PMP20021: How to increase efficiency in PMP20021 type circuit to 98%

Nitish,

The losses in the components are based on calculations and are only estimates, but represent the majority of the losses. As you push for higher efficiency, it becomes increasing difficult to identify all losses, because even minor ones begin to have an impact. Some of the losses (switching transition times, lower FET body diode dwell time, etc) can generally only be measured to get accurate info. Other terms, not included before, such as FET Coss losses, begin to factor in too.   

1) If I reduce the switching frequency to 100kHz, would the faster FET in your response (4) be needed ?

A lower switching frequency will certainly help with switching losses. A faster FET will reduce the remaining switching losses further. I would start with the same FET as in the reference design and try a faster FET if you still need higher efficiency. The risk is that the increase in the new FETs conduction losses outweigh the gains from the switching losses. Alternatively, selecting a lower RDS_on FET for the bottom FET alone could also help (a faster FET for the top, and a lower Rds_on FET for the bottom).

2) I will check out Coilcraft SERxxx line of inductors. Are there other lower loss series you recommend for 300kHz and 100kHz?

Most of the major manufacturers, Wurth, TDK, Coiltronics, Pulse, etc make large ferrite inductors. Coilcraft's selector also specifies core loss based on your design inputs, and I believe Wurth's has a tool for this too.

3) Thanks for pointing out the snubber loss.

You are welcome.

4) Depending on my space requirement, I might choose not to reduce the switching frequency, in which case I will look into the faster FET.
I plan to use the formulas provided by George Lakkas in TI's document slyt664.

Another possibility is to parallel two FETs. This is beneficial if power in one FET alone creates too much temperature rise. It also decreases the conduction losses, assuming each FET has 2X the original FET's Rds_on, since each FET's conduction loss is based on I^2R. This will not help switching losses, unless each FET is has less than 1/2 the total gate charge of the original. Just another option that could help.

5) From the test result, it appears that PMP20021 achieves 95% efficiency at full load (12V x 20A). 5% of 240W = 12W.
Based on your loss figures, it appears that ~4W is in the controller / drive (LM25119) and PCB copper losses.
From LM25119 datasheet, bias current ~ 6.5mA total (max). That would mean LM25119 consumes 24V x 6.5mA = 0.156W.
Are you saying that ~3.8W is in PCB copper losses?

No, as mentioned the losses I mentioned are based of estimates. The PCB copper losses are unknown, but real. The controller losses are what you show plus some of the FET gate drive losses (based on f*Qtot*V). This term is divided up between the controller, the FET gate resistors, and the FETs.

regards,

John Betten