Over the years, I've taken some notes of do's and don'ts based on my experience in the industry. Some of it may seem pretty obvious, but believe it or not, it's not.
On a high level, if you want to know what my "ideal PSU" is, it would be something like the Corsair HX Platinum (non-i) released in 2017, but with either a totem pole or MOSFET bridge for line rectification instead of the bridge diode. I'd consider replacing the two supervisor ICs with an MCU to achieve the multiple +12V rails. I'm not opposed to using GaN or SiC, especially for the bridge rectification, but there would have to be marketable advantages (higher efficiency, smaller size) and no significant cost adder.
Only use toroidal coil inductors. Never rod coil:
Toroids have less audible noise because the magnetic forces do not exert bending movement on the core. The core is only in compression or tension, and the circular shape is more stable mechanically. Toroidal chokes do take up more space on the PCB, so your overall footprint may be larger, but this is a sacrifice I'm willing to make.
Paper on Inductors and Inductor Coils
Measures Against Acoustic Noise in Power Inductors
Try not to use “Full Package” through hole (TO-220FP) MOSFETs for PFC:
If using smaller TO-220 MOSFETs, try not to use “Full Package” on the primary side. The only alternative would be to use the less common, more expensive TO-247, which is a larger package and therefore has more surface area for heat dissipation
.
Only use MOSFETs for synchronous rectification and not Schottky diodes:
MOSFETs ensure optimal efficiency and voltage regulation, as well as greater reliability over Schottky diodes in the SR circuit.
Utilize OTP on primary heatsink, as well as SR:
Typically there is only an OTP sensor on the SR (which can typically run over 100°C at full load). But the primary side typically runs just as hot, especially at lower mains voltage. So it is good to add an OTP in the form of a thermal fuse as shown below:
Speaking of bridge line rectification… Use active bridge or Totem-Pole instead of diode bridge…
Other than the SR, this is the hottest part of your PSU. It just makes sense to get out of the 80’s and use a more modern, more efficient topology.
How GaN enables high efficienct in totem-pole PFC-based power designs
A comparitive analysis of topologies for a bridgeless-boost PFC circuit
Heatsinks on the +5V and +3.3V DC to DC:
Typically, these are not sinked and heavy +3.3V and or +5V loads can make the MOSFETs run hot and cause the fan to ramp up prematurely.
ONLY use PWM fan control and always use Melexis MLX90411 or better driver IC:
Ensure lowest possible RPM and lowest noise possible at these RPMs with this combination.
Believe it or not, I'm not a huge fan of Zero RPM fan mode:
In lieu of a zero RPM fan mode, I would have the fan start off at the lowest possible RPM (potentially 250 to 300 RPM) and then ramp up only as needed. I would still utilize an MCU that can determine whether the fan should run based on temperature and load, and also implement a 5 minute hysteresis.
Utilize multiple supervisor ICs and/or MCU for multiple +12V OCP and allow the user to turn it on or off:
Typical current supervisor ICs have a limited number of current monitoring pins to support OCP. These lead PSU engineers to increase the OCP trip point to a higher than desirable value. By using multiple ICs, to multiply the number of OCP pins, or by implementing an MCU, one can have multiple “rails”, preferably the same number of OCP points as there are output connectors. Each connector can have the OCP trip point at a reasonable limit. The user should also be allowed to turn the OCP off for the multiple +12V rails which would switch the PSU over to a "single +12V rail" model for troubleshooting purposes.
ONLY use hexagons. No other shape is as efficient:
Squares are even worse:
Omni Calculator's 2D Geometry Calculator
Require soft termination MLCCs at particular “high risk” locations:
If an MLCC is < 2mm from PCB edge, or <3mm from a PCB screw hole, a soft termination MLCC must be used.
Preferably, you would not want MLCC's on a modular board. With the insertion and removal of cable connectors, the PCB flexes and can cause even soft termination MLCCs to come loose. But I put this in here because of an experience I had where they were putting MLCCs too close to PCB edge and/or screw holes and they were getting damaged during assembly.
Soft Termination Capacitors, Inductors, and Chip Beads for High-Reliability Products for Automotive Applications
Understanding Chip Capacitors
Require NP0 MLCCs for CT and Control Loop circuits:
To avoid any thermal drift, CT (timing capacitor for setting converter switching time) and control loop capacitors need to be NP0.
This has been a reoccuring theme for PSU failures over the years. Thermals cause most capacitors, resistors, etc. to change value. When these components are used in a control loop circuit, the components they control can go critical (worst case: causes a MOSFET to experience thermal runaway).
Pipe dreams:
Support both ATX12V and ATX12VO with the same PSU:
Have both 24-pin and 10-pin connectors on PSU. Main PCB standby rail is +12VSB, but this can convert to +5V on the modular PCB for when ATX12V is used.
Disable +3.3V and +5V DC to DC when the ATX12VO connector is used with no SATA/PATA:
Somehow implement a sense that knows that +5V and/or +3.3V leads are populated and if they are not, “turn off” the secondary DC to DC so only +12V is rectified.