This is looking 😎 :
http://www.ti.com/lit/ds/symlink/lmg5200.pdf
History of GaN promises over the last 5 years was really annoying.
Keeping fingers crossed that the technology has now reached a
satisfying maturity.
http://www.ti.com/lit/ds/symlink/lmg5200.pdf
History of GaN promises over the last 5 years was really annoying.
Keeping fingers crossed that the technology has now reached a
satisfying maturity.
Still a difficult package for the DIY folks building stuff in the basement. At least they have the correct gate driver on the device, which is a absolute must have.
I work with SMD all day and even with all the fancy tools I have for dealing with tiny parts down to 0201, I have a hard time with BGA and other leadless power packages like those used for this part. No, we don't own a BGA rework station since it costs a bit more than 15K USD but I wish we did.
GaN could have gone a bit further if they'd have offered a few parts in TO-220 instead of those basically solder bumped dies. I'd even take D-PAK.
Class D with GaN is a very natural application for these parts.
What? 35.00 USD each. FAIL.
I work with SMD all day and even with all the fancy tools I have for dealing with tiny parts down to 0201, I have a hard time with BGA and other leadless power packages like those used for this part. No, we don't own a BGA rework station since it costs a bit more than 15K USD but I wish we did.
GaN could have gone a bit further if they'd have offered a few parts in TO-220 instead of those basically solder bumped dies. I'd even take D-PAK.
Class D with GaN is a very natural application for these parts.
What? 35.00 USD each. FAIL.
Last edited:
Yup, yesterday night I came to the same conclusion.
Originally I planned to simply order 10 pieces with my next order at mouser for playing around. But at mouser it's 45,- EUR / pc.
Ok, I am not in desperate need.
Topic postponed until it becomes available without such political pricing.
GaN is inherently much faster than silicon. This allows higher switching speeds, or less time spent in the transition region for the same switching speed. This allows higher efficiency power supplies and class D amps than silicon.
GaN is inherently more costly than silicon, and as of two years ago when I spent some time looking at GaN devices for LTE transmitters, GaN devices must be made on top of another type of substrate, usually SiC or glass. This raises the thermal resistance, making it harder to get the heat out of the package. Early RF devices from RFMD and Nitronex were prone to failure due to "hot spotting" if operated continuously in the linear region.
GaN is inherently more costly than silicon, and as of two years ago when I spent some time looking at GaN devices for LTE transmitters, GaN devices must be made on top of another type of substrate, usually SiC or glass. This raises the thermal resistance, making it harder to get the heat out of the package. Early RF devices from RFMD and Nitronex were prone to failure due to "hot spotting" if operated continuously in the linear region.
What about SiC ?
Why nobody seem use Sic for Class D ? (sorry if I not aware about people works)
Why nobody seem use Sic for Class D ? (sorry if I not aware about people works)
Last edited:
SiC has the focus on devices for 600V and above.
The body diodes have reasonably low Qrr, but very high Vf.
GaN is also attractive for lower voltages.
There are no body diodes (except cascode arrangement).
Inherently backwards path is being opened by the field between gate & drain,
this also leads to a higher voltage drop than a MosFet body diode, but the great thing is that there is no Qrr.
For general GaN info read here:
EPC -Tools and Design Support
The body diodes have reasonably low Qrr, but very high Vf.
GaN is also attractive for lower voltages.
There are no body diodes (except cascode arrangement).
Inherently backwards path is being opened by the field between gate & drain,
this also leads to a higher voltage drop than a MosFet body diode, but the great thing is that there is no Qrr.
For general GaN info read here:
EPC -Tools and Design Support
When you look through EPC's handful of app notes you quickly realize why they don't even try offering a packaged version of the die. They explain how 1nH of lead inductance is one too many.
Unfortunately, attempting to accommodate their 3 dimensional multi-layer board layout suggestions without using an aerospace/mil quality board house is just about impossible. Even getting 2 ounce copper double sided and vias small enough to lead the current in from both sides of the die as is shown in the recommended layouts would be surprising. One ounce copper single sided trace leads coming at the source and drain bumps each from only one side might not just ruin the inductance cancellation plan, they might fuse before allowing rated die current. I haven't calculated the current density, but ~200um width and 5 amps per trace seems a little high, especially since the traces at that area will be pretty much at die temperature. ..and that's just for the steady state current rating.
Unfortunately, attempting to accommodate their 3 dimensional multi-layer board layout suggestions without using an aerospace/mil quality board house is just about impossible. Even getting 2 ounce copper double sided and vias small enough to lead the current in from both sides of the die as is shown in the recommended layouts would be surprising. One ounce copper single sided trace leads coming at the source and drain bumps each from only one side might not just ruin the inductance cancellation plan, they might fuse before allowing rated die current. I haven't calculated the current density, but ~200um width and 5 amps per trace seems a little high, especially since the traces at that area will be pretty much at die temperature. ..and that's just for the steady state current rating.
Last edited:
Played a few hours with the EPC devices when they came out, but had to
accept that they where mostly unusable with my DIY approach for various reasons.
Parasitic inductances + mechanical reasons + signal integrity of the gate drive is the reason why I think the LMG5200 is pretty promising.
However without samples its all just hot air
accept that they where mostly unusable with my DIY approach for various reasons.
Parasitic inductances + mechanical reasons + signal integrity of the gate drive is the reason why I think the LMG5200 is pretty promising.
However without samples its all just hot air
GaN has all of the advantages as advertised and provides excellent linearity of the open loop amplifier. Packaging a die only adds bad things - inductance, cost, resistance, size, reliability risk. Wafer level packaging is expanding rapidly in both discrete and IC devices. Mounting is quite easy. You need a little (very little) tacky flux, tweezers, and a heat gun. For a very thick board, you may need a hot plate or a heat source from the back. Smear a tiny bit of tacky flux, place the device, and blow hot air. Details and video are in Assembly Basics
Many of the GaN devices made today are targeted at high speed switching, and are not optimized for linearity. This is OK for class D amps.
GaN is also being targeted at GHz frequency linear RF amplifiers, and these devices are optimized for linearity. The devices I tested at 700 MHz and 2.3 GHz were not as linear as an LDMOS device at the same power level.
LTE requires excellent linearity and all the GaN devices I tested required adaptive predistortion to make the LTE specs. Several LDMOS devices would meet spec with sufficient margin without DSP help, AND they were much cheaper.
I am sure that the situation will improve because GaN is more efficient, and the large cell phone carriers like AT&T and Verizon have HUGE electric bills, millions of $$$$ per month! A good chunk of that energy is burned up in the inefficient RF power amps.
GaN is also being targeted at GHz frequency linear RF amplifiers, and these devices are optimized for linearity. The devices I tested at 700 MHz and 2.3 GHz were not as linear as an LDMOS device at the same power level.
LTE requires excellent linearity and all the GaN devices I tested required adaptive predistortion to make the LTE specs. Several LDMOS devices would meet spec with sufficient margin without DSP help, AND they were much cheaper.
I am sure that the situation will improve because GaN is more efficient, and the large cell phone carriers like AT&T and Verizon have HUGE electric bills, millions of $$$$ per month! A good chunk of that energy is burned up in the inefficient RF power amps.
next big thing to bring losses down for LTE is envelope tracking using multi phase synchronous buck converters.
Accepted, if one is really really is desperate to use the unpackaged items, it is doable. However I would need to upgrade my DIY work bench with pretty some equipment and beyond the shown parts additionally use magnifier glasses and a solder fume extraction unit. The toxic gases of the long term heated FR4 is really not what I want to breath for half an hour per device.
In the LMG5200 most critical paths, except the external caps, are integrated.
To me this appears pretty attractive.
Early GaN devices typically suffered badly from high dynamic Rdson, while only the static Rdson was stated in the data sheets.
In 2013 at the PCIM there was a paper (forgot the company behind...) which stated that this had massively improved. I did not restart experiments with new types, but now again two years later I tend to anticipate that it is really solved also in the series production.
Having in mind the 2013 PCIM, it is impressive to read that EPC had this topic under control already in 2011.
http://epc-co.com/epc/Portals/0/epc/documents/product-training/epc_relreport_030510_finalfinal.pdf
Let's pick a new type, i.e. EPC2012C:
Which average ratio of dyn vs static rdson do I have to expect when using the device with a 150V (+/-75V) rail in a classd amplifier at 1MHz?
GaN is really interesting for analog FM broadcast, it can potentially provide a huge efficiency boost there as you can potentially build >100MHz class D amps with it. Unfortunately current GaN devices have a pretty tiny SOA, so it's hard to make an amplifier that's able to happen any significant reflected power.
A typical 40KW FM radio transmitter pulls 60KW from the wall, and dumps out 20KW of heat that has to be removed from the transmitter building somehow. Adds up to a hefty power bill for the broadcaster.
Oh, and I believe these days there's really no excuse not to use RF precorrection. High speed A/D and D/A converters are cheap and easy to come by, as are FPGA's with tons of signal processing capability, and precorrection algorithms are very well researched and understood. Plus since you're sensing your own RF output, you've got the ability to self-test your spectral compliance and pull yourself off the air if something bad happens. And it makes for a more efficient transmitter as you don't need to run as much headroom/bias on the amp.
I will agree on the need for a magnifier, but disagree on the half hour soldering time. From an equipment perspective, my assumption is that hand soldering is used and ventilation is available. I neglected to mention that cleaning and baking is recommended to avoid uncured flux underneath (after surrounding components have been soldered.
For the EPC2012C, under the conditions that you described, dynamic Rdson will be negligible provided high frequency loop inductance is kept down and spikes are kept within the 200 V datasheet limit.
Deadtime should be minimized. The following page has an open loop half bridge development board with good layout and commutation timing. The page has schematic, BOM, and gerbers, along with a link to Digi-Key to buy.
For the EPC2012C, under the conditions that you described, dynamic Rdson will be negligible provided high frequency loop inductance is kept down and spikes are kept within the 200 V datasheet limit.
Deadtime should be minimized. The following page has an open loop half bridge development board with good layout and commutation timing. The page has schematic, BOM, and gerbers, along with a link to Digi-Key to buy.
...searched digikey and found pretty some portfolio of your boards...
Evaluation and Demonstration Boards and Kits | Programmers, Development Systems | DigiKey
And of course this one:
http://www.diyaudio.com/forums/class-d/273596-maybe-very-good-class-d-design.html
Complete amp with pre filter feedback.....
Really nice to see things moving ! 😀
Last edited:
testa di melone (Melon Head)! EPC9004
Many Development (subsystems) and Demo boards here Demo Boards
Awesome Class D Audio Demo here EPC9106
Many Development (subsystems) and Demo boards here Demo Boards
Awesome Class D Audio Demo here EPC9106
telco electric bills are scary. And just as you get a handle on it the goalposts change! The last test bed I built burned over 1MW when full. I think the most efficient cellular basestation I worked on hit 9% efficiency! But that was a few years ago now.
- Status
- Not open for further replies.