I've no hard data, but I sure think it's the way to bet.
The thermal expansion/contraction of repeated on-off cycles, tending to make something "crack", coupled with the fact that chemical reactions take place faster at higher temperatures, would make me think that lower temperatures would make for longer life.
There is a rule of thumb, that an increase of 10 degrees C, halves the life. To determine the mean time to failure, they generally run accelerated life tests by testing at elevated temperatures.
That said, it may not make any significant difference to you. Back in the 60's the better transistors had a mean time to failure of about
1500 years ( based obviously on accelerated life tests ). If your triacs had a mean time to failure of 1500 years based on a junction temperature of say 30 degrees C, then running them at a junction temperature of 70 degrees might reduce the mtf to 90 years. Just remember the temp limit is the junction temp, not the ambient temp.
I can't give you a scientific answer, but I used to run thousands of triacs in outdoor traffic control boxes in south Florida. We seldom had a failure that could not be explained by lightning or a short circuit.
Proabably. Running right at the limits will shorten the life. But thermal cycling can really kill power semiconductors, especially the cheap-a** TO-220 type. The bond between die and leadframe/heatsink fails, causing the semiconductor die itself to overheat. I've seen failures in the range of xE4 cycles where x is a single digit.
Also, if the die is already very hot, a surge of current that it might otherwise survive can heat the die enough to destroy it.
Triacs used in traffic control loadswitchs are very conservatively applied, or at least they were in the '70s. Lamp loads are tough because of the arc that occurs when a lamp burns out. Most of the triacs were at least 20 amp parts and some were bigger.
Lightning in South Florida was a problem. We got some controllers back from Dade County that looked like they'd been dipped in a volcano! I think some of the 8-phase dual-ring controllers I designed back then are still in service in MN.
Yes. The lifetime of a semiconductor device is directly related to its operating (junction) temperature. The cause of eventual failure (assuming operation within the envelope) is a result of migration of material within the junction, which is exponential with temperature as I recall.
Whether the reduction in lifetime is of concern depends on how hot the device runs, how close to rated limits it is used, and what transients it will be exposed to.
My brain is swiss cheese these days but if I remember correctly, aren't Triacs and SCRs a device such that the higher the temp, the lower the internal resistance? I seem to remember something about people putting them in parallel and a vicious cycle starting where one would conduct a higher percentage of the load, heat a little more and drop internal resistance, therefore heat more until it blows.
Just curoius...one of these days I have to switch an inductive load on a forging machine somehow. Output is about 6000 amps at 3 volts. Makes the theoretical (no loss) input about 75 amps at 240 V. The problem is that things are so highly inductive, any digital switching would have to be far greater than the 75 amp rating. That gets spendy. If you could gang the darned things, the cost would be MUCH better.
Any ideas on a cheap but digitally controlled solution? In the old days, one might use mercury relays. Regular relays will fry and lock. Of course there are ways to suck up the initial inductive load but I'd rather keep things simple.
I think that you've oversimplifed the situation. I don't think you will find much difference in lifetimes between a semiconductor running at 20 deg C and one running at
80 deg C. Not today with modern parts. As long as the part is working with within it's current and voltage safe operating area and within it's temperature range, you should not expect more failures than if it were running at 20 deg C.
Bob Pease, probably the world's greatest authority on analog IC design wrote some good stuff on this subject years ago. Unfortunately, I can't find it on the web. As best as I can recall, his position was that the standard lifetime derating curves for temperature that the military and aerospace companies use were in error and lifetime stays pretty much linear until you exceed a certain temp.
Take the temperature above the specifications and all bets are off. As another poster suggested, I suspect extreme temperature cycling with plastic packages will reduce lifetimes as well.
Inductive loads don't have an initial current surge. The problem is a voltage surge when you try to turn them off, unless you turn them off at an instant when current is passing thru zero. Back-to-back SCR's inherently do this and are suitable for switching inductive AC loads. You can buy these as "solid state switch" modules. These guys have modules with ratings up to 125 amps and 660 VAC:
I've seen "hockey puck" IGBT's and SCR's in the several thousand amp range. Haven't seen much on ebay but there are on the surplus market. Several hundred bucks a pop, compared to several thousand bucks a pop new. IIRC, the voltage range was in the thousands, though, and wouldn't know what the forward drop would be. There are lighter duty devices ranging from minimal amperage to several hundred. Look for IGBT and SCR's on ebay. To switch AC fully with SCR's you need two back to back, which is essentially what a TRIAC is.
As far as I know the characteristics of materials hasn't recently, although manufacturing processes have. Modern parts are better but the effects of temperature on material migration is solely a materials property. Testing that I was involved in some 15 years ago showed measureable changes in device characteristics after relatively short runs at the upper limit for a number of devices. Extrapolating that out results in the prediction of earlier failure as compared to the same device running at the lower end of the temperature range.
I used to have a number of papers from other studies done on a wide variety of devices that supports what physics says should happen.
That's not to say that the manufacture's data is wrong. Many devices have a published lifetime vs temperature spec and on average those devices will run according to that curve. But that wasn't the question the OP asked.
In particular, there are existing cracks in chips resulting from the dicing process (breaking apart the silicon wafer into all the individual chips).
If the crack is curved so it points back to the edge -- no problems. However, if it is pointed towards the active area, then it will grow with repeated temperature cycling (note that staying warm is better than shooting up and down frequently). The action is like that of a crack in your windshield. It may sit there all summer as a nice short little crack, but once winter comes, and you start blowing heated air on a frozen windshield, that crack will grow.
One of the things which makes transistors and ICs made to mil specs so expensive relative to commercial ones is that each chip is microscopically examined, and those with cracks which are pointing towards an active area are rejected (perhaps to wind up in commercial chips with the same basic part number.
Another issue is that the die are typically wire-bonded to the head it is on. The bonding wires are relatively fragile and the bond between the die metalization and the wire itself is a source of failure. Again this is exacerbated by thermal cycling issues.
When I wire bond stuff, I use a puff of gas from a hand-held gun to apply force to the bonds. Any ones that are weak will detach at this point and ones that pass this test will not fail under repeated thermal cycling.
Many commercial ICs have the bonds encapsulated - but the combination of Si die, aluminum wire, and polymer goop never have exactly the same thermal expansion.
I went back and reread your post and I don't disagree in principle. The failure rate may indeed be exponential with temperature. The
*practical* issue is "where are we on the knee of the curve"? If, at 80 deg c, we are still way down on the flat part of the curve, the temperature effects on reliability are very small-to-nonexistant. I maintain that this is the case with modern power semiconductors in decent metal-and-glass packages, operated within their specifications.
I would have no problem designing, selling, and standing behind a warranty on a product whose parts are operating within the manufacturer's specifications, even if it meant running a power semiconductor at 80 or 100 deg C, as long as I knew that worst-case, the value would never be exceeded.
Triacs don't tend to have very good reliability in a fairly large class of applications. The temperature is but a small factor (in a decent design). Bigger factors (aside from the thermal cycling that I already mentioned) are current and voltage surges. Ordinary fuses are not fast enough to protect fragile semicondutors the way they are often derated (eg. using a 10A or 8A triac at 5A). Ordinary MOVs may be not sufficient to clamp voltage transients at non-damaging levels, unles the semiconductors are rated at much higher than normal voltage. They are really quite delicate compared to a hefty contact, and what's often worse, they typically fail "on". Without massive overrating (eg. using a 40A/1000V triac at 4A/240V, which costs a lot more money) it may not be possible to protect them adequately at all.
It would be nice if a semiconductor manufacturer were to address the issue by coming out with relatively inexpensive, yet rugged large-die parts in low-cost packaging, but it's not happened yet that I know of.