3

In power converters, it is common that MOSFETs are fed a PWM signal to have cyclic switching. Part of the power is lost in the process. Does the body diode reduce this loss? It seems most power MOSFETs have the diode included.

HYQ
  • 749
  • 1
  • 10
  • Sometimes the body diode is an asset, and other times it is a liability. But as others have said, it is an intrinsic part of the transistor, so it is not included by choice. – user57037 Aug 27 '22 at 08:08

3 Answers3

6

You say they "have the diode included", implying that you consider it something separate from the MOSFET; that's a bit of a misunderstanding, I think. The body diode is an intrinsic part of the construction of a MOSFET; it's literally impossible to make a MOSFET without one. So it doesn't really make much sense to consider whether it helps with losses or not, as you can't avoid having it if you're using MOSFETs.

Hearth
  • 27,177
  • 3
  • 51
  • 115
  • 1
    Actually there used to be 4 pin MOSFETs available, where the body had a separate pin from the source. So your "not possible" is in error. – Math Keeps Me Busy Aug 27 '22 at 04:23
  • @MathKeepsMeBusy The body diode is *always* there in silicon MOS, you just get a choice of where you connect it in a 4-pin device, vs. a 3-pin device where it's connected to source. You can still buy such devices. CD4007 has two on one chip, for well under $1. – Kuba hasn't forgotten Monica Aug 27 '22 at 05:34
  • @Kubahasn'tforgottenMonica You are right. I was misunderstanding what you wrote. – Math Keeps Me Busy Aug 27 '22 at 11:35
  • @MathKeepsMeBusy You can still get those, but they still have a body diode. The body diode (or perhaps in that case you could say body diode*s*) are between the body and source, and body and drain. The body-source one is irrelevant in three-terminal FETs, as it's shorted out. – Hearth Aug 27 '22 at 12:03
4

The body diode is unavoidable in silicon power FETs. Gallium Nitride (GaN) devices do not have such a diode. Because DC/DC converters generally need a deadtime between switching events on each switch to avoid any risk of overlap and simultaneous conduction of devices (which would cause a short across supply and destruction). Since the inductor will cause current to flow continuously, the body diode allows the inductor current to flow even when the FETs are 'off'.

GaN converters either need a separate external diode or conduct in reverse when the gate-drain becomes forward biased (e.g. the drain is more than VTH below ground).

winny
  • 13,064
  • 6
  • 46
  • 63
jp314
  • 18,395
  • 17
  • 46
  • 1
    GaN MOSFETs would have a body diode; it's just that GaN devices are usually HEMTs, not MOSFETs. It's not related (directly) to the fact that it's GaN. (There is also no body diode in JFETs, or in IGBTs, for instance.) – Hearth Aug 27 '22 at 12:05
  • That is correct for GaN devices. Systems with IGBTs usually do need external diodes to be added. JFETs are not used in power converters -- power JFETs are not available and the normally-on characteristic would make them cumbersome for this. – jp314 Aug 27 '22 at 17:37
  • Power JFETs do exist, primarily because they're substantially easier to make in SiC (for reasons I could go into if you like) than MOSFETs. Many "SiC MOSFETs" on the market are in fact SiC JFETs cascoded with a low-voltage but high-current Si MOSFET. You can also get bare SiC power JFETs--at least you could before UnitedSiC got bought; I don't know if they're still available. A very niche device, for sure, but one that does exist. – Hearth Aug 27 '22 at 17:48
  • For an assembly with many parallel MOSFETS to avoid an intrinsic body diode, it would need to have source, drain, and gate connections on the same side of the bulk substrate, which may effectively require an extra metal layer, with a selectively-masked layer separating it from the layer below. That would add to manufacturing costs, but I don't think that would imply that the diode is "unavoidable". The question would be whether eliminating the diode could improve the in-circuit performance of the device enough to justify the cost. – supercat Aug 27 '22 at 18:42
  • In an efficient power MOSFET, the diode is unavoidable. But more fundamentally, the diode is generally NECESSARY in a DCDC converter. A diode with lower VF and/or faster reverse recovery would be better. That is why Schottky diodes are sometimes used in parallel with MOSFETs. – jp314 Aug 27 '22 at 18:55
2

A more interesting question is when the body diode increases, or decreases, switching or conduction loss:

The switch having a diode, which does not increase its capacitance (the body diode is intrinsic, you can't buy a Si MOSFET without it so there is nothing in its class to compare to), and only carrying forward current (as in the buck high-side or boost low-side), is simply irrelevant. Devices that do have optional diodes (co-pack IGBTs) will see slightly increased loss, due to the additional capacitance.

The rectifier (buck low-side, boost high-side) having a diode, arguably greatly reduces conduction loss, by mere fact that, if not for that diode action, it doesn't buck or boost at all(!). But being active devices, they can also be made to switch synchronously, and with perfect timing, conduction losses are greatly reduced, and switching losses are defined by total device capacitance, and stray inductance.

So, synchronous switching, without body diode, is tricky. It's also tricky with:

The problem is this: without body diode, voltage is either unlimited (ideal switches), or limited to somewhat over Vgs(th) (MOSFETs), either way the conduction loss can be quite a bit higher during that small gap (dead time) between switch turn-ons. But the problem with the body diode is, at the start of a cycle (sync turn-off, switch turn-on, i.e. buck rising edge / boost falling edge), load current flows into the body diode, briefly forward-biasing it, then the switch yanks it off in forced hard switching, incurring reverse recovery loss. Even worse still, due to the short forward-bias period, charges aren't distributed evenly through the body junction, and drift step recovery can occur (trr goes down when very short forward-bias pulses are used; it also becomes sharper, i.e. softness factor goes down; at short enough time scales, sharpness can go to ~zero and extremely fast risetimes result!).

Forward bias can be avoided by using a schottky diode in parallel, but this also adds more capacitance -- increasing switching loss. And it needs to be quite a substantial diode too. It's also not possible at high voltages, where low Vf Si schottky are unavailable, and SiC schottky are marginal at best. (At low voltages, there are even monolithic "FETky" type MOSFETs available, adding somewhat less capacitance than discrete parts, and making sync. rect. design much easier.)

I recall one -- believe it was for LTC3810 -- eval board, that had terrifyingly bad EMI, like 2-3ns wide pulses, which could be measured essentially anywhere on the board. And it seems like they had some idea about it, but sheepishly slapped in a mere 1A schottky diode -- woefully insufficient, its Vf is much too high to do anything at all -- and in fact makes things worse by adding a little capacitance and doing nothing else!

Tim Williams
  • 22,874
  • 1
  • 20
  • 71
  • 1
    The LTC3810 is a pretty messy regulator in general, going by my experience using it. It's quite sensitive to EMI too, or at least that's what I eventually decided was causing it to operate at a very unstable switching frequency. (Part of that was my fault with a poor layout because I was rushed to get something working, but all the comparators dictating timing in the 3810 certainly didn't help.) – Hearth Aug 27 '22 at 12:47