Group Actions
Engagement
  • Author Author: Jan Cumps
  • Date Created: Date Created
  • Views 3784 views
  • Likes 9 likes
  • Comments 24 comments
Related
Recommended

GaN Point of Load converter 48V to 1V 50A - part 1: Design Overview

Jan Cumps

I'm reviewing a Gallium Nitrate step-down converter for Point of Load (PoL) high power conversion

 

image

 

This design is intended to deliver low voltage (0.5 - 1.5V) at a power hungry point of load. It can source 50A. With a switching frequency of 600 kHz, the footprint can stay small.

 

 

A 48V DC power bus is common in industrial environments. That voltage often needs to be converted to other levels.

For high-performance processors, FPGAs and application specific ICs, this can be as low as 0.5V or 1V,  with a current draw of many amps.

The evaluation kit here (LMG5200POLEVM-10), that I got for review from TI almost a year ago, is intended for such applications - located close to the low voltage load.

 

GaN Power Stages

 

This isn't the simplest circuit. It consists of a half-bridge Buck converter, a transormer (a funny one) an output rectifier with filtering.

 

Primary Side

 

image

source: Using the LMG5200POLEVM-10 48V to Point of Load EVM

 

The half-bridge on the primary side is a GAN device that I've extensively reviewed in previous blogs: LMG5200. Check the blogs for details.

For this design, it's essential to understand that it is used as a typical Buck and that it switches at high frequencies.

We'll visit this component again when we check the PWM controller (it's a smart one with i2c interface) that is the brain of the converter.

 

Transformer

 

It's built from PCB traces, with a ferrite core over it. SO no wires in this transformer , just a single PCB trace on each side of the board.

edit: This is a 10 layer board. The Transformer has a 5:1 ratio. Check the comments below.

 

Planar transformer surrounded by a ferrite core

(two half cores that connect trough rectangular openings in the PCB)

10 layer board with 5 primary windings and 1 secondary winding

image

Layers 1,4,5,7,10 dedicated to the secondary side (each one turn in parallel) and the other to the primary side (one turn each in series).

 

 

The stacking goes S PP SS P S PP S (S=secondary P= primary).

 

image

source: Using the LMG5200POLEVM-10 48V to Point of Load EVM

It's crazy that you can transfer 75 Watt with this construct, isn't it?

 

Secondary side

 

This is a GaN transistor based rectifier and an LC filter to smoothen the output.

The gates of these transistors (in reality Q3 and Q4 in Figure 1 each are constructed with two discrete GaN FETs) are driven with the same signal that drives Q1 and Q2 on the primary side.

 

image

Core of the power circuit. Additional decoupling and filter caps not shown here.

source: Using the LMG5200POLEVM-10 48V to Point of Load EVM

 

 

 

There's no heat sink required. When the output current exceeds 20A, active cooling (a fan) is needed.

 

In the next blogs, I'll check out additional circuits on the development kit: the step-down controller and the on-board 10A pulsing(!) test load.

I hope that someone with sound electromagnetic knowledge chimes in to discuss the PCB transfo (the core is ER18-3.2-10-3F45-S from Ferroxcube).

 

Blog Posts
part 1: Design Overview
part 2: Current Doubler
part 3: Dynamic Test Load
Related Blog
Checking Out GaN Half-Bridge Power Stage: Texas Instruments LMG5200 - Part 1: Preview
Parents

  • "I hope that someone with sound electromagnetic knowledge chimes in..."

     

    That's me out of it, then. You'll have to ask one of the PWMs if it's clever theory you want.

     

    It's confusing when you first look at it. That's partly because of the way it's drawn - it looks like a switched transformer. If you redraw the input like this, it becomes clearer:

     

    image

     

     

    the 1:1 transformer is simply using the ripple current in the coil of the input buck converter to induce the voltage in the secondary side which is then synchronously rectified for efficiency. What a neat idea - wish I could think of things like that. Can't quite get my head around the control side for the moment - does the input have to operate at the point between continuous and discontinous operation? Perhaps I should read the datasheet...

     

    Be careful with probing this one.

  • in reply to jc2048

    With regard to the transformer, the pcb trace winding is probably quite close to what you would achieve with a solid wire with a diameter that was 64% [2/pi] the width of the track (that's because of skin effect which would push the current to the outer of the wire - I'm assuming that the switching frequency is several MHz - imagine hollowing out 99% of the inside of the wire, where the current is low, and then flattening it to get the trace). So the winding is more substantial than it might seem. Are there parallel traces on multiple layers of the board to get the resistance down? It's still very impressive though.

     

    I'd like to see the temperature measurements when it's running flat out - if the winding had a resistance of 1mOhm, then it would dissipate 2.5W at 50A, so it will get hot, though it's manageable (obviously so, because they've done it).

Comment
  • in reply to jc2048

    With regard to the transformer, the pcb trace winding is probably quite close to what you would achieve with a solid wire with a diameter that was 64% [2/pi] the width of the track (that's because of skin effect which would push the current to the outer of the wire - I'm assuming that the switching frequency is several MHz - imagine hollowing out 99% of the inside of the wire, where the current is low, and then flattening it to get the trace). So the winding is more substantial than it might seem. Are there parallel traces on multiple layers of the board to get the resistance down? It's still very impressive though.

     

    I'd like to see the temperature measurements when it's running flat out - if the winding had a resistance of 1mOhm, then it would dissipate 2.5W at 50A, so it will get hot, though it's manageable (obviously so, because they've done it).

Children
  • in reply to jc2048

    Jon Clift wrote:

     

    ...

     

    I'd like to see the temperature measurements when it's running flat out - if the winding had a resistance of 1mOhm, then it would dissipate 2.5W at 50A, so it will get hot, though it's manageable (obviously so, because they've done it).

    I'll measure the temperature (I still haven't made a decent load). There's this warning right above the area of the transfo though:

    image

     

    And in the manual:

    Some components reach temperatures above 50°C.

  • in reply to Jan Cumps

    Hi Jan,

    since I have some data at temperature, I would like to share it with you so that you can have a reference point.

    The thermal images below have been taken running at 75V input, 50A output current at ambient temperature with 200LFM of airflow.

    image

  • in reply to alberto@ti

    , you were right with the pcb loop getting warm. From the termal pictures above from , it gets close to 50°C - 65° (say ambient + 25° - 40°).

    Two of the secondary side FETs (the 4 skewed components are the secondary side GaN FETs) also get warm, although I think the heat is coming from PCB coil again - not from the FETs.

    Otherwise there wouldn't be a hot and cool side on the FETs. They are bare die, unpacked. And Id expect the two outside FETs to be similarly hot then - they do the same work.

  • in reply to Jan Cumps

    I think you're probably right about the transformer 'winding' heating the FETs.

     

    I would imagine that the outer FETs are actually doing more of the work (though that's not what you'd first think just looking at the picture). The Rds-on will go up with temperature (I'm assuming the GaN parts are similar to silicon in that respect - they must be or they wouldn't have paralleled them), so the hotter inner pair will have a higher resistance than the outer pair. Normally the pair would try and balance, because the hotter one would lose current to the cooler one, but here that's overridden by the external contribution. It's a bit of an unfortunate layout, but it might have been the only way they could make the high-current side work (with a board like this there are going to be conflicting constraints).

     

    Shame the ferrite isn't a bit hotter - I was going to suggest you try cooking a small omelette on it.

     

    Are you going to try and match their airflow figure?

  • in reply to jc2048

    Jon Clift wrote:

    ...

    Shame the ferrite isn't a bit hotter - I was going to suggest you try cooking a small omelette on it.

    ...

    tru dat

     

    Jon Clift wrote:

    ...

     

    Are you going to try and match their airflow figure?

    No. I can't do a better job than that - or even an as good one. I don't have tools to do that, only the typical temperature sensor that comes with a multimeter these days.