# How to do high voltage high-side switching in a simple and power efficient way?

## Background

I am planning a system that needs to maintain low power draw (especially when in its standby state), which means I need to use some switches to turn off parts of the system when not needed. The main power supply in the system is a lithium ion battery which has a voltage range of 50-84V DC from 0 to 100% SOC.

Here is a block diagram of part of the system:

Initially it seemed like the obvious choice for a switch would be a logic level N channel power mosfet in low-side configuration. This is because they do not draw current in order to keep them switched on or off, have a very low resistance so waste a low amount of power, and in low-side config can be driven by either the ESP32 GPIO or via a level shifter to 5V to give better Rds(on).

In comparison, mechanical relays draw constant power to stay on, and can have contact resistance 10X the Rds(on) of a mosfet in some cases. It appears they also wear out much quicker. And solid state relays that I have found so far are generally very large compared to a mosfet or relay and drop a constant voltage up to 1V across them, which would be a big problem for high side switching of say 5V.

However, the key problem with the low side switching NMOS solution is that if you switch off the ground to a part of the circuit, for example switching off the ground to the 5V LDO that powers the PIC in my diagram, then that part/power domain will now float at whatever the Vcc is (unless it is isolated), so in this case up to 84V. For a simple load this might not be an issue, but in my case the PIC microcontroller takes 1 digital (D0) and 1 analog (A0) input from the ESP32 (highlighted in red), so it would ground itself through those data lines into the ESP's power domain which would apply 84V onto the PIC/ESP which would cause damage.

Therefore, high-side switching seems more appropriate as that means all power domains will float at GND when turned off, so leaving data lines connected between domains would at worst put 5V onto an IO pin when the MCU is powered off in another domain, which isn't as dangerous, and this can also be mitigated by setting the outputs going into a domain to low when that domain is powered off. This is what I have shown in a generic way in the diagram with Q1, Q2 and the two gate drivers.

It seems the workaround for low-side switching is isolated communication between domains, which is commonly done with opto-isolators for digital signals, but these draw current when being used. In addition, from my research it seems passing analog signals between domains (which is required) in an isolated manner would require some complex circuitry such as an analog to frequency modulator and demodulator on the other side.

## Question

I have been having trouble identifying the best way to do high side switching with either PMOS or NMOS at 84V. I tried looking at gate drivers, or load switches (which it seems are mosfets + gate drivers combined for convenience) to do the heavy lifting for me, but it has been very difficult to find any drivers rated for a max Vcc of 84V or higher on UK sites such as Rapid Electronics, RS Components or Farnell, especially in through-hole format.

From doing some research, it seems there are some high side gate drivers that do not have the full Vcc across them and instead derive a separate voltage from the main power supply, so they do not need a Vcc rating of 84V.

For PMOS, one example is using a zener diode and some resistors to create a defined voltage drop to create the required negative Vgs, as just pulling the gate to ground would mean Vgs is -84V which is far too high.

For NMOS, some methods I found to generate a voltage higher than Vcc to turn on the gate include:

• Bootstrap supply using a capacitor, diode and some resistors
• Charge pump
• Magnetically isolated/floating supply derived from Vcc

All of these methods have tradeoffs of course, so my question is between all the options (including any better ones I haven't listed), what is the most power efficient and low component count gate driver + gate driver supply + mosfet combination for simple on/off switching? Alternatively, lowest power + component count load switch if applicable. This includes:

• The quiescent/leakage power used by the gate driver and its supply in either the mosfet's on or off state
• If a PMOS high side driver + supply is much more efficient than any of the NMOS options, then increased power wasted due to higher Rds(on) of PMOS compared to NMOS should be factored in too.
• Low component count is simply to keep the gate driver + supply combination small and easy to assemble by hand on a through-hole solution such as a stripboard since it will be need to be duplicated once for every mosfet switch used in most cases. It is possible that I could design and order a PCB to mitigate this issue if it's necessary to use surface mount or a lot of components, but I would like to avoid that if possible as it would add time and complexity.
• The solution must be able to switch up to 3A without a heatsink. For a gate driver this is irrelevant as it depends on the mosfet, but for a load switch with the mosfet integrated this spec would be more important.
• Have you considered a latching relay? This may require additional components to drive said relay, but does not require constant current in either state.
– TLW
Commented Oct 3, 2023 at 11:53
• @TLW That's a good suggestion - I wasn't aware of latching relays before but they certainly seem like an interesting option since they solve one of the problems with relays (constant current to power the coil). Just from looking at the datasheets for some dual coil latching relays it seems the two remaining issues are the 100mΩ contact resistance (compared to 32mΩ for an NMOS), and the electrical endurance of only 6000-50,000 operations depending on the relay. For a 3A 5V load, 100mΩ is still 94% efficient so that might be acceptable, but the endurance does concern me for their long term use Commented Oct 3, 2023 at 17:20
• As contact resistance is one of the major limiting factors of the current-carrying capacity of a relay, you'll typically see high contact resistance for 'small' relays - because if they had lower contact resistance they would be rated for more current. As a result, you can get much lower contact resistance if you 'oversize'. If you look only for low-carrying-current relays, yeah, you'll see 100mΩ contact resistance. Do you have to switch 3A, or can you e.g. turn off the downstream regulator first? 3A @ 84VDC is a bit much for many relays.
– TLW
Commented Oct 4, 2023 at 12:51
• If you do go down that route, beware that relay datasheets are, uh, misleading. You'll often see e.g. '3A 115VDC 100k cycles', when they actually mean '3A' or '115VDC' or '100k cycles', but not at the same time - switching capacity and/or longevity rather diminishing with higher current. Also beware that some relays have minimum switching current requirements (long story short: keeps the contacts from building up oxidization.)
– TLW
Commented Oct 4, 2023 at 23:19
• @TLW Exactly! I've been looking at some data sheets and long story short, an Omron G6C SPDT relay rated for 8A is actually only rated for 8A at 30V DC, and approx 0.6A at 84V DC due to derating above 30V DC. Meanwhile, they can take 250V AC at 8A no problem due to the zero crossing property of AC greatly helping to extinguish arcing. So this property alone requires seriously "oversizing" the relay specs required, in order to achieve a derated current of at least 1A at 84V DC. Commented Oct 4, 2023 at 23:40

A few options:

1. High side charge pump based driver like ltc7001. Pricy, but good performance and low power. Other parts like this exist but not many for ~100V

2. Photovoltaic based driver like the vom series (see VOM1271). Will use ~10ma but from your low voltage regulator so not a ton of watts.

3. Pfet switch. Can be low power but the lower power it is the slower it will be to shut off. May be fine. PFETs cost more for their current rating so they look worse at higher currents. At 3A this is viable.

• +1 for the VOM1271, the easiest way to use n-chan high switches from an MCU; as easy as driving an optocoupler. Commented Oct 3, 2023 at 22:10
• Thanks a lot for these suggestions! I've evaluated the first 2, but just wanted to check what exactly you meant by the last one. Are you just talking about a P-channel mosfet high side gate driver, or do you mean a high side load switch with an integrated PMOS? And could you give an example of an IC or circuit that you are referring to that would work for 84V? Commented Oct 4, 2023 at 22:59
• @ScottyN91 Yes I just mean a pfet or PNP. You’d pull down the gate/base with a smaller ground referenced 100V+ transistor. Pfet needs Zener protection, PNP doesn’t but need sufficient base current and associated loss. Commented Oct 5, 2023 at 0:03
• @asdf30 Thanks, I've accepted your answer as you provided several different solutions. I've also added an answer (since it was too long to fit in a comment) with my assessment of the pros and cons of the solutions you provided as well as those from other users on this question. In the end I think I will go with the VOM1271 as it seems to be the most power efficient and simple solution. Commented Oct 6, 2023 at 17:09

High side switching with a charge pump will work well. The current pump is off when the NFET is off. Idle current is still zero. Active current is only AC leakage via diode capacitances so tiny as well.

As for how to make the charge pump: AC square wave from the microcontroller GPIO pin set to high drive, through a capacitor, clamp diode to high side, then another diode to a holding capacitor. Do that with Schottky diodes and configure one or two more stages as a voltage multiplier and you’re set. Or step up the GPIO output to 5V and then use that - no need for a multiplier then. The NFET will turn off slowly, and if that’s not desirable then the gate should be driven from a rail-rail output comparator with built-in reference. Once the pump voltage goes below some minimum, the gate gets turned off. It also will get turned on quickly on power-up. Add some hysteresis of course.

isolated communication between domains, which is commonly done with opto-isolators for digital signals, but these draw current when being used

True. That’s why you’d design it so that when load is to be off, the isolators are off as well :) That’s the trick. And then you realize that if you are controlling power to the high side of the isolator anyway, there’s no point to the isolator anymore in your application.

• Thanks for the suggestion - would you say what you described is similar to what the LTC7000 IC does, or is it better than that in some ways? For reference, the data sheet says it uses 250uA when the charge pump is on/regulating, but this is taken from the Vcc of the mosfet as far as I can tell, so that would be 21mW leakage power at 84V. As for the optocouplers, I understand you would turn them off when not needed, what I was describing about having isolation was just for if you need to use low side switching. As you said, it's not necessary when combined with high side switching :) Commented Oct 5, 2023 at 0:51

After reading the other helpful answers from TLW, asdf30 and Kuba hasn't forgotten Monica (thank you all), I thought it would be helpful to add my understanding of the details of their different solutions including their pros and cons. This is mainly for others to see who want to choose a solution to do high voltage DC high side switching with minimal power wasted.

## LTC7000/LTC7001

Charge-pump based gate driver IC that sits directly across the mosfet’s supply. LTC7001 is the simpler and cheaper version of the two. It only draws 250uA when on (or 1uA when in shutdown for LTC7000 only) but this is from the mosfet’s supply, so at 84V that’s still 21mW.

Pros:

• Fast turn on and turn off times (maximum of about 160ns inc. propagation delay for a mosfet with 10nF gate capacitance)

Cons:

• Expensive (about £6.66 or more in the UK)
• Can only go up to 135V supply/Vds
• Only available in surface mount format, so requires a PCB to be designed and manufactured.

## VOM1271

Photovoltaic based gate driver that can run directly off a GPIO pin with a resistor. Similar to an optocoupler except instead of generating current into the base of a phototransistor it generates a voltage/current directly onto the output. Draws a minimum of 5mA, but that’s from GPIO level voltages, so at 3.3V that’s only 16.5mW which is less than the LTC7000 series.

Pros:

• Fully isolated like an optocoupler, so can withstand up to 3750 rms. So most likely your high side voltage is only limited by the mosfet’s Vds rating, rather than the driver.
• Has a max forward voltage of 1.4V, so can run directly off the GPIO pin of any microcontroller that has a Vcc greater than that voltage, with a series resistor. This also means only 2 components are required for the complete gate driver circuit.
• Cheaper than most of the other options at about £1.64 in the UK.

Cons:

• The generated gate voltage varies a lot with ambient temperature, from 5V at 100C up to almost 11V at -40C. So if used in circuitry that will be outside, this gate voltage range should be factored in when choosing a mosfet. The LED forward current also affects the gate voltage but to a lesser degree.
• Slower turn on and off than LTC7000 series. The data sheet states 53us turn on and 65us turn off times. However, this is only with a 200pF gate capacitance whereas most power mosfets with low Rds(on) can be up to 10nF, so the times would likely proportionally increase with the gate capacitance. Also, those times assume an LED forward current of 20mA, whereas at only 5mA the stated turn on time increases to around 160us, but the turn off time decreases to around 40us. So for a 10nF gate capacitance with 5mA forward current, I would expect about 8ms turn on and 2ms turn off times.
• Only available in surface mount format, so requires a PCB to be designed and manufactured.

## Latching Relay

Similar to a normal electromechanical relay, except it remains in the on or off position after the coil is turned off. In this case I will be discussing only normal PCB mount latching relays, not large contactors or other classes of relay.

Pros:

• Draws 0 current in the on or off position. The coil power is 400mW or less on most models I could find, so with a minimum pulse width of 20ms, only 8mJ is used to turn on or off. Averaged over 10 seconds this is already only 800uW which is far less than the other options, and if the periods between switching are even longer (e.g. on the order of minutes or more), the power used is essentially 0.
• Is through hole format, so can be easily used with a stripboard (or a breadboard if prototyping).

Cons:

• The contact resistance for the majority of the latching relays I could find was stated to be 100mΩ max, which is much greater than that of an NMOS. Using this figure, at 1A current you would waste 100mW of power in the relay. This alone is much higher than the combined on power and mosfet power loss of the LTC7000 series or VOM1271 solutions, which would be 53mW and 48.5mW respectively assuming a 32mΩ NMOS. Only one series of latching relays I saw had a contact resistance less than 100mΩ, which was the G6C series from Omron with a resistance of 30mΩ, which is on par with NMOS. This would remove this con, but you would be forced to stick with this series (or other similar ones which I haven’t seen) which reduces your component choice.
• For DC usage, the rated voltage is only 30V for most models I could find, at 30V or less, the rated current limit applies, however above 30V this current limit gets heavily derated. At 84V, the current limit of an 8A SPDT relay was only 0.6A for example for a resistive load. This is due to the higher voltage making arcing more likely, so current must be derated in order to guarantee the relay can break the arc. Conversely, the AC rated voltage is 250V before the 8A current starts to get derated. The reason much high AC voltage is allowed is because AC voltages cross 0V twice each cycle, which means the arc is much more likely to extinguish during these crossing periods. Conversely, DC voltage remains constant and so an arc can only be extinguished by air resistance alone. This makes it difficult to find an appropriate PCB relay that can handle 84V and 1A or more, as you would need to find a relay with a much higher current rating than 1A in order to pre-correct for the severe de-rating.
• For dual coil models, 2 microcontroller outputs (as well as 2 transistors) are needed, one for set (turn on) and one for reset (turn off), whereas the other methods only use a single output pin to turn on and off. It is possible to use single coil latching models with only one MCU output, but this would require additional circuitry to translate the MCU’s high and low voltages into the appropriate positive or negative voltages on the coil.
• The electrical endurance can be anywhere from 8,000 to 100,000 operations minimum for the relay models I’ve seen, which could be limiting depending on how many switching operations per second are performed on average in the application. In contrast, mosfets have a much longer endurance/life.
• They require a minimum current in order to not fail prematurely due to contact oxidation (a current above the minimum will guarantee burning away the oxidation). 10mA minimum seems to be a typical figure, which is right on the edge of my application since I have one 84V load (the CA V3.1 block in my diagram) that draws only 20mA maximum.
• Somewhat costly at £5 each for a dual coil Omron SPDT latching relay for example.
• Slowest to turn on and off at around 15ms max including the bounce time.

## P-Channel MOSFET with Zener Diode Power Supply

The basic version uses a 10V Zener diode to create the correct Vgs, as well as two resistors and a transistor to pull up or down the gate.

Pros:

• Cheaper than the LTC7000 series or the latching relay. The cost of the 4 driver components should be less than £1, so then that just leaves the cost of the PMOS. A low Rds(on) PMOS of 60mΩ is around £2.33 in the UK. So the whole circuit should be about £3 or less. Compared to the VOM1271 + NMOS combo, this is probably slightly more expensive though.

Cons:

• Restricted to only using PMOS instead of NMOS, which means you will have a higher Rds(on). Just looking at through hole PMOS that can withstand 90V or more, the lowest Rds(on) I could find was 60mΩ. However this will come at a cost premium as the vast majority of PMOS at Vds 90V+ have Rds(on) of over 100mΩ, in which case you would be better off using the other solutions as they would have lower total wasted power (on power + load power loss).
• For the basic circuit, quiescent power draw has a trade-off with switching time, as a lower value resistor will increase current to/from the PMOS’s gate but also increase quiescent current. There are some more advanced circuits with additional components that can reduce the leakage current but that would require more time to design and verify.

## Conclusion

Of all the options, the VOM1271 seems like the best for my application, as it draws the least leakage power other than the latching relay and has the lowest load power loss due to the low Rds(on) NMOS. Although it does have relatively slow switching times (up to 8ms for turn on time), this is fine for my application as I will be using it as a simple switch rather than with any high frequency signals such as PWM. This is because although higher than normal power is dissipated in a mosfet during the turn-on/rise period, according to this article from Analog Devices, so long as the turn on/off time is limited to 10ms or less, the high power generated will not have long enough to heat up the mosfet by any significant amount.

LTC7000 series is higher performance but as I am only using this like a switch and not doing PWM, I don’t need fast switching times, so the extra cost and leakage power of the LTC7000 series isn’t worth it.

PMOS has higher load power loss due to higher Rds(on) and potentially higher leakage power.

The latching relay was probably second place, because whilst it could potentially have the lowest power out of all the solutions due to 0 leakage power, this is assuming you can find the rare relay that has a contact resistance in the region of 30mΩ or less such as the Omron G6C series. In addition, the Achilles heel of relays appears to be high DC voltages, because this promotes arcing which can not be easily extinguished, so it seems they are much more suited to AC or low voltage DC. Because of this problem, it would be more difficult to filter for a relay that can pass 1A or more at 84V due to the severe derating. They are also more expensive than the VOM1271 solution.

The VOM1271 is surface mount which I was trying to avoid initially due to the extra time and cost of designing and manufacturing a PCB. However, it only requires one other component (a resistor), so designing a PCB for it should be simple even for people with a low amount of PCB design experience such as myself.

Finally, it’s worth mentioning that I probably won’t be using these solutions for low voltage high side switching, as even a leakage power of 16.5mW of the VOM1271 per mosfet is much more than the rest of my system’s standby power at 500uW, so I will be looking for a more efficient gate driver for low voltage purposes.

• Take a closer look at the 8ms turn on time that may be a real problem. If your load draws current while your fet is in the linear region during that 8ms it creates heat and may violate the SOA (safe operating area). If your load is capacitive you have problems independent of startup times. Commented Oct 6, 2023 at 21:33
• @asdf30 I had a look at the SOA for the IRLU3110ZPBF (one of the NMOS's I was considering using) but whilst I know the max current will be 3A for one of my loads, I wasn't sure how to determine the average Vds during turn on for the SOA graph without simulation, since during the turn on period, Vds will be changing from 84V at the start when the mosfet is fully off to less than 0.1V when it is fully on? Commented Oct 7, 2023 at 9:02
• For steady state that mosfet can definitely handle 3A without a heatsink at even 3.3V Vgs, but I was aware that for all mosfets during turn on/off there is a period where the product of Vds and Id is much higher than during steady state so you get a transient spike in power. So I wanted to find out what is the max acceptable turn on time without damaging the mosfet, which is where I came upon that analog devices article that I linked that talks about at 10ms or below, the extra power during turn on does not have enough time to heat the mosfet significantly to damage it, so 8ms seemed ok? Commented Oct 7, 2023 at 9:04
• What’s your load? If it’s an active load with a delayed startup you may be ok (but consider C). If it’s an R load I’d love to see an SOA that can tolerate full V and I for the 8ms. This mosfet has very poor linear behavior. Note SOA is a pain to search for but keep looking and you’ll find some parts are much better than others. Commented Oct 7, 2023 at 23:31
• Perhaps. To put it simply your fet goes from 84V/0A to 0V/3A over some time. If it’s a resistive load the lines will be straight and in the middle it will be 42V/1.5A. The time it spends in the linear region depends on the turn on time - in your case about 8ms. Google “hot swap safe operating area” Commented Oct 8, 2023 at 15:21