How hot should silicon run?

[simonbramble]

You know you get those questions where you ask 10 people and get 10 different answers... This could be one of them.

How hot should a silicon device be run before you should start getting worried about its lifetime? Does it differ from device to device, say MOSFETs vs linear regulators and more integrated electronics?

I appreciate the Arrhenius equation states that lifetime drops off as temperature goes up, but are there any general rules of thumb that say 'Don't operate above this temperature if you want a lifetime of, say, 10 years'?

I am particularly interested in hearing from engineers who have designed high reliability equipment, life support, military equipment etc. Stuff that absolutely has to work for decades

 

[ronsimpson]

What electronics? It makes a difference.
I pulled the data sheet on a power transistor. 2N3055.
The silicon should be from -65C to +200C. That is not the temperature of the case, that is the silicon. It takes some math to get to the case temperature.
You need to know what is the temp of the place your product lives at. Room temp is about 25C. Most products can be used in a much hotter place. When I worked for Samsung we think about a house for temperature. At HP/Agilent we make the electronics work over a much wider temperature range. When I was doing car electronics -40C was the spec but I have seen my product fail at -45 C. On the hot end a car can get really hot.
Simple answer. For consumer electronics I want most parts to be so I can put my finger on the part and leave it there. That is a very crude way of saying the part is 25C over a room temp of 25C. (50C) Transformers I run hotter. Capacitors less hot.

[the air is 25C, IC's case is 50C, the silicon is 75C, then you put the electronics in a plastic box that stops air flow and adds another 25C]

This gives me some room. Many consumer products are designed to run at a room temp of 50C.
Remember it is typical for a person to pile paper on top of the computer. Or put the electronics on a blanket. Or close up the product in a cabinet.

 

[simonbramble]

Hi Ron. Thanks for getting back to me. Yes I am aware of theta JA and the calculation behind it. Funny you should say about the finger test, but I have always said that if you cannot hold your finger on a device for more than about 1/4 second, then it is running at about 50 degC. I dont like seeing (feeling, smelling?) components running above about 50 degC, but some engineers are OK with products running up to well above 60 degC or higher

The problem I have is that engineers always quote an 'acceleration factor' (from Arrhenius) whereby an IC will fail 30x more frequently at 100 degC than it will at 50 degC, but I dont have a number to start with (at 50 degC or 25 degC). How likely is the part to fail at 50 degC? If I had to design a product that absolutely had to work for 10 years, what is the hottest temperature my components can run at.

I am specifically interested in MOSFETs, but parts of smaller geometries are of interest too (controller ICs and other analogue components) which is why I was asking for rules of thumb from engineers who have designed equipment that definitely is not allowed to fail.

I am still interested in other thoughts from other members please...

Thanks

Simon

 

[audioguru]

When I think about temperature changes then I think that they cause a device to expand and contract. Then eventually the device will crack or have fatigue.
If operated continuously at one temperature, even if very hot, I think the device will last much longer than if its temperature changes frequently.

 

[Ian Rogers]

I see the need to make things cheap... BUT!! something has to give.... I have repaired motor speed controllers on Japanese cranes... These are fit for about 6A... The motors get tired about 1 year on and start pulling 6+ Amps... The cost of production at 6A is quite a bit less than say 10A..

I retrofit slightly better components and old tired motors continue for at least another 2 years.... Running a device near to max temperature will derate the device over a period.. Take a humble LED... Run it at 10mA forever or 30mA for a good while.... The visibility isn't that impaired..

 

[Nigel Goodwin]

I retrofit slightly better components and old tired motors continue for at least another 2 years....

For service purposes it's always best to fit higher quality components - as the parts costs is usually only a tiny fraction of the labour costs - so do it right, do it once. Different in production, where labour costs are small, and total parts cost is high - so fitting cheaper parts makes a big difference.

 

[kubeek]

If the semiconductor works life at some elevated temperatuure, then you can use the arrhenius eqation to estimate the acceleration factor. But if the semiconductor is used sporadically and has significant thermal cycling, then you will be looking at much worse numbers.

The actual number vary based on the reliability standard that you use. We for example use SN29500. You can use the free MTBF calculator to get reliability data based on running temperature and voltage stress in transistors. https://aldservice.com/Download/download-reliability-and-safety-software.html

 

[Dick Cappels]

Really, you should consult the datasheet for each device you are using. The manufacture publishes its best estimate of maximum temperature based on its understanding of the device. It is not only that it is made of silicon (or other material) but also its geometry thermal resistance to the case and probably a few factors). That is why you cannot get a consistent answer.

 

[MacIntoshCZ]

random numbers from me: 70 degrees OK, 90 degrees TOP

 

[simonbramble]

I have what I think is the definitive answer.

The Arrhenius equation states that failures accelerate as temperature goes up. However this equation only details failures at the *silicon* level. It does not take into account mechanical stress associated with the part heating and cooling and the device maybe expanding at a different rate to the PCB and thus, over time, causing poor solder joints etc. It also does not take into account corrosion effects that get worse as temperature goes up.

It also ignores the effect of raised temperature on the surrounding components. Leakage currents in silicon double every 10 degC rise in temperature, offset voltages get worse, analogue reference voltages drift, accuracy goes down and data memory retention deteriorates, so although you part is happy running at 125degC for the next 175,000 years (see below) the effect of temperature has a huge impact on the performance of the rest of your circuit.

Looking at silicon only, the Arrhenius equation states that a part is 943x more likely to fail at 125degC than at 25 degC. Arrhenius is based on exp (-qV/kt) where q = 1.602 x10E-19, k = 1.38 x 10E-23, T is Kelvin and V = 0.7)

So we know the acceleration factor, but we don't know what number to start with. Looking at the failure report of a random component (LTC3891) this details a statistical failure rate of 0.65 parts per Billion device hours at 55 degC. This report can be found from the landing page on www.analog.com.

So 0.65 failures per billion device hours is 1,538,000,000 hours per failure (taking the reciprocal of 0.65/1Billion). Divide this by (24 x 365) to give years per failure (or mean time between failures, MTBF) = 175,000 years.

So, at 55 degC you can run a part for 175,000 years before it will statistically fail.

At 125 degC, this goes down to 2,250 years.

So although, as temperature goes up, so do silicon failures, the numbers are minuscule.

However, as temperature goes up, mechanical and chemical problems start getting worse and these are far far more likely to cause a circuit to malfunction compared with anything dictated by Arrhenius.

I continue this argument in another comment (see below)....

 

[KMoffett]

Add in the lead-free debacle. But then, consumer goods are disposable in a few years at best.

 

[gophert]

Looking at silicon only, the Arrhenius equation states that a part is 943x more likely to fail at 125degC than at 25 degC. Arrhenius is based on exp (-qV/kt) where q = 1.602 x10E-19, k = 1.38 x 10E-23, T is Kelvin and V = 0.7)

So we know the acceleration factor, but we don't know what number to start with. Looking at the failure report of a random component (LTC3891) this details a statistical failure rate of 0.65 parts per Billion device hours at 55 degC. This report can be found from the landing page on www.analog.com.

The in-house testing is approaching your calculated 2250 years (over multiple chips and packaging styles). Already at 18M hours...

https://www.analog.com/media/en/technical-documentation/reliability-data/r526.pdf

 

[simonbramble]

The in-house testing is approaching your calculated 2250 years (over multiple chips and packaging styles). Already at 18M hours...

Yes - I noticed that too. The theory seems to tie up with the practice. Thanks for the clarification. Reliable stuff indeed

 

[simonbramble]

Add in the lead-free debacle. But then, consumer goods are disposable in a few years at best.

With consumer electronics it is likely to be the electrolytic capacitors that dry out.

 

[audioguru]

The datasheet for some chips that have "thermal shutdown" say not to repeat the thermal shutdown often to prevent the heat expansion and cooling contraction from fracturing the chip.

 

[simonbramble]

Just for completion for those who stumble on this post in future, thermal shutdown normally activates at 165 degC +/- 10 degC.

I agree that you should not get anywhere near that even at maximum ambient temperature. Even if you do, you are definitely exceeding the maximum junction temperature (TJMAX) of the device which is normally 125degC or 150degC

 

[simonbramble]

Me again... There is still some confusion in the industry about temperature and 'how hot is too hot?'. Many older engineers seem to think that running a component above 100 degC limits its lifetime. I can see where this opinion comes from. Back in the 90's failure rates of components were measured in parts per Million device hours. Now it is measured in parts per Billion device hours, so things are 1000x better today.

Putting this into perspective, a part that will last for 2250 years (see above) when run at 125 degC made on *today's* silicon processes will only have lasted 2.250 years back in the 90's (when parts were 1000x less reliable). Obviously a circuit only lasting for 2.25 years instead of 2,250 years is a completely different issue. I think this is where this opinion comes from that running circuits hot limits their lifetime. This was true in the 90's but not today (based on Analog Devices reliability data anyway).

So.... it is OK to run a part at 125 degC and still expect thousands of years of operation out of it. However, what will kill your circuit is all the other stuff that high temperature causes, some of which are outlined in my post above.

 

[Nigel Goodwin]

Me again... There is still some confusion in the industry about temperature and 'how hot is too hot?'. Many older engineers seem to think that running a component above 100 degC limits its lifetime. I can see where this opinion comes from. Back in the 90's failure rates of components were measured in parts per Million device hours. Now it is measured in parts per Billion device hours, so things are 1000x better today.

I think you're living in fairy land

Parts fail now just as they have always done, and one of the reasons is running them too hot, and even more so cycling them between hot and cold.

As a service engineer you can often tell what sort of parts are going to fail, often it's becaus eos serious design flaws (and manufacturers never learn!), or because components are run too close to their limits.

 

[gophert]

Back in the 90's failure rates of components were measured in parts per Million device hours. Now it is measured in parts per Billion device hours, so things are 1000x better today.

Putting this into perspective, a part that will last for 2250 years (see above) when run at 125 degC made on *today's* silicon processes will only have lasted 2.250 years back in the 90's (when parts were 1000x less reliable). Obviously a circuit only lasting for 2.25 years instead of 2,250 years is a completely different issue.

Thanks, I needed a good laugh. By your calculations, any device with 12 chips on board will last 60 days. I don't think so. Also, you might want to look back at your 1990's data. It may .have said 5000 ppm, and your recent failure rate 5.5ppb.

 

[Dick Cappels]

If you are making just of a few of your product and you are willing to take the risk, go ahead and run it as hot as you want, but if you are going to make millions of widgets and want to protect your company's reputation for reliability you need to pay attention to the component manufacturer's advice. At the very least when they bring you loads of dead widgets you will have the manufacturer's engineers there with you to help figure out a solution.