Maximum junction temperature for long-term reliability

[matthuszagh]

IC datasheets specify a maximum junction temperature above which a device is likely to become damaged. However, I’ve read that long-term device reliability benefits from staying well below this maximum. For example, from the Art of Electronics 3e in 9.4 Heat and Power Design:

It is wise to be conservative in heatsink design, because transistor life decreases rapidly at operating temperatures near or above maximum.

In the example that follows they stay 25C below the max temp.

What’s a safe margin below the max temperature assuming I’m after long-term (more than a decade) reliability? Is 25C sufficient or should I aim for more? Are there any papers that investigate this? Does operating life slowly degrade as temperature rises, or is it more of a sharp falloff above a certain temperature?

[Kleinstein]

For a large part I would expect some thermal activation for the aging processes. So a little like the life time extrapolation used for LEDs with about half the lifetime for 10 K more. So 30 K lower would buy you about a factor of 8.

Towards rather high temperature (e.g. > 250 C) there can than be a thermal run-away of the circuit that may lead to essentially instant destruction. With silicon transistors / diode the ageing is usually not that bad. It is/was a real problem with germanium parts.

p.s.  Another point besides the absolute temperature can be the temperature cycles that cause mechanical stress. This problem is larger with lead free parts (lead free solder does not creep as fast) and depends also on the number of cycles.

[T3sl4co1l]

Depends, as usual; at some point, the package itself decomposes, for example, governed by an Arrhenius relationship.  Most things are like that, but semiconductors have the added wrinkle that they're self-heating (internal temp > package temp) and the current flow itself causes changes (such as electromigration).

I don't know offhand what typical lifetime is, for parts operated at ratings; probably, many manufacturers use MIL-STD-883, which may give you some insights.  That wouldn't be used for measuring (well, inferring) actual lifetime, though.

Also relevant: http://www.hp.woodshot.com/hprfhelp/4_downld/products/xrs/at41511_r.pdf
YMMV, it's been a long time since HP (semis line now Avago, now Broadcom...) made plain old transistors.

Within ratings, expect, ballpark, on the order of 10k's of hrs of operation -- the semiconductors will be the least of your worries at such temperatures -- and probably more like 100k's or more.  And say for 5 or 10% over ratings, probably that goes down very aggressively, like 10x or more?  Depending on which limits, and how aggressively they've been pushed (e.g. Vds(br) causing junction defects and eventual failure, vs. Vgs(max) causing shifted Vgs(th), vs. Ids(max) causing migration and eventual failure, etc.), of course.

As for papers, probably it's so hard to find because it's all ancient history, mostly academic.  Or perhaps relevant documents are commonly available in the high-rel sector, but I don't do aerospace or whatever so I don't know what all shows up there.  (Nor have I had reason to ask a manufacturer directly; but you are always welcome to!)

Tim

[tmadness]

Semiconductor reliability and degradation is a small but active field of research. The causes of degradation are manifold. TDDB and HC degradations are two such phenomenon but there are more:

https://en.wikipedia.org/wiki/Time-dependent_gate_oxide_breakdown
https://en.wikipedia.org/wiki/Hot-carrier_injection

You have rightly identified temperature as one of the biggest factors for accelerated degradation. If you bought your transistor from even a half decent manufacturer (they own the fab) they know how long their transistors will last. You can ask for this info or sometimes this info is ready available on their website under "lifetime" or "reliability".

A caveat: semiconductor degradation is a statistical phenomenon. There is no concrete way of saying that a transistor will last X days/years/decades, there only exist statements like this : "80% of these transistors will last at least 10 years at 85 C ambient"

[trobbins]

I'd suggest the issue is more related to the operating environment and the risk/likelihood of disturbances/perturbations/changes causing Tj to increase beyond what you may have designed for (based on some known and considered variations).  For example, if a semiconductor device is directly or nearly directly exposed to mains supply voltage levels and disturbances (such as a rectifier part, or regulator, or part driven directly from a supply voltage - even on the secondary side of a transformer or after passive filtering) then there is more risk that it could be exposed to excursions beyond Tj max than for say a part buried in a circuit and buffered by many other parts. 

For parts in mains powered power supply equipment, we applied a worst-case 125C design limit (ie. max rated mains voltage and ambient temp and loading etc) for 175C rated parts.

[David Hess]

As a rule of thumb, discrete power devices can operate up to 85C junction temperature with good reliability.  For integrated circuits it is 65C.  For power devices higher temperatures would limit power dissipation anyway.

[Neomys Sapiens]

The abovementioned MIL-STD-883 is the test method standard, it won't help you much.
Mil-HDBK-217 (reliability  prediction) contains the most basic assumptions, MIL-HDBK-251 (Reliability/Design - Thermal Applications) gives some more information.
Also important sources are the 'Reliability Toolkit' and 'Reliability Sourcebook' from the Reliability Analysis Center (RAC).
Very specific information on the derating of components is found in the JPL derating guidelines (JPL-D-8545-D), PD-ED-1201 from NASA GSFC 'PRP: EEE parts derating' as well as RADC-TR-82-177.

Regular books containing more than a glance at this topic are for example:
High Temperature Electronics (F.Patrick McCluskey, R. Grzybowski, T.Podlesack/CRC Press/1997)
Reliability and Failure of Electronic Materials and Devices (Milton Ohring, Academic Press)
Semiconductor Devices in Harsh Conditions (Kirsten Weide-Zaage, Malgorzata Chrzanuwska-Jeske   CRC Press, 2017)
Thermal Design of Electronic Equipment (R.Remsburg (Ed.), CRC Press, 2001)

Industry sources:
IH (Renesas) AN9207 Junction Temperature Considerations
ON Semi HBD-856-D Basic thermal properties of semiconductors

[Berni]

For some of the special high temp rated chips you can occasionally find a graph of expected lifetime in years versus temperature in the datasheet.

But these graphs tend to have the temperature start at something like 120°C and end at up to 250°C. Mainly the problem being that the P and N doping in the silicon starts to diffuse and move around at high temperatures (this is how they originally get the doping in there at the factory, very high temperature). So basically eventually the parts of the transistors that are supposed to be doped loose there P and N properties, or areas around them gain P or N properties and start conducting. So the chips performance would gradually drift before finally giving up the ghost. But this does take a long time at very high temperatures (like many years at 150°C).

You are sooner going to see issues due to thermal cycling. If something is getting really hot that means it will be swinging in temperature a lot between room temp and operating temp. This can eventually work something loose, perhaps not even in the chip but instead crack a solder joint in your PCB.

General rule of thumb (literally) is if you can hold your finger on a component then it is definitely plenty cool enough. This is around 50°C . For high power dissipating components this is sometimes not possible to achieve, but it is perfectly fine to run most things at 85°C all day. Some chips (Like LDOs) are designed to run hot and boast about a 150°C Tj you could run those at 100°C all day just fine. But best to avoid these high temperatures if you can, give those chips some extra ground fill or stick on one of those SMD heatsinks.

[Siwastaja]

As a rule of thumb, discrete power devices can operate up to 85C junction temperature with good reliability.  For integrated circuits it is 65C.  For power devices higher temperatures would limit power dissipation anyway.

This is quite conservative, to the point of not being practical. Tj=65degC on a power IC is almost impossible to meet because local ambient inside a case is easily at 50degC without heroic cooling efforts. I bet anyone who is "following" this rule of thumb only thinks they are doing so, and just have no idea what their actual junction temperatures are.

Both dicrete transistors and ICs are nowadays usually rated to Tj=125degC operational (some discretes rated to 150degC!), which means manufacturer trusts there are no long-term issues.

Of course, the elephant in the room is, how do you prove Tj stays at 125degC? Even if you do perfect thermal modeling of the system, then dust partially clogs a heatsink and Tj rises by 10degC.

I would dare to say most of such "rules of thumbs" are not to actually keep Tj that low, but to add margin against the uncertainty about Tj, also compensating for poor calculations / lack of proper modeling.

Because the aging processes are very steeply dependent on temperature (in exponential way), even just 10-15degC derating is a lot. For a part rated to Tj=125degC operational, I have no doubts running it at Tj=110degC long-term, HOWEVER this means I need to be sure my Tj=110degC really means Tj=110degC. Because Tj is difficult (sometimes nearly impossible, or just expensive; for a discrete diode, you could of course monitor Vf and calculate Tj, but that would be added cost) to monitor directly, then thermal calculations would be done in such way that Tj=90degC or so at maximum ambient temperature specified in the manual.

I bet "65degC" rules of thumbs do nasty things like assuming Ta=25degC always. We should do better than this, and always define max Ta and max Tj explicitly, so we can calculate using Tj_max = Ta_max + sum(R_th)*P.

Another point is, lowering temperature may offer other benefits, such as MOSFETs Rds_on being lower, improving efficiency of most power ICs and of course discrete MOSFETs. OTOH, diodes tend to work more efficiently at high Tj, so there is no other reason than reliability to keep temperature low.

[TheUnnamedNewbie]

As a rule of thumb, discrete power devices can operate up to 85C junction temperature with good reliability.  For integrated circuits it is 65C.  For power devices higher temperatures would limit power dissipation anyway.

I have little to no experience with discretes, but do have experience with IC design. Such a rule of thumb is somewhat pointless as it depends a lot on how the IC was designed. For example. ICs for automotive applications often have to meet very, very stringent life-time requirements at high temperatures - for example, designed and simulated (with very advanced modeling of degradation and variability) to have uptime rates of 99.9999% for >10 years at 125C ambient in the case of AEC-Q100 grade 1 devices (this grade is usually for devices like automotive radars, sensors, etc that need to operate not near the engine, but in places that can get very hot still). Limiting the maximum junction temperature of such devices to 65C is likely to bring more hassle than you will win from a reliability perspective. (for example, you might spend a lot more on a bigger heatsink to cool the device, while all this time, your main failiure point was a capacitor elsewhere on the board, and so you gained nothing from a system reliability perspective).
Not to mention that ICs that use non-CMOS technologies, such as GaN/SiC/etc can often withstand significantly higher temperatures before running risk of damage.

[David Hess]

As a rule of thumb, discrete power devices can operate up to 85C junction temperature with good reliability.  For integrated circuits it is 65C.  For power devices higher temperatures would limit power dissipation anyway.

I have little to no experience with discretes, but do have experience with IC design. Such a rule of thumb is somewhat pointless as it depends a lot on how the IC was designed. For example. ICs for automotive applications often have to meet very, very stringent life-time requirements at high temperatures - for example, designed and simulated (with very advanced modeling of degradation and variability) to have uptime rates of 99.9999% for >10 years at 125C ambient in the case of AEC-Q100 grade 1 devices (this grade is usually for devices like automotive radars, sensors, etc that need to operate not near the engine, but in places that can get very hot still). Limiting the maximum junction temperature of such devices to 65C is likely to bring more hassle than you will win from a reliability perspective. (for example, you might spend a lot more on a bigger heatsink to cool the device, while all this time, your main failiure point was a capacitor elsewhere on the board, and so you gained nothing from a system reliability perspective).
Not to mention that ICs that use non-CMOS technologies, such as GaN/SiC/etc can often withstand significantly higher temperatures before running risk of damage.

I should have qualified my statement as applying to silicon in plastic packages for the commercial temperature range.  Industrial, automotive, and military parts have to operate at higher temperatures, and obviously this does not apply with GaN or SiC.

It is not normally specified in publicly available datasheets, but power ICs have separate junction temperature ratings for their power and control elements.

This is quite conservative, to the point of not being practical. Tj=65degC on a power IC is almost impossible to meet because local ambient inside a case is easily at 50degC without heroic cooling efforts. I bet anyone who is "following" this rule of thumb only thinks they are doing so, and just have no idea what their actual junction temperatures are.

Thinking about what I said, because of how it was suppose to be tested, it should have been case temperature and not junction temperature.

Both dicrete transistors and ICs are nowadays usually rated to Tj=125degC operational (some discretes rated to 150degC!), which means manufacturer trusts there are no long-term issues.

Something I have noticed recently is that some parts have been rated 20C or so higher than they used to be.  I think about 10 years ago some manufacturers started rating their parts for an expected operating life of 5 years instead of 20.