. I'll use a flyback converter." Me: "Is the user going to touch any part of the output?" Engineer: "No, it's inside a sealed plastic enclosure." Me: "Then why are you paying for a transformer, an optocoupler, and a TL431?"
Most engineers reach for an isolated flyback by default. It's the safe choice. But it's not always the right choice. A non-isolated buck converter like Power Integrations' LinkSwitch-TN can do the same job with half the BOM cost, a third of the PCB area, and no transformer design headaches.
But get this decision wrong in the other direction - using non-isolated where isolation is required - and your product fails safety certification. Or worse.
If the user can touch any part of the output - any connector, any metal housing, any LED lens - you need isolation. No exceptions.
That's the rule. It's not about efficiency. It's not about regulation accuracy. It's about what happens when a single fault puts mains voltage on the output.
With an isolated topology, the transformer's reinforced insulation is the barrier. With non-isolated, there is no barrier. The output ground sits at the same potential as the rectified mains negative - a rectified, floating node that can sit 170V away from earth.
USB chargers, phone adapters, LED bulb drivers inside a plastic housing - all isolated. A sensor node sealed inside an HVAC duct - non-isolated is fine.
A smart plug's internal relay power supply - isolated, because the output side of the relay connects to something the user might touch. An LED driver inside a sealed metal downlight - non-isolated, because no user-accessible parts exist.
A non-isolated buck converter like the LNK306DG-TL connects the rectified DC bus directly to a buck inductor. There's no transformer between the mains and the output. The feedback pin monitors the output voltage through a resistive divider referenced to the same ground as the switching MOSFET.
The upside: the BOM is tiny. One IC. One inductor. Two capacitors. A freewheeling diode. A feedback resistor divider. That's it. No optocoupler. No TL431. No transformer. A competent layout engineer can get the whole power stage onto 20×30mm of PCB.
The downside: the output isn't isolated from mains. You can't connect it to anything a user touches. You can't ground the output. You can't probe it with a scope without an isolation transformer.
Output regulation is ±10-15% - fine for relays, MCUs with on-board LDOs, and sensor loads, but not for anything that expects a tight rail.
The other downside most people don't think about: the entire circuit lives at high voltage during debug. When you're probing a non-isolated supply, your scope ground clip becomes a short circuit to earth. One wrong probe placement destroys the board.
You need an isolated scope, a differential probe, or an isolation transformer on the bench. For a production design that never needs probing again, this doesn't matter. For a prototype you'll debug for weeks, it's a constant headache.
An isolated flyback puts a transformer between the mains and the output. The transformer provides both isolation and the voltage step-down ratio. The IC switches current through the primary. During the off-time, the magnetic field collapses and energy transfers to the secondary through the output diode.
But how do you regulate the output voltage when you can't directly measure it? The controller is on the primary side. The output is on the secondary side. There are two answers, and they define which PI part you pick.
The TNY284 uses an optocoupler and a TL431 on the secondary side. The TL431 compares the output voltage to its internal 2.5V reference. When the output exceeds the setpoint, the TL431 drives current through the optocoupler LED.
The optocoupler transistor pulls current from the EN/UV pin, and the TNY284 skips switching cycles until the output drops back into regulation.
Result: ±2% line and load regulation. This is the tightest regulation you can get from a flyback controller at this power level. If your load is a microcontroller, an analog sensor, or anything sensitive to supply variation, this is the path.
The cost: an optocoupler, a TL431, two resistors for the feedback divider, and maybe a compensation cap. And the optocoupler ages - CTR drops 20-30% over 5 years of continuous operation. It won't cause a failure, but it's a drift mechanism that primary-side designs avoid entirely.
The LNK625 eliminates the optocoupler entirely. Instead, it senses the output voltage through the flyback transformer itself. When the MOSFET turns off, the bias winding voltage reflects the output scaled by the turns ratio.
The LNK625 samples its FB pin 2.5µs after turn-off, compares it to an internal 1.8V threshold, and adjusts switching accordingly.
Result: ±5% regulation. Not as tight as optocoupler feedback, but more than adequate for battery chargers, motor drives, LED strings, relay coils - anything that tolerates a few hundred millivolts.
The BOM saves an optocoupler, a TL431, and two resistors on the secondary side. More importantly, there's no optocoupler to age, which matters in products designed for 10+ years of service.
Pick non-isolated when you need isolation: your product fails hi-pot testing. The certification lab finds 170V between the output connector and earth. You redesign the power supply and miss your launch date by three months.
In Shenzhen, this happens more often than anyone likes to admit - usually because the enclosure design changed late and what was a sealed plastic shell now has a metal trim ring the user can touch.
Pick isolated when non-isolated would work: you pay ~3× more for the power supply BOM than you needed to. The transformer takes up 40% more PCB area. Your product still works. It's just bigger, more expensive, and took longer to design.
For products shipping 100k+ units, that BOM difference matters. For a 500-unit industrial controller run, pick isolated and move on with your life.
User touches output? → Isolated flyback. Next question: do you need the tightest possible regulation? Yes → TNY284 with optocoupler. No → LNK625 with primary-side regulation.
User doesn't touch output? → Non-isolated buck. LNK306. Half the BOM. A third the PCB area. No transformer design. No optocoupler aging to worry about. Just make sure your enclosure doesn't change late in the design cycle.





