Driving currents into protection diodes on CMOS

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

bcarso

Well-known member
Joined
Feb 20, 2005
Messages
4,055
Location
San Fernando Valley, Los Angeles
Has anyone had experience with equipment where occasionally current flows into forward-biased protection diodes on pins of CMOS parts? Specifically, I have a horribly complex EQ with tons of 4016 quad bilateral switches in design/breadboard/layout. In order to preserve dynamic range I am running the 4016 parts off of +/- 7.5V, and the opamps from +/- 15V. The switches are positioned at the inverting inputs of the opamps to minimize the voltage swing across them and hence minimize distortion.

I'm using 4016 parts because they are cheap and because they don't have the nasty behavior of 4066 parts when switching, which would require a preemptive muting function every time some setting was changed.

Under high level drive there are times when the unselected switches see more than |7.5| volts. But it is always with a resistor in series to limit current, typically less than a milliampere.

Right now I have belt-and-suspenders clamping with SOT23 dual diodes at each affected node, clamping to lines positioned 0.7 volts below Vdd and 0.7 volts above Vss. For one channel this entails 76 duals!

When I test a candidate 4016 with my Fluke DMM in the diode mode, which pumps 1.0mA into the DUT, I see voltage drops from given pins of the 4016 to each supply pin in the neighborhood of 0.65-0.73 volts. So these clamp diodes are not large, nor would one want them to be for other reasons. The question is: would there be any long-term degradation effects from repetitive forward conduction?

Of course the power supply for the CMOS needs to do a good job regardless of the polarity of its output current, since the nominal small leakage current under static operation will be tiny compared to the reverse-direction clamp diode conduction currents.

I can test this in short term, and even determine what current reliably damages a part, but it would be nice if someone has had experience already. Datasheets are not very helpful, as in most cases they never specify a current, and assume a voltage source and tell you to stay away from anything more than 300mV above Vdd/below Vss.

Also, with some CMOS, there are parasitic bipolars that can be triggered into conduction with a suitable transient and work as silicon-controlled rectifiers, and suddenly put a clamp with about a 1.2V drop across the power supply rails.

Seems to me Pease said something about overload of CMOS gate inputs once, when showing some pulse generation circuits where a capacitor slammed the input during switching. But I think he was just wary, said you shouldn't do this, and didn't give any limits.
 
I've done some switching with those parts but don't have much experience with exploring the limits of the clamps. I did some very low distortion tricks in a console back in early '80s, but managed the open voltage issue buy clamping unused inputs to grnd (to reduce crosstalk).

I wonder if there might be another way to skin that cat? You want a low impedance path for the transfer gate ports themselves, but the power supplies are mostly providing bias voltages. Especially if the TGs are switching a virtual ground input bus (like almost no voltage in use). Perhaps provide a compliance in series with the PS? Something like 100K resistors in series with +V and gnd (-V?), then a 16-18V zener directly across the CMOS chip could keep fault current low and parts happy. There may be some other clever angles applying mitigation to PS side of circuit, but without seeing details not worth more speculation.

JR
 
One bit of 'magic' can occur - some input signal may show up in the output when it's off. One way to do this is to have complementary 4016's - one to feed the signal to the op-amp, the other grounds the signal when the one feeding the amp is open - so no voltage appears there in either case.
 
I considered having two switches per switched resistor as well, rather than clamp diodes, but it would double the number of switches. Already there are 48 4016 packages in this beast. OTOH the number of dual opamps is only 26. This is for doing five sets of preset EQ each with four bands, for three channels. There is one resistor at any given node connected all the time, and then four others that are switched in or out in parallel.

Right after I finished the basic topology for this I read Fred Forsell's piece about EQ, using resonators a la Steve Dove's allpass phase shifter network (actually, a very similar circuit in which an overall bandpass function is realized had existed well before Steve's famous series of articles, but who knew).

Fred speculates as to why anyone would have four filters in series when resonators hanging off of a single sum-difference stage would do, and reduce the number of stages through which the signal passes. Well, the constraints on this application are an example of why one would---Dove's circuit is hairy to use these sort of switches with.

The power supply for the switches already is made of shunt regulators, so an SCR crowbar event would not be destructive, probably. However, it would punch a hole in the audio and probably require power cycling to restore, so that's only a small comfort.

My executive consultant (a Radio Shack press-the-button and spin the answers until they stop and speak to you) tells me I could get away without clamp diodes. My critical-parent inner voice says don't take the chance.
 
what do you think is going to occur in the long term that you couldn't simulate on the bench? it sounds like the perfect opportunity to fry some chips, see how much they can take. if you want to hit them with years worth of transients, perhaps a sharp pulse HF waveform only a little below the one-shot failure point?
 
Another thought, how about just dropping the PS and signal swing for these filters to +/- 7V? Most GP opamps don't swing all the way to the rail anyhow so you're not even giving up 1 bit of dynamic range.

Big dog codecs only use one +5V rail, so maybe just keep an eye on the noise floor and drop your rails.

FWIW I made a midi controlled line level studio switch box back in '80s using +/- 8V rails, to keep everything inside the CMOS process limits.

JR
 
> it sounds like the perfect opportunity to fry some chips

Geez, do you know how much a 4016 costs??? What they pay bcarso, he can't afford it.

<G>

Don Lancaster's CMOS Cookbook shows several possible on-die input networks, using straight diodes, 20V Zener diodes, and few-hundred ohm resistors (which may enter into your hasty estimate of diode size). He goes on to say "limit... to 10mA".

That's what I recall. You can dump 10mA fault current recklessly. Maybe you have down-the-road electromigration failures, but I don't remember this ever being an issue.

I say "do it". Let the internal diodes work. Note that millions of CMOS multivibrators smack the diodes thousands of times a second for decades.

But note that I am not voluntering to handle the warranty repairs and expenses and ill-will.

There are other switch arrays. There is a very old (Gates?) analog switch rated for like +20V/-10V. They are mostly Military market and not as cheap per point as 4016. There are new matrix analog switch chips with multiple possibilities. Unless you need a very large array, enough to justify a nano-CPU to set-up the matrix, these will not be cheaper than 4016.
 
I'm with PRR, as long as you current limit it to a couple of mils, whack away.

At a nickel a switch it's the way to go.

I'm using the DG-408's in the EQ-2NV on high voltage rails because they're on the input and output side of the networks, but they also cost 10 times as much.
 
I know from experience that unless the multiplexor is injection-current rated, when you exceed the power rails with the input signal, the output signal will often receive a distorted version of the input signal. I got bit by that once when using a 74HC4051 in a prototype because I didn't have the right chip (74HC4851). I was reading an analogue signal of +12 volts (through a 51.1k resistor) in a +5 volt system. ALL of the analogue inputs on that mux read full-scale on the A-D converter, regardless of their input voltage (even if that input voltage measured 0 volts), simply because a single input on that IC was shorted to vehicle battery voltage - at a current of 130 microamperes!!!

For the subtle difference, compare the 74HC4051 and the 74HC4851 analogue multiplexors. Note that the 74HC4851 is injection current rated, and I use them all the time in my (automotive) designs. Note that the 74HC4851 has a specification called 'Injection Current Cross-Coupling'. Of course, you can't use this part in your application, but it gives you an indication of something to watch for.

Check the MC14016B data sheet from On Semi... it indicates that there are no clamp devices that you can rely on - it says so right on the first page - "No ESD Protection". The data sheet for the 4066 does not have this note.

Also, the text on page 8 of the data sheet explicitly tells you to clamp them with low voltage drop diodes (germanium). That piece of information indicates to me that you should not rely on these clamps. The 74HC4581 data sheet tells you explicitly that you don't need to.

I am thinking that your inner parenting voice might be correct in this case. Only because I didn't listen to that voice in my past on this issue (and many other issues, too) and got bit. Now, in this application, it may do something that's ok, but it might not. Check carefully.

-Dale
 
I am sitting on the fence regarding this. Sure I have used my share of cmos gate oscillators that banged those diodes zillions of times, albeit with tiny currents.

I went back to my CMOS bible (circa 1975 RCA data book), so old it calls it COSMOS but they go into detail about static protection schemes. Anyone familiar with old unprotected MOS remembers how fragile that stuff was. Per RCA current limiting inputs .5V beyond rails to 10 mA continuous is their advice for happy sailing. The TGs have clamp diodes from input/outputs to both supplies.

It is worth confirming that the gate doesn't do anything funny to the audio while it's protection diode is conducting. I find it difficult to imagine that anybody is making versions of these without protection in light of the fragility of bare ass MOS but when in doubt read the data sheet.

While I would just avoid the need for this exercise by lowering the rails, if inclined to rely upon these internal diodes I would at least search out several current data sheets from likely component vendors and if they all look OK give the contract manufacturer a list of approved parts so at least you have CYA on that level.

Of course that doesn't guarantee that somebody, sometime won't substitute a cheaper part but the savings from leaving out the protection devices seems a false economy for an IC maker, so I don't expect there to be many such options available to a wayward CM.

JR
 
[quote author="JohnRoberts"]I am sitting on the fence regarding this.

.....

Of course that doesn't guarantee that somebody, sometime won't substitute a cheaper part but the savings from leaving out the protection devices seems a false economy for an IC maker, so I don't expect there to be many such options available to a wayward CM.

JR[/quote]

Thanks to all for their (unclamped) inputs on this :grin:

The devil is mentioned by John above: there is no guarantee with this third-party manufacturer who will require constant policing on most everything as it is. My client is making a maiden voyage here, in stormy seas to begin with, and I need to protect them as best I can.

FWIW I did try diligently to blow up some Harris parts to no avail, dumping as much as 40mA of either polarity into the pins that would see at most a mA. Nothing---no sign of SCR action, no damage that could be discerned.

The sane thing to do is lower the opamp rails, but I really do need the roughly 6 dB I think.
 
[quote author="dale116dot7"]
Check the MC14016B data sheet from On Semi... it indicates that there are no clamp devices that you can rely on - it says so right on the first page - "No ESD Protection". The data sheet for the 4066 does not have this note.

Also, the text on page 8 of the data sheet explicitly tells you to clamp them with low voltage drop diodes (germanium). That piece of information indicates to me that you should not rely on these clamps. The 74HC4581 data sheet tells you explicitly that you don't need to.

-Dale[/quote]

And that On Semi departure from the standard parts is VERY significant and exactly the sort of thing to fear.

I'm reminded of the Monty Python insurance sketch, when Mr Devious explains that the insured man had purchased one of their no-claims policies "which is a very good deal as long as you don't have any claims". The ON parts are probably very good for low leakage and capacitance specifications compared to parts with internal clamp diodes.

As mentioned above, in this design the clamping is done with silicon, but to a voltage (a bypassed zener) one Vbe below Vdd, one Vbe above Vss.

Thanks again for that bit of data Dale.
 
I stand by my preference to just drop the rails a few volts, loss will be less than 6dB due to output stage losses and low noise opamps could buy back that loss unless you're already using low noise parts.

Another though in passing. Are you married to the parallel structure? If the resistors were configured as a series string, they would form a natural voltage divider that would provide some degree of signal swing reduction.

JR
 
Using TL072's at the moment, mainly because they are cheap but also decent performance, especially in the inverting mode where there is essentially no input C modulation. Also there are no concerns about current noise, and they are much less susceptible to RF rectification effects than most bipolars. A bit lower e sub n would be nice though---the main noise contribution now is from summing amps, with the integrators negligible by comparison. Resistor thermal noise is significant but below e sub n times noise gains in the summers.

With the fairly light loading I see the 072s swing to within nearly +/-14V with +/- 15V rails. So were I to run the opamps and switches off of +/- 7.5V and assuming the same approach to the rails, I would lose about 6.7dB. I could run the opamps a volt more per rail and recover some of this.

I could also live a little dangerously and push the opamps to +/- 18V rails, or nearly, and pick up a little bit.

At the end of the day it may turn out that the source component shipped with this system has more noise than the amp, but only because it may be poorly designed ( yet another third-party vendor). My hope is that at least the amp is mostly "blameless".
 
Hey Hey my first post in a long time! I'm glad that it's a reply to a good thread!

Although I'm not familiar with this specific IC, I can tell you a bit about my experience with CMOS...

DO NOT TRUST THE BODY PROTECTION DIODES!

:green:

Well they never seem to be "enough" when the going gets tough.

My opinions now out of the way, do we know anything about the protection diodes in this IC at all? From blowing up my share of CMOS ICs in the past, during which I was trusting the datasheets a little too much, I found that the protection diodes are generally very very very close to the edge of adequacy. I've also found that OV situations tend to lack in catastrophic failures but cause high amounts of current leakage and commonly form a resistive device. Especially in higher power level parts. i found that I always needed an external diode when switching large FETs and IGBTs regardless of what the literature says about the internal diodes. But these diodes are almost always reversed diodes, not forward diodes that you mention

Recently we were working with an RF CMOS switch that supposedly had a degree of protection built in. The switches would randomly die strange deaths but always with the voltage rails partially shorted. Initially it was thought that the parts may have been faulty because there is also external protection diodes. Testing showed that both the external and internal diodes were blown but still showed that the diode was intact. Somehow the substrate had been compromised and some kind of current path was created. I later found that the external parts had been spec'd incorrectly and their voltage rating was signifigantly lower than what was expected to be used and the RF switch had no tolerance for the voltage rails to float for short periods of time. Seems that the internal structure, while "protecting" the I/O pins did not protect against one of the voltage rails floating while the other was fully on and a signal present like during power-up. Changing the external protection diodes seemed to fix the main problem that we were seeing. Who knows, but I still have a feeling that the parts don't like having a floating rail during turn-on while seeing very large amplitude signal peaks, much like the parasitic capacitance you mentioned.

As for crowbars, I've had some fun with those. Generally when you want to use a crowbar, you already see the need to recycle power. these are great for powersupplies when a fault might be downstream(or upstream) and you don't want the unit oscillating while applying it's power then current limiting and shutting off over and over. This also keeps from hitting a shorted part with power once it's blown ( FIRE!) I used one on a robot for overvoltage protection, a picofuse, a zener shunt and a SCR. the overvoltage would kick the SCR ON and short the circuit to ground. the picofuse would open and stay opened until power was recycled. It worked really well in instances where the cable to the robot would get cut, shorting higher voltages in the cable to the lesser voltaged circuits.

But I digress.

I'm also of the mind that maybe a switch IC should get sacrificed in the name of science on this one because there are so many variables when taming the CMOS beast. But i'll offer some more info to the contrary here in a second.

would there be any long-term degradation effects from repetitive forward conduction?

I would say no as long as the current/power rating hasn't been reached and you have a limiting resistor in series.

Since the protection diode fiasco of late, we've been banging our circuit on a hodgepodged test rack that slams 100 vac waveforms into the circuit to see how it reacts since the circuit is designed to clip at 90 vac. After letting this thing run for days on end, I didn't see any of the failures that were previously seen and all testing showed the diodes were still within datasheet specs including capacitance and resistance between nodes. While this is hardly long term, I think your scenario is much less harsh than this which should offer some hope since if I can't kill a diode via reverse conduction, you shouldn't have much to worry about in your application.

Drops of .6v would put you around schottky range, exactly what I would put as external protection on any of my low level signal designs. But hey, I've been wrong before!

I guess in closing my excited rant which has spanned about 2 hours in the making due to this pesky thing they call "work" I would sum it up as such:

If you can afford to use external diodes, do it. Personally I would rather plan to use them but run as many tests as I could to see if I could get away without them. I would also turn in the prototype with them installed so that the person with the checkbook doesn't get a bad suprise later if you find that you actually need them. i would also leave the pads on the PCB if you find that you can do without.. you never know what will happen and you might need them as well. If for nothing else and at the end you can reliably say that you don't need them, The Holder of the Checkbook will be mightly pleased that you cut a signifigant cost from the BOM..
 
Thanks for all of that information Chris.

I think I will go with diodes, although it will make the board(s) bigger, and if they feel really brave consider depopulating down the road. This is part of a system that will retail somehwere in the 5-7k range and it would be a shame to have vulnerabilities due to saving at most a couple of bucks.

Again, one of the major issues is lack of complete oversight of component sourcing/quality, so one IC may work and another not.
 
Wouldn't any overvoltage current be limited by the filter resistor values? I am speculating that this would keep fault currents pretty low.

If the switches are in series with a virtual ground the diode clamps could actually be clamping to ground rather than some synthesized sub rail voltage as the voltage at the node would be near zero when switched on and who cares when switched off. Of course this would in theory introduce more distortion via TG capacitance, but that should be pretty low (<1 pF?).

Another way to protect is a resistor to ground at the clamp point dividing the open switch voltage, but this too has a noise gain consequence similar to just dropping the signal level and boosting back again post switches.

Good luck.

JR
 
Again, one of the major issues is lack of complete oversight of component sourcing/quality, so one IC may work and another not.

Oh absolutely, although the IC type is not related, I have found certain manufacturer's IC's to be stout and hardy while another manufacturer's "identical" part be failure prone, mostly Vregs and PWM ICs but I would extend this across the board and err on the side of caution in a product that would retail at 5k+.

Wouldn't any overvoltage current be limited by the filter resistor values? I am speculating that this would keep fault currents pretty low

Sure they would, but in a commercial product I would not rely on only that to protect an IC from catastrophy. I've been in the heat of a product recall due to design problems resulting in unit failure and the cost of designing in a redundant safety mechanism is far less than an extensive repair/recall program, especially if you consider in your reputation and the reputation of the company. Sometimes you can't put a price on that. If it were a personal project, I wouldn't think twice about leaving the diodes out, but that's not the case here.
 
Believe it or not, I am familiar with the constraints of large scale manufacturing. That said it is worthwhile to fully understand the failure mechanism you are protecting against otherwise you are throwing money on the ground or worse, adding circuit complexity that could itself become a failure vector (leaky, shorted, or mis-inserted diodes anyone?). If you do reliability engineering you find diminishing returns for protection circuit complexity beyond first order or anticipated high likelihood failure modes.

I have routinely dumped modest currents into CMOS catch diodes in simple oscillator circuits, and don't recall one related failure, ever. The data book claims a 10 mA tolerable limit and manufacturer's don't routinely promise that unless they can do better. At issue is what do all the modern parts, that may find their way into those boards, act like under anticipated stressors.

My personal inclination is to gain structure the circuit so clamps aren't needed, internal or external, but YMMV.

JR
 
Believe it or not, I am familiar with the constraints of large scale manufacturing.

I apologize if my post sounded preachy or anything like that, my intent was purely posting my opinion based on my own experience and nothing otherwise.

:thumb:

My personal inclination is to gain structure the circuit so clamps aren't needed, internal or external, but YMMV

In my circumstance and used in the subsequent example, the stage before the problematic diode/rf switch is the outside world. This circuit needs to be as hardy as it can possibly be within reason as this is a testing product that connects to digital coax networks with various DC and AC signals and voltages. This could very well include lightning, spikes, surges and so forth. I guess the basic idea is the same although in my instance the circuit needs to deal with faults that are usually a couple orders of magnitude higher than what Brad's circuit would ever see.
 

Latest posts

Back
Top