Class D amp with 300v mosfets

Thank you for your advice. I had emergency protection and nothing bad happened with the prototype, only the MOSFETs shorted between drain - source, and the 8244 driver still runing. I've changed the mosfets and now it's ready for the tests. This time i have increased the load to 6.4 ohms and nothing has happened yet ...

An externally hosted image should be here but it was not working when we last tested it.


This means 1000-1200w in 6.4r

@Chocoholic:
I also thank you for your advice. For a safe operation at this voltage and load, even in sinusoidal mode, what option do you recommend:
- irfp4868 with schottky diodes and antiparallel diodes
- 2 pairs of C3M0065090D
 
Last edited:
Then I suppose there were fuses between the small filter caps in the PCB and the big bulk caps, that got blown before the FETs actually got fused internally. The driver ICs pop easily when the source lead to die connection of the FETs fuses. The energy stored in bulk caps is usually enough to fuse source lead to die connection. In more sophisticated circuits, and when space used is at premium, the fuses are often replaced by electronic protections, which also saves the voltage drop across fuses and their inductance, which is quite inconvenient for fast switching. The problem with too much inductance between main bulk caps and local decoupling caps is that more HF current flows through local decoupling, shortening the life of the electrolytics. btw: running the channels in "quadrature" helps to cancel out ripple in supply capacitors, mostly at low-mid signal levels.
 
Ionut,

Your mosfet is not so fast with Reverse recovery time at 152nS and Reverse recovery charge around 844nC which
actually increases at high temperature operation.

Switching losses due to the Qrr of the body diodes.
The body diode is specified with a Qrr of 844nC-1689nC at 25C and di/dt=100A/us. If you assume a hot junction and di/dt around 300A/us you can easily expect a Qrr of 3000nC, unfortunately the data sheet gives no information under realistic operating conditions.
The resulting losses are immense.

In order to remove Qrr an additional 20A peak leads to a total current peak of almost 40A while the Vds still equals almost full rail.
Under this load condition it takes slightly less than one second until
the junction reaches something like 175C :D

What do you think are the causes? Qrr?

Enough of references from me and Choco.....its QRR only, your mosfet is lousy:D

Use Combination of regular Mosfet with series schottkys + inverse parallel FRED or GAN OR SIC devices.
 
Last edited:
Some more words about power handling of MosFets and resulting junction temperature. The max. allowed power dissipation in the data sheet is valid only if we keep the casing of the MosFet at 25C, which we simply never can achieve in reality.

Furtheron in my experience the thermal resistances between case and heat sink are often higher than anticipated.
For a TO-220 I measured the following resistances from case to heat sink depending on the way you mount it.
TO-220 directly mount on heat sink with thermal grease: 0.26 K/W
TO-220 mount with glimmer pad and thermal grease: 0.96 K/W
TO-220 mount with AL2O3 pad and thermal grease: 1.17 K/W
TO-220 mount with silicone pad: 3.5 K/W
TO-220 mount with thermal grease and undefined ceramic pad from professional electronics shop in Guangzhou : 4.7K/W

The example with IRFB4227 was with a DC output load of -20A DC and a fs=280kHz.
Similar like I showed the turn on losses one has to examine the turn off losses and
of course we should not forget the conductive losses. In total the ITFB4227 has to handle a power dissipation of approx 100W in the shown example.
Attached the dynamic thermal modelling from junction to heat sink.
I anticipated that the heat sink would be at 50C and large enough that its change of temperature within 2s is neglectible when 100W are applied.

Translation from thermal circuit into equivalent electrical circuit:
Thermal capacity, J/K ==> Electrical capacitance F
Thermal resistance, K/W ==> Electrical resistance V/A = Ohms
Thermal power, W ==> Corresponds to current, A. (Ref. K/W corresponds with V/A)
Temperature, K ==> Corresponds to voltage, V. (Ref. K/W corresponds with V/A)

Values in the equivalent circuit are derived from data sheet + measurement of thermal resistance from case to heat sink and calculation of thermal capacitance of TO-220 geometrie and materials.

In less than 1s the junction reaches 175C, which is fitting perfectly to all other observations.
 

Attachments

  • EquivalentCircuit.JPG
    EquivalentCircuit.JPG
    43.6 KB · Views: 573
  • ThermalResponse.JPG
    ThermalResponse.JPG
    67.8 KB · Views: 538
@Chocoholic:
I also thank you for your advice. For a safe operation at this voltage and load, even in sinusoidal mode, what option do you recommend:
- irfp4868 with schottky diodes and antiparallel diodes
- 2 pairs of C3M0065090D

It depends on your personal goal.
If your deep desire is to to learn how to layout and handle more complex switching stages, then approach with IRFP4868+shottky+antiparallel diodes is the way to enlightment.
If your goal is just to get the amp running without too much hazzle and standard PA performance in THD is sufficient, then the C3M0065090D is the way to go.
 
SiC FET specs are good, but consider pricing:
- C3M0065090D: $9/1k units
- SCT3030AL: $20/1k units
- IPP410N30N: $4/1k units
- IRFB4227: $1.6/1k units
- PHP33NQ20T: $0.67/1k units
- IRFB4615: $1.1/1k units
- PHP28NQ15T: $0.53/1k units

For class D, a half bridge of SiC FETs is not likely to compete in price with a full bridge of Si FET for the next 10~20 years. A further economic slow down for the next 10 years is anticipated. There is also new IGBT technology, trench IGBT, offering crossover performance between IGBT and super-junction Si FET. "Trench" means that the die is arranged as many little transistor cells, resulting in just about 2x the turn off loss of a Si FET in usual conditions. Anti-parallel diode can be chosen between slower or faster depending on model. Main difference is that these IGBTs can be produced using current facilities and materials, while SiC requires different process and ingredients, inherently more expensive. German guys should be more aware of Infineon progress.

And concerning the RF-based isolated gate drivers applied to audio amplifiers, the 200ps p-p jitter (quantization actually, as ON time must be a multiple of RF oscillator period) is likely to increase modulation noise floor and reduce SNR, although for PA applications this is easily masked by loud output and system noise floor, not the case for low volume listening.
 
German guys should be more aware of Infineon progress.

You should know: At least one German guy is wondering what must have happened to you during the last six years.

Can't talk for all German guys, but from my perspective a SiC Fet is more attractive for DIY than an ultrafast IGBT. E_ts of a corresponding SiC Fet is typically 2-3 times less. From commercial view the IGBTs are of course attractive in terms of cost/feature ratio.

Regarding other progress in Infineon... I am scratching my head when they will extend their offer of GaN to everyone.
TI has started this already pretty some time ago. Infineon seems to run a different strategy.
 
Unfortunately semiconductor market got somewhat effeminate (oestrogenic: the hormone of attraction/speculation, zero benefit games based in finding the backdoor in the brain of the other to make him believe there is benefit) with the introduction of a bunch of new players due to massive investment in renewable energy development, not free from corruption. This means aesthetics got into play, new offers are not necessarily technically superior to previous ones considering all planes of existence, and equivalent offers may be disguised to appear different. I'm only focused in designs that can be both built for DIY or mass produced for profit with long market lifetime.
 
Nice combo:

An externally hosted image should be here but it was not working when we last tested it.
With TO-247 and SI8274( as it can run in +/-750V)I believe it can do 1000RMS into 8ohms and at least 1800W in 4ohms easily?
Why not use SI8244 as its drive capability is about 4A.

Are there any TO-247 Mosfets in 300V package to use with SI8274,SI8244
 
Last edited:
Compare Eoff vs. collector current chart for latest 650V IGBT vs. SiC FET.

IGBT: Page 10, figure 12:
https://www.infineon.com/dgdl/Infin...N.pdf?fileId=5546d461464245d30146a51147116d12

SiC: Page 10, figure 20:
http://www.rohm.com/web/global/datasheet/SCT3030AL

70A: 0.9mJ (IGBT @ 400V) vs. 0.67mJ (SiC @ 300V)
10A: 80uJ (IGBT @ 400V) vs. 30uJ (SiC @ 300V)

Consider that, at 400V vs. 300V, loss can be up to 33% higher. Consider that SiC can turn off a bit faster with negative gate voltage, but only up to -4V are allowed.

And the difference vanishes in power supply applications where a few-nF capacitor in parallel with switching device is allowed. This is the reason SiC is going to impose in a few fields only, those where the capacitor in parallel with transistor is not permissible (or those where increased IGBT loss at low current is not permissible), and thus it will remain expensive for decades or forever. (Does someone remember lateral FETs?)

SCT3030AL costs ~$19/1k units, IKP30N65F5 costs ~$3.5/1k units

These new parts are mostly a "conspiration" of nature for chasing too emotional designers and companies hehe
 
Lets do the statistics: There are a finite number of application fields in switching electronics for fast (>20khz) unidirectional switching, There are 5 cases:

1a- No reverse conduction needed, no C across switch permissible
1b- No reverse conduction needed, C across switch permissible
2a- Reverse conduction needed but faster is not advantageous, no C across switch permissible
2b- Reverse conduction needed but faster is not advantageous, C across switch permissible
3- Faster reverse conduction needed (>100khz >150V hard switching)

Lets combine this data with available semiconductor technologies and final application types:
- Audio amplifiers are among the highest switching precision/bandwidth requirement in power electronics. Audio amplifiers are located between case 2 and case 3.
- Most (if not all) non-precision power conversion applications (power supplies) can be solved with cases 1 or 2, at same or lower cost than with case 3, only difference is that case 3 can produce smaller equipment due to lower part count and size.
- For higher-power lower-precision conversion, cases 1a and 2a can always be turned into cases 1b and 2b, with increased part count, but increased efficiency.
- Cases 1a and 2a are already well covered by superjunction FET
- IGBT has unrivaled cost/performance for cases 1b and 2b (with co-packed ulta/hyperfast diode).
- SiC and GaN are only technically advantageous for case 3.
- Case 3 represents a small fraction of total equipment using switching transistors produced.

Additionally I have to tell to ionutgaga that the 300V FETs wouldn't probably have failed in 4 ohm test if the modulator had been self-oscillating, because self oscillating idling at 250~300khz drops to below 100khz near the rails, where current and switching loss is highest, this involves a reduction in switching losses around 4:1 w.r.t. clocked (which is similar loss reduction than going from Si FET to SiC or GaN in a clocked modulator).

So in the end: better modulation and circuit/layout techniques, with latest Si FET or IGBT, result in similar or lower losses than upgrading to SiC/GaN a clocked modulator. Thus the bottleneck for further development is the clocked modulator. And in self-oscillating: Surprise!! At high current switching frequency is lower, so improvements in body diode only manifest themselves in total FET loss divided by ~8 (twice the 4:1 clocked vs. self-osc switching loss ratio, representing that only half or less switching loss is related to body diode).

"I'm not going to pay several times more for a body diode several times better, because this is already being compensated in another way."

The savings due to simpler development are expent in exotic semiconductors haha
 
Last edited:
Note: the 4:1 reduction in switching loss comes from a linear factor plus a quadratic factor, as at lower current snubbers can handle most turn-off switching loss, while at higher current, as snubber (and circuit capacitances) action is not enough, turn-off losses increase faster. A typical ratio of conduction to switching loss is 1:1, so impact on total loss of frequency drop is not as important as the impact it has on the relative importance of body diode recovery loss. Reducing frequency as output approaches the rails does not only cut body-diode related loss, it cuts all switching loss.
 
Last edited:
Evita, You have neglected another important parameter which is Td_off , For SIC it's less than 50nS and the IGBT you are showing has it in excess of 174nS which will only increase to once the temperature increases ->>204nS. Now think of its impact on selecting the deadtime for the application such as audio amplifier, everyone knows increasing the deadtime increases the distortion proportionately. Just by comparing E_off solely and saying SIC is utter waste is nothing but propaganda to show SIC in bad light. Everyone knows SIC and GAN have distinctive advantages which cannot be neglected.

If IONUT wants to use SIC with clocked modulator, what is your problem?
Choco and I feel GAN and SIC is definitely here to stay and change the future.
 
Last edited: