Absolute maximum ratings

[Viitalahde]

Say, an op-amp has specified maximum ratings for rail voltages. For an OPA604 they're +/-25v (out from the memory). Recommended rails are +/-24v.

How much safety margin you think they've specified?

 

[clintrubber]

I see where you're going, you want to run stuff at higher rails and hope for the 'real' limit being still above that, right ?

I guess if you're not going really a lot over those ratings it'll still be fine... for most samples... and for a while... :wink:

In other words: while it looks like it most of the time, ICs aren't designed for eternity, not economical. Stressing them harder reduces lifetime.
It could be an approach though, enjoying eventual improvements in sound and just replace when things break. Sounds like tubes ?

Regards,

Peter

 

[mikep]

I dont think there is much margin at all when you look at it like peter notes, over time and lots of samples. BUT if you were to test lots of parts im sure some percentage will handle a wee bit more. I have a hunch that the ones that are gonna blow at say, 27V, are going to do it right away. and you won't have to worry about finding a use for the parts that end up in the lower bins, cos they'll be smoked, just toss them!

mike p

 

[CJ]

it depends on how much power you draw?

40 volts might be cool if your output current is microamps,
10 volt rails and 1 amp current will generate 10 watts, see where i am going with this?

fun to do destructive testing, usually when i wonder about boundry conditions, i just discover them for myself.

lucky people get paid to blow stuff up see here:

 

[bcarso]

There are a number of failure mechanisms. Some of them cause slow degradation and parts eventually getting out-of-spec. Some are more catastrophic, like punctured gate oxide or secondary breakdown and localized thermal runaway.

If you are not using something in a life-critical (or job and reputation-critical) app you can afford to take chances. But for some specs the semi vendors can't afford the big "six-sigma" windows the QA folks would like, or the performance comparisons beloved by Marketing would put the company at a disadvantage. The abs max voltage ratings and power dissipation ratings are usually in that category.

 

[JohnRoberts]

There are general process related breakdown voltages and perhaps pin to pin variations due to local internal circuit configurations.

IMO there is not a lot to gain from hot rodding parts to their upper limits. If you look at S/N or dynamic range in dB terms pushing the rails will buy you little in dB terms, but the thermal difference could be significant, since reliability drops precipitously for even modest temp increases.

I knew a design engineer who used to game the "vendor says 24 volts but part really does 26". He eventually got bit by some parts labeled 24 that only did 24. I personally prefer to run 24V parts at 22V. YMMV

JR

 

[pstamler]

Me too. I remember getting curious about this one time and trying some opamps rated at +/- 18V at higher voltages. At 18.1V they blew.

Peace,
Paul

 

[Viitalahde]

Yep, I know upping the rails a volt or two doesn't really add up much in headroom. +/-15v vs. +/-24v, about 4dB of difference.

Mostly I'm thinking of light loads, above 10k. With sufficient cooling, who knows? My EQ has the 604's running at +/-24v, and has worked fine for two years now. No cooling.

I think MikeP might be on the track that some parts will pop right away while some guys are tougher..

 

[gswan]

Sometimes all it can take is a little glitch on your PSU (caused by a large glitch on your mains) and you'll get an extra volt or so on your PSU rails for a very short time. If your HF decoupling is poor then you'll bugger all the devices run at their maximum supply rails.

In short, don't design things to run at absolute maximum. That's why those figures are published - to show you how much design headroom you have when other things are taken into account.

 

[keefaz]

Is is really better to power OPA604's with +/-24V ?
If I read the datasheet, at the performance curves section (especially the PSR/CMR vs supply voltage and gain-bandwidth/slew rate vs supply voltage), I can see that theorically the performance would be better if they run at +/-15V...
I haven't done any real test yet, but I found it interesting

 

[Guest]

Long time ago I used a curve tracing measurement device to select transistors with breakdown collector-base voltages more than 140 and with beta more than 100 from devices specified for 30V breakdown and 25-75 beta. PNP transistors had very wide spread of parameters than NPN transistors (recently I checked new Toshiba devices and found that it is still the same, NPN transistors are much closer to each other). However, I did not torture opamps though, but I expect you may find some damaging the rest...

 

[JohnRoberts]

[quote author="Wavebourn"]Long time ago I used a curve tracing measurement device to select transistors with breakdown collector-base voltages more than 140 and with beta more than 100 from devices specified for 30V breakdown and 25-75 beta. PNP transistors had very wide spread of parameters than NPN transistors (recently I checked new Toshiba devices and found that it is still the same, NPN transistors are much closer to each other). However, I did not torture opamps though, but I expect you may find some damaging the rest...[/quote]

Hfe being 33% high is not very surprising. I wouldn't expect breakdown voltages to vary that much from process targets, more likely explanation is manufacturer substituting a newer technology to sell parts ordered under archaic part numbers. I suspect the vast majority of 2n3055s sold today bear little resemblance to the original part, nor would I want them to, but this can be problematic for non-engineering types just copying an old design, or when someone wants to mimic the old (not so great) performance of some vintage circuit. IMO parts should agree with their specs. If much better, note that in the specification, or perhaps modify the part number. I recall seeing power FETs labeled 3055. Surely just a naked attempt to capitalize on the familiarity with the old workhorse part number.

Back in the good old days, transistor manuals contained process descriptions for families of parts, and individual part numbers were often just selected out for breakdown voltage or beta, or noise, or whatever.

Back at my old day job we had plenty of power transistors in the system using house numbers were they were just like a certain stock part but with one or more specs tweaked. Nowadays you could probably open up power amps from several companies and find the same (standard) plastic devices in them.

While you may find transistors labeled 35v handling 150v for the convenience of the manufacturer, don't expect parts labeled 150v to do 300v or opamps to exceed rated voltages plus more than a modest safety margin.

JR

 

[NewYorkDave]

Peavey? They used to supply their service centers with a nice semi cross-reference guide that was a huge help to me when I was a bench tech. We used to buy as many of our replacement semis as possible from Peavey, even for non-Peavey repairs, because it was actually cheaper than regular supply channels.

 

[JohnRoberts]

[quote author="NewYorkDave"]Peavey? They used to supply their service centers with a nice semi cross-reference guide that was a huge help to me when I was a bench tech. We used to buy as many of our replacement semis as possible from Peavey, even for non-Peavey repairs, because it was actually cheaper than regular supply channels.[/quote]

Yup... not to mention some parts couldn't be bought through normal distribution.

PV was the biggest non-military buyer of TO-3 metal power transistors, but several years back, Motorola sold their only TO-3 factory (in Mexico) and bailed on that business...

I liked metal but nowadays it's all Jap plastic. As far as I can tell, the wheels didn't fall off the power amp cart, so they appear to work adequately well, when properly applied and you don't waste all that time trying to figure out which part to use... :roll:

There were some other house numbers for parts you couldn't get elsewhere.. A version of 5532 selected for 1/F noise, A version of 3080 OTA selected for control feed-through... Nice parts to have around. One very well respected SR company, bought some of the selected opamps for use in their custom electronics. I approved the order as a one time favor since it clearly wasn't service related. Nowadays you can buy superior parts off the shelf.

JR

 

[bcarso]

[quote author="JohnRoberts"] One very well respected SR company, bought some of the selected opamps for use in their custom electronics. I approved the order as a one time favor since it clearly wasn't service related.

JR[/quote]

That reminds me of trying to buy nuvistors. They are used in the electronics of oil drilling heads because they withstand high temps. But some audio people try to buy them as "service replacement parts" from the oil equipment companies. Evidently they do a pretty careful screening.

 

[PRR]

National usta make some power amp chips rated 28V. Bah, I said, they must be on the 40V process. So I ran one on 36V, light load, hefty heatsink, and it lived. About 7 years. One day it was dead. Random defect? Stuffed another one in. About 8 years later, it failed to live.

As you see, I got 15 years use for about $6 cost. And I continued to use the box, which had other features and functions; I just couldn't hear what I was doing (and often it didn't matter). But when I looked around for my third chip, I'd lost my old stash and National no longer makes that exact part. Que sera sera.

 

[JohnRoberts]

[quote author="PRR"]National usta make some power amp chips rated 28V. Bah, I said, they must be on the 40V process. So I ran one on 36V, light load, hefty heatsink, and it lived. About 7 years. One day it was dead. Random defect? Stuffed another one in. About 8 years later, it failed to live.

As you see, I got 15 years use for about $6 cost. And I continued to use the box, which had other features and functions; I just couldn't hear what I was doing (and often it didn't matter). But when I looked around for my third chip, I'd lost my old stash and National no longer makes that exact part. Que sera sera.[/quote]

I'm not sure whether this anecdote suggests that this part works at 36V or confirms the validity of the lower voltage limit? 8 years is not a long time for solid state circuitry. Even if National made these using a 36v process those particular parts may have been de-rated for secondary breakdown or power dissipation in specific internal devices. The failures you experienced may have been anticipated in national's lower voltage rating.

Of course when building one-off DIY projects it's OK to use what you've got, as long as you'll be the one dealing with possible failures.

JR

 

[Guest]

[quote author="JohnRoberts"]

Hfe being 33% high is not very surprising. I wouldn't expect breakdown voltages to vary that much from process targets, more likely explanation is manufacturer substituting a newer technology to sell parts ordered under archaic part numbers. [/quote]

The explanation was simple: "Planned nomenclature". The industry demanded certain quantities of transistors with certain specs, so when crystals were good they anyway had to mark "bad" letter on them. It was in Soviet Union, no market existed, just strict Government regulations. Speaking of modern Toshiba devices, they matched well specks by voltages, and NPN devices matched well by beta, but PNP devices still had wider spread of beta.

 

[PRR]

> I'm not sure whether this anecdote suggests that this part works at 36V or confirms the validity of the lower voltage limit?

I'm not sure what I meant to imply; only an observation. A 2-point dataset.

I would normally expect solid-state to die "instantly" or live "forever".

Or, since there are long-term leakage issues, say 100,000-1,000,000 hours. Decades or a century. Forever-enough for most human purposes.

Figuring my actual power-on hours, it was close to 1,000 hours each time.

A 20% overvolt caused a 99+% reduction in life. Twice.

It almost suggests that, like car batteries, there is a timer inside. "5-year" batteries never fail to fail at 4.5-5.5 years. It must be a timer. The chip seems to have a timer which ticks super-slow at <28V but faster at >28V.

> de-rated for secondary breakdown or power dissipation in specific internal devices. The failures you experienced may have been anticipated in national's lower voltage rating.

Yeah, yeah, I knew they printed "28V" for some well-supported reason. If they could have sold into 36V markets (albeit at higher impedance), and lived, they probably would have said so.

Rational engineering suggests that surface contaminants and buffers are scaled to a certain life or voltage, and ensure failure if you ask for more than 5 years (4.5yr is acceptable with fine-print warranties) or more than 28V.

But some CPUs do have safety mechanisms which compare the external clock to an internal free-running oscillator. If you apply an over-clock which is too close to the free-running oscillator (which runs at a known multiplier of safe logic speed), it shuts-down. So while I "know" a battery or a 27-transistor chip from 1980 "can't" have a sophisticated timer, my conspiracy theory is not totally legless. Some systems are designed to fail.

In any case: observed 1,000 hour MTBF without an actual timer implies that if I built 1,000 of these units, they could start failing in hours, and my warranty return department would be swamped. $2 chip, minimum $20 repair return costs, $200 bad-will..... it ain't worth it.

Also: if this is in the recording path for live performance, Murphy will make sure the failure happens in the best take of your life.

So I guess my point was: even if it works first-time, even if it works for months, going outside the specs is a Bad Idea.

 

[JohnRoberts]

[quote author="PRR"]
I would normally expect solid-state to die "instantly" or live "forever".

---

It almost suggests that, like car batteries, there is a timer inside. "5-year" batteries never fail to fail at 4.5-5.5 years. It must be a timer. The chip seems to have a timer which ticks super-slow at <28V but faster at >28V.
-----
So I guess my point was: even if it works first-time, even if it works for months, going outside the specs is a Bad Idea.[/quote]

I vaugely recall one slow IC failure mechanism related to metal migration due to very high current density. Usually related to a design flaw resuling in an unusually narrow metalization trace. This is not likely in this case.

They may have made the design, independant of the process, voltage sensitive. Imagine a glorified shunt VR, perhaps accidental that stresses a circuit block with increased conduction above certain PS voltages. This would be apparent as overheating so again not likely.

My suspicion is that it may be a "fast" failure mechanism triggered by an infrequent combination of stimuli. I still suspect the poor robustness of small geometry integrated devices.

JR