somewhere in my storage shed is a NASA white paper on transistor deterioration, and the use of noise testing to predict transistor failure.
apparently the noise voltage or current in a transistor begins increasing by a factor of 10 or more during the last 10 percent of the transistor's lifetime.
most data sheets (at least from companies like MOT, or IR) have a figure in the data sheet, called MBTF, or Mean Time Before Failure, which is an estimate of the average lifetime of that particular transistor. this figure is different for MIL spec components vs consumer grade components, even for the (almost) same part. that is why you will see suffixes, such as A, C, K, M, etc... for the same parts. the suffix is a code for temperature range and MTBF, and sometimes other things such as radiation hardened parts (so we can take our favorite amp to Mars, through the Van Allen belt, without damaging the transistors)..... almost universally, A, C or K grade components are for consumer electronics, but the only way to know for sure is looking at the data sheet or the data book.
apparently the noise voltage or current in a transistor begins increasing by a factor of 10 or more during the last 10 percent of the transistor's lifetime.
most data sheets (at least from companies like MOT, or IR) have a figure in the data sheet, called MBTF, or Mean Time Before Failure, which is an estimate of the average lifetime of that particular transistor. this figure is different for MIL spec components vs consumer grade components, even for the (almost) same part. that is why you will see suffixes, such as A, C, K, M, etc... for the same parts. the suffix is a code for temperature range and MTBF, and sometimes other things such as radiation hardened parts (so we can take our favorite amp to Mars, through the Van Allen belt, without damaging the transistors)..... almost universally, A, C or K grade components are for consumer electronics, but the only way to know for sure is looking at the data sheet or the data book.
Onsemi has a reliability calculator for the MJ21193:
http://www.onsemi.com/PowerSolutions/reliability.do?part=MJ21193
458,617,914 hours MTBF @ 55C
This decreases to just:
31,136,503 hours MTBF @ 95C
They use the same parameters for the plastic MJL21193.
I really need to pick up a temperature probe and stop relying on my finger. The hard part's actually measuring the silicon itself, and not a point on a heatsink closest to the package.
But on the whole, I'm not worried much about the silicon. The only semiconductor failure outside of once driving one of my Leach amplifiers into a highly reactive load was the random failure of a 2N5087 (noisy). Electrolytic capacitors will usually fail sooner and more often, and sometimes more subtly.
Nelson Pass will more likely have experience with running MOSFETs very hot for extended periods; I seem to recall him indicating that some devices have definite lifetimes operated in Class A at their dissipation limits. Which is to be expected.
http://www.onsemi.com/PowerSolutions/reliability.do?part=MJ21193
458,617,914 hours MTBF @ 55C
This decreases to just:
31,136,503 hours MTBF @ 95C
They use the same parameters for the plastic MJL21193.
I really need to pick up a temperature probe and stop relying on my finger. The hard part's actually measuring the silicon itself, and not a point on a heatsink closest to the package.
But on the whole, I'm not worried much about the silicon. The only semiconductor failure outside of once driving one of my Leach amplifiers into a highly reactive load was the random failure of a 2N5087 (noisy). Electrolytic capacitors will usually fail sooner and more often, and sometimes more subtly.
Nelson Pass will more likely have experience with running MOSFETs very hot for extended periods; I seem to recall him indicating that some devices have definite lifetimes operated in Class A at their dissipation limits. Which is to be expected.
Nelson says, that his machine will work till the time, when universe will be cold ( he probably use " neutron " devices.... ) 😎
hi Damon
ON Semi's page does not say if this is cycled!
According to the old RCA data book (but assembly methods may have changed, and die may be thinner now) thermal cycles were the limitation.
I assume (but maybe wrongly) that the ON semi calculations are for constant temperature conditions?
trouble with device.hours is that there may be a systematic error hidden by the calculation. Let's say 1000 devices last 10000 hours. This is not the same as saying 1 device will last 10,000,000 hours, as they might all pack up at 10001 hours... but often manufacturers use device.hours to generate reliability figures.
What is important is mean time to failure as measured by an actual failure rate!
Maybe Nelson's amps can keep the universe warm...
cheers
John
ON Semi's page does not say if this is cycled!
According to the old RCA data book (but assembly methods may have changed, and die may be thinner now) thermal cycles were the limitation.
I assume (but maybe wrongly) that the ON semi calculations are for constant temperature conditions?
trouble with device.hours is that there may be a systematic error hidden by the calculation. Let's say 1000 devices last 10000 hours. This is not the same as saying 1 device will last 10,000,000 hours, as they might all pack up at 10001 hours... but often manufacturers use device.hours to generate reliability figures.
What is important is mean time to failure as measured by an actual failure rate!
Maybe Nelson's amps can keep the universe warm...
cheers
John
wire bonding has improved since the RCA data book (mine is dated 1965) and device reliabilities have vastly improved, even over the last 20 years. production samples are "torture tested" and MBTF is calculated from the torture tests. by torture testing, i mean tests that operate the device "on the edge" for a short period of time
(i.e. at temperatures that derate the MBTF to 1% of normal) and use the failure rate results to extrapolate the MBTF at normal operating conditions. if there weren't some kind of "shortcut" method of determining MBTF, it would be a very long time before new devices could go to market.
(i.e. at temperatures that derate the MBTF to 1% of normal) and use the failure rate results to extrapolate the MBTF at normal operating conditions. if there weren't some kind of "shortcut" method of determining MBTF, it would be a very long time before new devices could go to market.
I understand the original question was around parameter change over time, causing changes in operating points, etc., and not catastrophic failures. While the latter is also important, it would be interesting to know if anybody has any hard evidence or at least previous reports or articles on the subject.
MRupp said:
Since any competently designed amp will not significantly change its operating point even with component parameters that vary wildly between components, like transistor gain etc, I would be surprised if there were any reports at all.
Jan Didden
I agree that every competently designed run of the mill circuit must try to accomodate parameter spreads, especially if you manufacture them by the thousand.
Having said that, I am aware that at least some competent designers take great care to closely match their components, e.g. IDSS matching for complementary JFET pairs, selecting low noise components, etc.., which should allow them to explore topologies for their handbuild circuits which they cannot put into a mass product. And for diy, close matching should also practical.
So, deterioration of any of the critical parameters might be an issue, if it really exists, off course. And I have some anecdotal information that I should avoid certain types of JFETs, specifically older plastic package devices , and use metal case JFETs instead (low noise was the issue here, and plastic trannies were said to deteriorate).
Bottom line it could be an issue ...
Having said that, I am aware that at least some competent designers take great care to closely match their components, e.g. IDSS matching for complementary JFET pairs, selecting low noise components, etc.., which should allow them to explore topologies for their handbuild circuits which they cannot put into a mass product. And for diy, close matching should also practical.
So, deterioration of any of the critical parameters might be an issue, if it really exists, off course. And I have some anecdotal information that I should avoid certain types of JFETs, specifically older plastic package devices , and use metal case JFETs instead (low noise was the issue here, and plastic trannies were said to deteriorate).
Bottom line it could be an issue ...
MRupp said:
Moisture incursion would be the main problem on older plastic parts. I have not seen any evidence for deterioration of specifications without some external root cause. Transistors see hundreds of degrees C when their diffusions are formed. From classical activation energy arguments, at room temperature parameter changes occur on the time scale of eons.
Hi all
IN reality, negative feedback counters most parameter changes between devices, so a parametric shift within a device is unlikely to cause any audible differences at all.
This leaves catastrophic effects as the most likely problem, likely to occur suddenly a few decades down the road. One contributor's Class A has experienced what might be a premature failure due to thermal cycling - or reverse bias base junction.
One exception to parametric change is whether noise degrades as junctions are more sensitive to surface defects and contamination MIGHT cause noise to increase before cat fail, especially in plastic trannys.
And I agree that high temperatures are needed to accelerate testing - but we do need MTTF, not device.hours as the reliability guide.
BTW my nearly-30 year old RCA-transistored amp is still working and in regular use...
cheers
John
IN reality, negative feedback counters most parameter changes between devices, so a parametric shift within a device is unlikely to cause any audible differences at all.
This leaves catastrophic effects as the most likely problem, likely to occur suddenly a few decades down the road. One contributor's Class A has experienced what might be a premature failure due to thermal cycling - or reverse bias base junction.
One exception to parametric change is whether noise degrades as junctions are more sensitive to surface defects and contamination MIGHT cause noise to increase before cat fail, especially in plastic trannys.
And I agree that high temperatures are needed to accelerate testing - but we do need MTTF, not device.hours as the reliability guide.
BTW my nearly-30 year old RCA-transistored amp is still working and in regular use...
cheers
John
- Status
- Not open for further replies.