Output devices?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Konrad said:
Driver, note the reverse diodes for the mosfets are important!

Hello,

Hi Konrad, diode D220 might actually have the adverse affect of holding the gate higher than the transistor would keep it on its own. Say it's clamping it to ~.7V when the transistor would clamp it to ~.3V. You could reduce that by using a schottky perhaps.

Regarding the parallel body diode, I fully agree with everyone there.

I've learnt that it helps the reverse recovery of the body diode simply by not allowing it to become fully saturated, which lets it recover much faster.

Forward current of a few amps is actually sufficient as the body diode handles the full current, but it should be rated to block full rail to rail voltage, same as the mosfets Vds. Rectifier schottkys are perfect for the job. Of course it's no substitute for proper timing.

I've found MOSFETs with "fast body diodes", IXYS for example, has a good variety of them, but looking at the rest of their specs and from the way they simulate, it would take one of their monster current gate drivers to get them to switch half decent. I trust the "fast diode" is nothing more than an internal schottky.

Anyone know of other MOSFET lines that incorporate such a thing worth looking at?


Cheers.
Chris
 
classd4sure said:
I've found MOSFETs with "fast body diodes", IXYS for example, has a good variety of them, but looking at the rest of their specs and from the way they simulate, it would take one of their monster current gate drivers to get them to switch half decent. I trust the "fast diode" is nothing more than an internal schottky.

Anyone know of other MOSFET lines that incorporate such a thing worth looking at?

Sorry just dropping in.

The body diode of a MOSFET is an inherent "parasitic" device. As such it can have any of the recovery characteristics other diodes have. The opportunities for optimisation are much less unfortunately, e.g. the doping profile required to obtain softer recovery characteristics of the diode adversely affects the on-resistance. Because of these tradeoffs, the body diode of a MOSFET will always be somewhat lower in performance than similarly sized dedicated diodes.

So far, MOSFET/Schottky combos are co-packaged devices. Integration of schottky diodes onto the MOSFET die has been experimented with (simple extension of the source metal should do) but with only limited success, largely because diode current would make quite a detour through the epi layer, resulting in a less than optimum resistance.

The use of separate schottky diodes is not very efficient. Suppose even a co-packaged diode (IR Fetky). Total wiring inductance is on the order of 4nH.

Now suppose you're conducting 10A in reverse though the channel. Next you turn off the FET. Under 10A the body diode will develop 0.8V, the schottky would develop 0.4V. The schottky won't be seeing any of this current for a while though. It was flowing through the MOSFET bond wires and due to the inductance it will continue to do so for a while.
At a 0.4V difference across 4nH current will transfer at dI/dt=U/L=0.4/4nH=0.1A/ns. It would take a bl**dy one hundred nanoseconds! You'll guess that waiting 100ns is the ideal way of getting lots of stored charge into that diode.

The schottky is effective in preventing storage of charge during reverse conduction through the channel when the drop across the on resistance exceeds the diode voltage. Charge storage in the diode when the gate is positive is relatively minimal, so unless you're using very old FETs (like the IRF9540 in the 100W SODA...) adding a schottky does positively nothing.

How To Prevent Stored Charge:
1) Minimise dead time. That's when charge gets stored. This is practically feasible and the effect on efficiency is very measurable.
2) Hold the gate sub-threshold instead of driving it to zero when you are turning the FET off. This is impractical and verges on the impossible, but is an interesting detail in MOSFET physics.
 
Re: MOSFET on ucd180

lbruynseels said:
Hello Bruno,

can you disclose to us which MOSFET's you used on the UCD180 modules?

thanks,

Ludo

STP14NF12FP. At the time when we selected them (2000) they were the best compromise between switching, soft recovery and cost. Actually they were the lowest cost fets available anywhere for the given current and voltage rating, and they happened to have way better than average performance :)
 
Bruno: Very generous of you to disclose your choice of MOSFET's.
This reminds me of an old speculation: how to calculate safety margins for the MOSFET's ???

What is your opinion about or method for this? I can see from the datasheet of STP14NF12FP that they have extremely low Cfb of just 30 pF, and can (safely) deliver 6 Ampere, (at Tc 100 C) . But to produce the spec'ed power of 180 Watts in 4 Ohms, you need a peak current of ~ 10 Ampere. (Vrail = 40 Volts and not taking any significant rail loss in to account).
Does this mean that your module has to be limited from reproducing full power at low frequencies, so to prevent overloading the MOSFET's?

What is the safe loadability of the UcD400?

All the best from

Lars

EDIT: Perhaps i should disclose my own method of calculating the safety margin: ;)
Maximum specified load should not exceed the DC load capacity of the output devices at 100 deg. C.
If bandwidth is limited from 20 - 20.000 Hz, (should be clearly stated in specs), current can be shared between two devices, and divided by sqrt2.

A IRFB38N20D (of the ZAPpulse) can carry 32 Ampere DC at 100 C.
This gives the amplifier a max loadability of (((2 x 32) / 1.41)^2) / 2 = 1030 Watts in 2 Ohms. (We spec. max. 1 kW). Limited bandwidth gives a loadability of 1 kW in 1 Ohms.

The UcD in the same calculation:

STP14NF12FP can carry 6 Ampere DC at 100 C. This give the amplifier a max loadability of (((4 x 6) / 1.41)^2)/4 = 72.4 watts in 4 Ohms. You spec 180 Watts in 4 Ohms.
At limited bandwidth you get 1.41 times 6 = 8.46 Ampere.
Then at my safety limits (calculated the same way as above) gets you a safe output power of: 144 Watts in 4 Ohms.
 
When I browsed the ST Microelectronics I was impressed by the difference between their MOSFETS AND HEXFETS.

For example, in 1996 the company (STMicroelectronics) announced a new technology known as Mesh Overlay that significantly improves the performance of power MOSFETs, leading to smaller die sizes, lower costs and higher performance.

The parasitic capacitances are greatly lowered permitting much faster switching. Channels are placed adjacent to each other in a row. That way, they can be shorter and smaller and still maintain the same d-s voltage rating. Yet one must be cognizant that HEXFETS possess greater power handling since their bigger channels have more thermal mass and more surface area to transfer heat.
 
subwo1: Thanks for this tip ;)

I checked out ST's homepage, and found the closest alternative for our use, the STP40N20. Even though the rds on is slightly lower on the ST device, it can only withstand a load of 25 Ampere, compared to IRFB38N20D's 32 Ampere (both at 100 C).

The capacitances are :

STP : 130, 580 and 3500 pF
IRFB: 73, 450, and 2900 pF.

So the IRF is better in every aspect except the rds on.
What does this mean in real life?

At full load 580 Watts in 4 Ohms, we have an RMS current of: 12.06 Ampere. With the IRF's 9 millioms higher rds on, we get a serial loss of:

12.06 x 0.009 = 0.11 V * 12.06 A = 1.3 Watt serial loss.

I think this loss is more than compensated for by the IRF's lower capacitance, and hence lower idle loss.

On the other hand the ST devices would limit our safe operating area to: 625 Watts compared to the IRF's 1030 Watts in 2 Ohms. So in effect 2 Ohms loadability would be off the table with the ST's where the IRF's can make it possible.

No i think all in all the IRF devices are better than the ST's. The ST's may be cheaper though, i haven't checked. No matter what we stick with IR ;)
 
Hi Lars,

I think most can agree that the "180W" value tagged onto the small module is slightly optimised, given the commercial nature of the venture. The same circuit is known here as a 150W/6ohm amplifier. Nevertheless, since audio circuits do not as a rule process continuous sine waves, the 100 degrees C are unlikely to be achieved on music.

Now, I would have taken action to keep JP from publicising a 180W number if it was unsafe. One should keep in mind that the 6A figure is for an overmold package and 9A for an uninsulated package. The reason is that "case" temperature for one is the metal surface for the TO220 and the plastic surface for the TO220FP, both being the first "externally accessible temperature".

Curiously, if one uses noninsulated TO220's with impregnated glassfibre thermal washers, the situation is very similar. The thermal resistance of these is typically 5K/W (Bergquist Sil-Pad 400) to 2K/W (Bergquist Sil-Pad 2000), on average very comparable to the 3.5K/W that the overmold (full-pack) TO220's add compared to the noninsulating ones. To compare apples to apples we can elect to measure temperature at the heatsink (accessible in both cases) or at the TO220 metal plate (accessible if you chip off some of the plastic on the face side).
You’d find the temperature is roughly the same for both (which is better depends on whether you’re using standard or good sil-pad material).

We may wish to do the same exercise with the IRFB38N20, and see what the current handling will be for a heatsink temperature of 100 degrees, presuming normal mounting practics.
The 32A figure corresponds to the maximum junction temperature of 175 degrees (75 deg above 100 deg case). Cross-check (never trust a data sheet if the numbers don’t jive): Ron at 175deg becomes 155mOhm. At 32A this is a power dissipation of 159W. Rated Rth,j,c=0.47K/W, yielding 75 deg temp rise. Spot on. Good data sheet.
Now insert a 3.5K/W thermal washer (I do not know what insulator you are using, fill me in on the exact numbers). Thermal resistance to the heat sink becomes 3.97K/W. At 100 degrees heat sink temperature we can dissipate 19W before junction temperature goes out of spec. This corresponds (again at 155mOhm Ron) to 11 (ELEVEN) amperes, rather a far cry from the promised 32A.
Your own suggested rule for wide-bandwidth would give us no more than 242W at 4 ohm or half that at 2 ohm.

Luckily for both of us, it doesn't work that way.

The metal tab (external on the TO220, internal on the TO220FP) does not respond to 20Hz in a significant manner. You will find this info on the thermal response graphs on page 4 of the data sheet (mind the horizontal scales are different). Otherwise put, the currents found (6A for the STP14NF12FP, 14A for the IRFB38N20 using Silpad 2000) may be used with presumption of "power sharing", even at 20Hz.

Under this rule, at a heat-sink temperature of 100 degrees, the UcD module remains within spec up to 144W (also at 20Hz) and the Zappulse modules will do 484W.

Nevertheless, it is possible to get most from the TO220 uninsulated package, without having to forgo electrical insulation. This is done using ceramic (alumina) insulators. A typical 1.5mm thick TO220 Al2O3 insulator has around 0.4K/W worth of thermal resistance. Counting on this requires excellent planarity of the TO220 and the heat-sink in order to allow the thinnest of grease films to be applied. It shall be no surprise that the UcD400 modules are constructed in this manner. Recalculation of the DC current rating at 100 deg heatsink temperature and a combined Tj,sink of 1.6K/W produces 21A. By anyone's rules this is more than sufficient for 400W into 4 ohms. The 400W rating of the UcD400 is therefore no hyperbole (the voltage restriction of 63V is produced by the use of 63V elcaps – moving to higher voltages to match the +/-75V rating of the FETs is under consideration).

It may be of comfort to you that there is no standard requiring an amplifier (for consumer use) to deliver full rated power continuously at a heatsink temperature of 100 degrees C.
The most stringent test used to verify commercial power ratings of end-user equipment is the IHF (FTC) power test. The amplifier was first run at 1/3rd rated power for an hour, then full power for 5 minutes ("was" because the test now specifies the pre-heating at 1/8Pr, who said there was no audio industry lobby). During this test, there is no requirement for all parts to remain within their ratings (!) as long as the amplifier delivers its rated power at the distortion quoted for this power. Under all circumstances, music is a much more relaxed condition, even in this age of hypercompressed CDs.

Cheers,

Bruno
 
Hi Bruno

Thanks for your thorough answer :) Just to exact the details, here is a short recalculation:

We use the st. Gobain typ. 1877 insulator which has a thermal resistance of 1.4 K/W for a TO220 package. (0.24 K In2/W).

The total t resistance from junction to sink is: 1.87 K/W.

Using your method of calculation we have 15.8 Ampere
to work with, at 100 C in each device. Corresponding
output power (without extending the power according to your
interpretaions of the graphs on page 4) is 499 Watts in 2 Ohms, and 998 Watts in 4 Ohms.

About UcD400, the ceramic insulators are a really nice touch!
Just another question, can the supply voltage be raised if the capacitors were to be exchanged to 100 V types?

All the best from

Lars
 
Lars Clausen said:

About UcD400, (...), can the supply voltage be raised if the capacitors were to be exchanged to 100 V types?
The FETs are 150V devices so the overvoltage detection would be set to trip at 75V.
For a normal PSU design, you'd go for an idle voltage of 67V under nominal mains condition so it doesn't exceed 74V at 10% overvoltage.
That makes it practical to get the actual 400W out of the modules with a realistic supply (with the current 63V setting this is difficult).
 
How can I get a shemetic of UcD?

classd4sure said:
There's a good choice of high end drivers out there that will allow you to do it easily enough (non you can actually simulate with as far as I know), and if you look at the UCD circuits around here you'll see a discrete method which should provide some good insight as to how it's done discretely, and which do simulate easily enough.

Cheers,
Chris
 
Hi,

Welcome to the forums.

You can find some in here:

http://www.diyaudio.com/forums/showthread.php?threadid=36852&perpage=10&pagenumber=1

If you dont' want to read the whole thing, skip to the end there's a fairly simplistic, low power one there, made for parts I had around here, I built it on a protoboard.

Here is an example at a full bridge attempt, I think it still needs improving, ie. the comparator outputs are still susceptible to current hogging. I think I negated the benefit of the cascode in my revised version, but it created some time delay problem with other configuations for some reason.

I'd like to do away with the summing op amps, or preferably, all of them, having only a discrete class A input stage. This circuit has not been built. With some tweaks to the input stage (which were found to be meaningless in spice) it gets down to the mid 0.00* THD range. As it is it simulates very well. I might work on it more some day. Better gate drivers would be a big help too, it is too easy to load these ones down!~

http://www.diyaudio.com/forums/showthread.php?s=&threadid=40307

Non of these schematics are good for AC analysis unfortunatly, but work well enough in transient. If you have orcad and want to try one of the attached .dsn files it won't open it with pspice. Best workaround is to open the .dsn file in orcad, create a new project, copy and past it into the new project, then you can simulate it.

Feel free to add your feedback to those threads if you like.
Regards
Chris
 
Kenshin said:



when the PSU is turned on, the couple capicitors needs charging. Will the charging current turn on the BOTH MOSFETS and cause short circuit?

I'm pretty sure you'll see some pull down resistors on the gates to prevent this from happening. In fact I think it's mentioned in one of the posts here.

Kenshin said:
something looks complicated -- more complex than what I had expected (open collector output stage & voltage follower & boost at the high side to power the voltage followers).


It might seem complicated but look at this way, where's the level shifters?
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.