I have been thinking about creating an All SMD component Class D amplifier. The purpose would be to keep a totally clean back side of the PCB, so that mounting the amp to a heatsink or an aluminum plate would be as easy as just using a foam heatsink spacer in the full size of the PCB, with maybe a single or 2 screws as a safety measure if the adhesive should somehow slip over time.
Only starting
The size is limited to 100 x 55 mm as I want to use a small 100 x 60 x 10 mm heatsink from Aliexpress, at least for some builds
The attached picture shows the very first version where the schematic is not finished yet, but where the larger main components have been placed. As can be seen, there will not be room to spare, and it's not even for sure there will be room enough. First picture is with D2PAC Fets, but it would save a lot of space to find a suitable type in a Super SO8 PQFN 5x6 package ... see second picture. The voltage must be limited for this amp to something like max +-55V, as the coil will only be able to handle less than 10A also to find suitable MosFets I'm sure I will have to use 150V versions and not 200V as I usually do.
PCB will be 4 layers and will be made at JLCPCB. I will try to use basic components from JLCPCB as much as possible to limit the cost, as it cost 3 USD for every extended component type you use.
The MosFet footprints will be perforated by a number of vias which will act as a thermal conductor from the topside of the PCB to the bottom side, where the heatsink will be placed.
Now I'm trying to find suitable fets, and there actually seem to be quite some ..... seems the development is racing on for mosfets in small packages, but has almost halted for TO220 types!!!
Take a look at either BSC360N15NS3 or NTMFS034N15MC, 150V, +30A, around 30 mOhm, super fast including diode, and easy to drive with Ciss less than 1 nF ....... I'm thinking what's the catch here??? .... and why bother with GaN fets????
https://www.mouser.dk/datasheet/2/196/BSC360N15NS3_Rev2.0-75853.pdf
https://www.mouser.dk/datasheet/2/308/NTMFS034N15MC_D-2037032.pdf
Only starting
The size is limited to 100 x 55 mm as I want to use a small 100 x 60 x 10 mm heatsink from Aliexpress, at least for some builds
The attached picture shows the very first version where the schematic is not finished yet, but where the larger main components have been placed. As can be seen, there will not be room to spare, and it's not even for sure there will be room enough. First picture is with D2PAC Fets, but it would save a lot of space to find a suitable type in a Super SO8 PQFN 5x6 package ... see second picture. The voltage must be limited for this amp to something like max +-55V, as the coil will only be able to handle less than 10A also to find suitable MosFets I'm sure I will have to use 150V versions and not 200V as I usually do.
PCB will be 4 layers and will be made at JLCPCB. I will try to use basic components from JLCPCB as much as possible to limit the cost, as it cost 3 USD for every extended component type you use.
The MosFet footprints will be perforated by a number of vias which will act as a thermal conductor from the topside of the PCB to the bottom side, where the heatsink will be placed.
Now I'm trying to find suitable fets, and there actually seem to be quite some ..... seems the development is racing on for mosfets in small packages, but has almost halted for TO220 types!!!
Take a look at either BSC360N15NS3 or NTMFS034N15MC, 150V, +30A, around 30 mOhm, super fast including diode, and easy to drive with Ciss less than 1 nF ....... I'm thinking what's the catch here??? .... and why bother with GaN fets????
https://www.mouser.dk/datasheet/2/196/BSC360N15NS3_Rev2.0-75853.pdf
https://www.mouser.dk/datasheet/2/308/NTMFS034N15MC_D-2037032.pdf
Attachments
Think about cicle by cicle current protection, ovp/uvp protection, temp protection, isolation... etc! For example https://navitassemi.com/gancam-overview-of-nv6247-gansense-half-bridge-power-ic/
Well the IRS2029 + my normal error circuit provides all of this. What I'm considering is to also make peak and average power measurements, using a ATMEGA328p. Problem in this is the samplilng freq which can be had with this chip. But it could be cool at I could attach a small display 😉
Anyway, I have decided to go for an update of my original design, by using FDMS8620 fets. They actually comes in a double sided cooling version, but Mouser does not have these right now
Just ordered from JLCPCB
Anyway, I have decided to go for an update of my original design, by using FDMS8620 fets. They actually comes in a double sided cooling version, but Mouser does not have these right now
Just ordered from JLCPCB
Attachments
Woou this was fast. Good job JLPCB 
Next thing is to do some soldering 😉
.... I'll have some challenges with the FETs though. As I have made all efforts to use the PCB as heatsink, it will be super difficult to heat up the pads with a small soldering iron!
I have just ordered a small pre heating plate, but that will not arrive before May.
Will try to preheat with a hot air iron.
With the PCB in hand I immediately realized that I had not made heat spreading pads to the 2 small regulators .... damn. .... but think they will be ok though
Next thing is to do some soldering 😉
.... I'll have some challenges with the FETs though. As I have made all efforts to use the PCB as heatsink, it will be super difficult to heat up the pads with a small soldering iron!
I have just ordered a small pre heating plate, but that will not arrive before May.
Will try to preheat with a hot air iron.
With the PCB in hand I immediately realized that I had not made heat spreading pads to the 2 small regulators .... damn. .... but think they will be ok though
Attachments
Last edited:
Let JLPCB do the hard soldering?
What is the expectations on the board performance - the "usual" measurements?
//
What is the expectations on the board performance - the "usual" measurements?
//
Well I have bought a few FDMS8620 from mouser, and I was going to say they don´t have them at JLBPCB ... but I actually found they do ... so next time for sure
I have a good stash of IRS2092, and they are easy to handle so opted to not have them soldered as well as a few other components
I have a good stash of IRS2092, and they are easy to handle so opted to not have them soldered as well as a few other components
Quite nice PCB.
I just experimented with a completely DIY IRS2092 homemade PCB and started to look into possibilities of doing my next experiment also entirely with SMD technology. I got to the point that they even make a printed circuit board on aluminum laminate for a few euros. Something I would like to try!
I stuck with the existing Infineon BSC320N20 MOSFETs, which I guess share same Super SO8 case as pads on your PCB?
There are some videos on internet that seem to show that the IRS2092 is pushed quite to the limit, supply voltage is only a few Volts away from the maximum declared on its datasheet. And some of these builds seem to largely ignore even the recommended layout, and strangely enough, these experimental builds actually even work.
What order of supply voltage do you use for power?
Couldn't capacitor in the input also be replaced with some SMD type? I mean some ceramic. Why not replace this electrolyte with X5R or X7R ceramic? I personally don't see a reason why it shouldn't be done?
I looked at my Colicraft VER output inductor and even that can be tweaked a bit in a home experimental project to make it an SMD component.
I just experimented with a completely DIY IRS2092 homemade PCB and started to look into possibilities of doing my next experiment also entirely with SMD technology. I got to the point that they even make a printed circuit board on aluminum laminate for a few euros. Something I would like to try!
I stuck with the existing Infineon BSC320N20 MOSFETs, which I guess share same Super SO8 case as pads on your PCB?
There are some videos on internet that seem to show that the IRS2092 is pushed quite to the limit, supply voltage is only a few Volts away from the maximum declared on its datasheet. And some of these builds seem to largely ignore even the recommended layout, and strangely enough, these experimental builds actually even work.
What order of supply voltage do you use for power?
Couldn't capacitor in the input also be replaced with some SMD type? I mean some ceramic. Why not replace this electrolyte with X5R or X7R ceramic? I personally don't see a reason why it shouldn't be done?
I looked at my Colicraft VER output inductor and even that can be tweaked a bit in a home experimental project to make it an SMD component.
Hi overpower
I have been 120% tied up in a new job lately .... and probably will for a good while, so no progress on my side wrt class d :|
I did solder up a board with the SO8 fets, but could not get it to work.
Have made a new revision but it is waiting for me to have time to solder the components.
I bought a pre-heater plate to heat the whole board to say 125 deg C, before soldering the SO8, this is necessary as you are using the board as a heatsink ... so it will of course also cool the solder making it difficult to hand solder 😉
Yes for sure do not go above the voltage spec and keep a little overhead. Bad layout and design will have massive overshoot on output and might easily go beyond voltage spec. Will it work ... maybe, but it will not be reliable!
I have not used more than +-70V yet. Think you should be able to go to 90V though.
Ceramics types like X5R and X7R will add measurable distortion when used in the input, much better to use a bipolar electrolytic which can also be had in SMD. Film capacitors are better, but also much larger and not readily available in smd. Difference btw electrolytic and film will be very small in measurement and I think not hearable in reality.
https://audioxpress.com/article/practical-test-measurement-stop-worrying-about-coupling-capacitors
Ceramics C0G are almost perfect caps, but not at all available in values for input cap.
I use X7R extensively for decoupling.
Yes BSC320N20 also uses the Super SO8 case.
I have been 120% tied up in a new job lately .... and probably will for a good while, so no progress on my side wrt class d :|
I did solder up a board with the SO8 fets, but could not get it to work.
Have made a new revision but it is waiting for me to have time to solder the components.
I bought a pre-heater plate to heat the whole board to say 125 deg C, before soldering the SO8, this is necessary as you are using the board as a heatsink ... so it will of course also cool the solder making it difficult to hand solder 😉
Yes for sure do not go above the voltage spec and keep a little overhead. Bad layout and design will have massive overshoot on output and might easily go beyond voltage spec. Will it work ... maybe, but it will not be reliable!
I have not used more than +-70V yet. Think you should be able to go to 90V though.
Ceramics types like X5R and X7R will add measurable distortion when used in the input, much better to use a bipolar electrolytic which can also be had in SMD. Film capacitors are better, but also much larger and not readily available in smd. Difference btw electrolytic and film will be very small in measurement and I think not hearable in reality.
https://audioxpress.com/article/practical-test-measurement-stop-worrying-about-coupling-capacitors
Ceramics C0G are almost perfect caps, but not at all available in values for input cap.
I use X7R extensively for decoupling.
Yes BSC320N20 also uses the Super SO8 case.
Those ferroelectric caps are non-linear, and microphonic - neither suggest they are usable in an audio front-end.
This question was in the air for several weeks...
OK, this answer was also found a week ago - I researched this matter more thoroughly and found the necessary information and technical explanation why X5R or X7R (really also R, not to be confused with the "U" index) are also not good, although I assumed that they could be suitable.
When they are used, especially in the lower frequency, additional distortions occur but it is not add other dangers, but these distortions is also not wanted and then I need to stay with elcos.
NP0 would be ideal, but a 22uF "battery" would take up half the Universe of the PCB.
Only 0.47uF is available in 1206 size and 50 of these do not seem conceivable considering the space layout.
Nichicon UWP1E220MCL1GB seems like a viable candidate to replace the crappy THT input electrolyte, and I can live with it.
2 pieces of UUG2A331MNQ1MS on power rails and 2 pieces of UWT1C101MCL1GB in power supply of the analog part of IC.
Basically, I can do my first test with a aluminum laminate PCB soon. I can get this PCB even smaller than my first attempt in FR4 version, which despite the single-sided copper is still smaller than this Chinese "500W IRS2092" which has spread around the world.
My very first hastily made IRS2092 module and next to it a Chinese "500W IRS2092 specimen" for comparison.
One MOSFET has been removed for its examination - they are usually fakes in these modules.
I believe that "for sake of gaining knowledge" I will go at least +-90V on this next test, but possibly even further.
Considering the IRS2092 "recommended" design solutions out there, I believe that experimenting with a compacter solution will not be an obstacle to achieving if better, then at least same...
Last year I spoke with one Infineon's developers and he confirmed that "In terms of reliability, he does not recommend exceeding values presented in datasheet" and at the same time he stated that "Staying within these limits, manufacturer guarantees reliability" - staying within the supply range of up to 200V means staying within the parameters indicated by manufacturer.
From an experimental point of view, he saw no reason why a +-100V power supply should not be used in experiments, but it is obvious that in a commercial application it should be held back a bit. Tolerance limit in reality is around 230-240V, i.e. +-115V to +-120V, this is actual limit where the genuine IC is really destroyed, and to some extent also the 200V understood from datasheet, with a 20% margin.
OK, this answer was also found a week ago - I researched this matter more thoroughly and found the necessary information and technical explanation why X5R or X7R (really also R, not to be confused with the "U" index) are also not good, although I assumed that they could be suitable.
When they are used, especially in the lower frequency, additional distortions occur but it is not add other dangers, but these distortions is also not wanted and then I need to stay with elcos.
NP0 would be ideal, but a 22uF "battery" would take up half the Universe of the PCB.
Only 0.47uF is available in 1206 size and 50 of these do not seem conceivable considering the space layout.
Nichicon UWP1E220MCL1GB seems like a viable candidate to replace the crappy THT input electrolyte, and I can live with it.
2 pieces of UUG2A331MNQ1MS on power rails and 2 pieces of UWT1C101MCL1GB in power supply of the analog part of IC.
Basically, I can do my first test with a aluminum laminate PCB soon. I can get this PCB even smaller than my first attempt in FR4 version, which despite the single-sided copper is still smaller than this Chinese "500W IRS2092" which has spread around the world.
My very first hastily made IRS2092 module and next to it a Chinese "500W IRS2092 specimen" for comparison.
One MOSFET has been removed for its examination - they are usually fakes in these modules.
I believe that "for sake of gaining knowledge" I will go at least +-90V on this next test, but possibly even further.
Last year I spoke with one Infineon's developers and he confirmed that "In terms of reliability, he does not recommend exceeding values presented in datasheet" and at the same time he stated that "Staying within these limits, manufacturer guarantees reliability" - staying within the supply range of up to 200V means staying within the parameters indicated by manufacturer.
From an experimental point of view, he saw no reason why a +-100V power supply should not be used in experiments, but it is obvious that in a commercial application it should be held back a bit. Tolerance limit in reality is around 230-240V, i.e. +-115V to +-120V, this is actual limit where the genuine IC is really destroyed, and to some extent also the 200V understood from datasheet, with a 20% margin.
Attachments
Last edited:
The ferroelectric materials have extremely high dielectric constants (thousands), that's what ferroelectricity is really. Alas its very unstable, and goes hand-in-hand with distortion and piezoelectricity. All the high density ceramic caps are ferroelectric, typically barium titanate or its close relatives, ideal for decoupling caps on digital rails, but little else. Once you avoid ferroelectric materials you are limited to dielectric constants upto about 15 max, so the capacitor size is an order of magnitude or two larger.
https://www.ti.com/lit/an/slyt796a/...14550129&ref_url=https%3A%2F%2Fcn.bing.com%2F
Maybe it will help someone else too in future.
That wraps up this part of the story I believe?
The question of the capacitor was rather a passing one, but the topic deals more with the class D amplifier.
Maybe it will help someone else too in future.
That wraps up this part of the story I believe?
The question of the capacitor was rather a passing one, but the topic deals more with the class D amplifier.