Designing and Maximizing Switch-Mode Efficiency of MODPSU, the Modular Power Supply

Abstract: This work is an investigation into practical design and loss minimization of DC/DC-switched power supplies, enabling cooler operation, smaller size, and additional or wider areas of use. There is an abundance of switch controller ICs on the market and a few of the manufacturers offer very sophisticated design tools that suggest operating parameters and grade components. However, one of the major losses (inductor AC loss) is not accounted for, so the engineer must combine these results with the ones from inductor manufacturers' loss calculators. One obvious question is whether you can trust these recommendations and simulation results. This work takes a special interest in inductor losses in relation to input voltage, load current, and switching frequency for a few very common use cases (12V to 3.3V, 12V to 5V, and 9-20V to 12V), where total loss minimization is the greatest concern. Both a combined simulation of inductor and FET transistor losses and a temperature rise experiment suggest that there is a switching frequency sweet spot (for a given load current and input voltage), where the total inductor and FET losses are at a minimum. However, the efficiency and loss measurement experiments show no sign of it and instead indicate that minimum losses occur at minimum switching frequency (in this case 200 kHz). One conclusion is that the switch controller design tools recommend a higher switching frequency than what's optimal from a loss minimization perspective or suggest a too low inductance value that would give much higher core losses than what's possible to achieve. The steep left slope of the inductor loss curve is dangerous, so if in doubt choose a higher inductance value. A comprehensive selection of off-the-shelf inductors from three manufacturers was tested. Only one vendor publishes core material specifications, which incidentally is the only one whose loss calculator stands up to scrutiny (Vishay). All off-the-shelf inductors in this range come in either of two core materials; Mn-Zn ferrite and iron powder that are both cheap and easy to produce. On average, the iron powder inductors have significantly higher core losses than the Mn-Zn ferrite ones, although one of the iron powder inductors performed on a par with the ferrites but without suffering from the latter’s known weaknesses (sudden and highly temperature dependent saturation). Among the three tested iron powder sizes, size has very little impact on losses (except for the smallest one). The main difference is that smaller inductors heat up and saturate faster and that the lower DCR of a bigger one might result in a higher optimal inductance value. Surprisingly, low rated inductor DCR correlates to high total losses, indicating a lower quality core material. The body diode of modern FETs is so good that there's no need for an external freewheeling diode. Due to idle losses, dual phase operation only becomes attractive at relatively high load currents. Real-world inductors are far from ideal, so beware of sour spots (unfortunate sets of winding turns, wire gauge, and core size). As a result of this work, the MODPSU products perform significantly better than the competitor's closest counterparts, with 60% lower total losses.