fast CC/CV power supply... again

[exe]

Hello my friends!

This is my another attempt to make a fast and precise power supply. Previous two attempts are here:
1. https://www.eevblog.com/forum/projects/fast-cccv-power-supply-take-two/msg4988506/#msg4988506
2. https://www.eevblog.com/forum/projects/how-to-design-fast-bench-supply-with-cc-and-cv/

In my third attempt (not really true, I've built 20+ prototypes) I'll try to combine the two (kinda). Harrison/HP topology is good for both CV and CC modes of operation because in CV mode it's a voltage follower (i.e., low output impedance), and in CC mode it works as collector current source (high output impedance, good for CC). The problems are:
Problem 1: switching between CC and CV modes
Problem 2: shunt and output capacitor lower bandwith in CV mode, but increase bandwidth in CC mode. You can't have high bandwidth in both CC and CV modes. So, pick your poison.

## The idea

Since I want fast CC mode to protect DUTs, I can insert imprecise but fast current limiter. It doesn't have to be accurate, all it needs to do is to limit current until CC-circuitry engages. Here is what I'm trying with a simple bjt current limiter (see screenshot). I quickly checked it on breadboard, it works fine.

I tried to use jfet to limit current, but it's hard to find a jfet with any high Idss. I also tried depletion mode mosfets and normal mosfets for current limiting, but they need quite some voltage drop to work reliably due to low transconductance comparing to bjt. So I ended up with bjt. Also, bjt inherently limiting current due to finite beta and limited base current.

## Practical tests

I used a red LED that I connected to 32V supply with current limit of a few mA, but with ~2.5uF output capacitance (two 16V PSUs in series, each with 4.7u MLCC output capacitor) . Without the limiter the LED dies pretty quickly, just after two or three connects to power supply. With the current limiting circuit I wasn't able to kill it. Success!

Now not so great thing about this current limiter. It seems I can just use 24 ohm resistor to limit the peak current and LED still survives. So, my LEDs survive peak currents of more than 1A without problem. Also can pass 50mA no problem, at least for a short period of time.

So I need to make a better test.

## Downsides

1. Protection circuit has some voltage drop across it. So, not ideal, probably works well only for small currents. I don't think this is the problem as I only need protection for low currents anyway. I plan to make circuit bypassable when higher current range is selected.

## Further works

Okay, red LED is good, but I need a more scientific way of measuring stuff. So, I plan to measure how quickly voltage drops across a resistor when current limiter engages. I can capture that with o'scope and even calculate peak current and energy dumped. With this I can benchmark power supplies and different protection circuits.

## Why previous attempts fail.

The first design had Harrison topology ([1]) heavily inspired by @blackdog 's PSU ([2]). While it worked, I had troubles making it fast enough under all operating conditions, particularly at low currents (more on that later). Probably that was due to my lack of skills. So I had to try something else.

The second attempt was a radical way to simplify things: I just made fast current source (easy), and added a voltage clamp (not so easy, but was working). That's how design 2 was born. I've built the circuit and it worked really well from first attempt. There was only one problem: it couldn't measure current accurately. That's because some current goes through the load, and some through the BJT shunt. That base current error is quite significant. So I tried to fix that with adding a transimpedance amplifier to measure load current, but with that the topology is no longer simple and fast.


References:
[1] http://hparchive.com/Journals/HPJ-1962-07.pdf
[2] https://www.eevblog.com/forum/beginners/how-does-blackdog_s-psu-work/

[exe]

What is safe current to pass through sensitive components? Here are much thoughts.

Most ICs specify they can pass up to 10mA through their pins continuously (with some rare exceptions, like LMC662 can only pass 5mA). Also, most ICs can sustain ESD discharge of 2kV with Human Body Model (HBM). Since HBM is a 200p capacitor with 1.5k series resistor, we can calculate peak current of 1.33A.

With some hand-waving, I assume that as long as we keep peak discharge current below an amp or two, and quickly limit to some tens of milliamps in less than a few microseconds, most sensitive parts like small bjts, opamps, microcontrollers etc should be fine.

So, limiting instantaneous current to below 1A, and more prolonged current to 30-60 mA (until CC kicks in) looks like a safe choice.

[Kleinstein]

A lab supply is not made to have a super fast current limit. Even if the electronic part is fast the output capacitor could kill a sensitive device when starting from a higher voltage.
Another point is that lab cables have inductance and even of the voltage is stable at the supply the voltage at the load may still drop from the cable inductance. So a load that need a stable voltage even with current spikes kind of wants a local buffer capacitor. Not many needs for really fast regulation. The more important point is making shure that even tricky loads don't cause oscillation.


A point why one may want fast regulation is that the faster the regulator, the smaller the output capacitor can be. Commercial lab supplies may have 100 or even 1000 µF of capacitance at the output. At least in the simulation one can use rel. fast regulation and get a design with maybe some 1 µF or so and still not have much overshoot for the CC to CV transition.

A point for a fast CC-CV and CV-CC transition is to look at methods to limit integrator wind-up. This can be somewhat tricky if high precision is needed, as diodes have leakage.
A radical way is to have only 1 integrator for the CC and CV mode and this the diode "AND" function before the regulator / integrator part. Many SMUs seem to work this way, doing the min/max with the raw error signal and the integrator part of the regulator only after this.

[Terry Bites]

An addon current source will limit the max current in your load irrspective of the PSU behaviour.
Thats a trivial circuit. eg LM317. That'll respond to a transient in 10uS or so for a resistive load.
See data sheet.

[xavier60]

I did some experimenting with an idea that seems to work well in the other thread, https://www.eevblog.com/forum/projects/fast-cccv-power-supply-take-two/msg5093919/#msg5093919

[exe]

I created a small fixture to do 10A (or more) pulses through LEDs to see how well different current limiting methods work. Haven't used it much. However, from what I tested, even cheap LEDs that I considered very weak could survive several amps of current if pulses are not repeated. With repeated 10A pulses they degrade pretty fast (can't blame them for that). The only issue is how to measure that degradation. My plan is to compare brighteness before and after abuse. Since I don't have any equipment for that, I'll just compare brightess with a new LED. The plan is to put them in series and compare brightness visually.

I'm curious what do you think about eye safety. Can 10A+ pulses through a small LED damage my eyes? I prepared a shield, but I wonder if I really need it. The LEDs when pulsed for 5us (0.1% duty cycle) were barely lit, but doesn't mean they are safe. Or, may be, the first pulse destroyed the LED, and that weak glow doesn't emit much optical power, idk...

@xavier60: thank you very much for the circuit. Back when you posted it I didn't get how it works exactly, but now I'll spend my time to properly go through it. Thanks! PS do you happen to have the *.asc file for the schematic? That would save a bit of time redrawing the circuit in LTSpice.

[exe]

So I did a few tests.

I tested lm317l as a current limiting device. The results are... interesting. There is like 50us startup time when subjected to a pulse. So, perhaps, it will work, I just don't like the startup time before it starts conducting, and also significant drop across shunt (1.24V).

Next I tried the schematic above. Actually, even simpler: I drove base from a fixed voltage with a series resistor (see screenshot from ltspice).

The results are promising, the peak current is about 0.35A, and only lasts 0.5us or so. Then it settles at ~45mA. I connected a diode, and banged it with pulses for about an hour (at 10 pulses/s rate, with 0.1% duty), and I don't see any difference comparing to how bright it was before. I took picture with my phone with fixed settings (1/100s shutter, 100iso) so I can compare its brightness to other diodes.

Now things that didn't go well. One thing that bothers me is that with current limiting, there is some current flowing for almost 2us. It's as if switching mosfet has "stored charge" or something. But that's not possible, right? Anyway, even if I limit current with a resistor, I clearly see that the falling edge of current is not sharp at all. I checked gate voltage, it's fine. I also noticed that the higher current flows through the mosfet, the less time it takes for current to drop to zero. UPDATE: I figured it's because of drain-source capacitance. If I put a 10n cap between source and drain, I have the effect greatly amplified. Hmm...

I'm also not happy how fast I drive the mosfet. Perhaps, it's a bit oversized for the job (LSGC04R029), so esp32 is struggling to drive it a little bit. I think large mosfet contributes to switching noise that is very apparent on screenshots.

Some fun stuff. I managed to kill a 2R resistor while discharing 3300uF cap charged to 35V. It measures open. I also killed a mosfet (AP2312) rated to 20V when I tried to make it block 35V. It worked well for a few minutes, and then stopped.


PS datasheets:
- AP2312 https://www.lcsc.com/datasheet/lcsc_datasheet_1912111437_ALLPOWER-ShenZhen-Quan-Li-Semiconductor-AP2312_C360339.pdf
- LSGC04R029 https://wmsc.lcsc.com/wmsc/upload/file/pdf/v2/lcsc/2009091607_LONTEN-LSGC04R029_C779910.pdf

[T3sl4co1l]

What do you mean MOSFET, same circuit as for the BJT or--?

Yes, 42nC typ. Qg(tot) at 10V or equivalent ~4nF average over the full switching swing, will take a lot of current and/or time to transition.  Consider this figure with respect to the current you're running at (drive or load).

Tim

[exe]

What do you mean MOSFET, same circuit as for the BJT or--?

Apologies for confusion, here is the complete circuit overview (minus decoupling capacitors). So, there is a current limiting bjt, and a switching mosfet.

[ArdWar]

It's as if switching mosfet has "stored charge" or something. But that's not possible, right?

Well, it is, in the form of gate charge which especially shows up when your gate drive is weak. Are you sure your gate are actually at zero during the "taper off" period?

Your "switching noise" are on the other hand probably came from your measurement setup.

[exe]

Are you sure your gate are actually at zero during the "taper off" period?

Yes, I measured it. There is a 2us delay between the time the gate is off and mosfet stops conducting.

[exe]

I compared three different bjts for current limiting. To my surpise, the winner is 2n3904 (can't identify company, probably onsemi, beta 300). It has the smallest overshoot. Next bc849c bc549c (fairchild, beta 600), then d882 (ST, beta 200). For those who play along at home, vertical scale is 200mA/division. That's because it measures voltage across 0R25 resistor.

I expected bc849c to win because I thought it has smaller die due to lover current rating. However, in the datasheets, it has nearly identical capacitances with 2n3904.

Peak current is 240mA, recovery takes only 130ns, I think it's a success).

NB I changed scope parameters such as bandwidth (now its 20MHz lowpass, 10X probe), switched back to AP2312 mosfet, and limited voltage to 20V. Because of that, results are not directly comparable with previous screenshots.