In one very distant galaxy, there were old processors. They worked normally at the temperature of people 36.6, knew how to count and write, and they ate only 640 kb memory. But large corporations, parallel calculations, gamers, users, miners came … And the abyss of technological progress slowly absorbed the old iron. The new soldiers of high -speed calculations have fallen into the armor of radiators, fans, COOs and, at some point, +60 degrees – became a normal operating temperature. High technologies, a rich culture, and now still hot silicon in colors of RGB response. In addition, buying/selling components in the secondary market, too, has long been a new reality. I think to consider the obvious it makes no sense, agree?
Another thing is interesting, but as semiconductors, in general, they relate to temperatures? The year of mining in a basement with a temperature of +85 is normal or a lot? Why would silly in microcircuits grow old – this is sand? So look, pebbles in the sugar for millennia lie in the heat and nothing is done to them. Answers to these questions lie somewhere in the depths of modern physics. We will not go there – still nothing is clear, but to look behind the curtain, probably, it will be curious. I do not promise accurate numbers, but I hope that the conclusions at the end will be worthy of attention.
So that the course of my thoughts can be seen more clearly, I compose a mnemonic rhyme on the motive "The house that Jack built".
Here is a map,
Which mines the coin.
But an overheated memory,
Which began to melt slowly,
Inside the card
Which mines the coin.
And this is a substrate,
Which boils a little,
In memory,
Who wants to melt,
In the heat of GPU,
Which mines the coin.
And this is a transistor moosfet,
Which cannot in a discrete signal,
In the substrate,
That boils not a little at all,
In memory,
That everything is ready to melt everything around,
Inside the card,
Which mines the coin.
Here is a prohibited zone,
Which in holes from electrons,
The transistor was broken by the Mosfet,
Who is not at all discrete,
In the substrate,
Which was touched a little,
In memory,
Where there is nothing to melt,
In the map,
Which mines the coin.
And finally, here are the electrons,
Which the nodes do not sit in the crystal calmly,
Which stores the prohibited zone,
Which in holes from electrons,
The transistor broke a little mosteum,
Which has gone through all the signals discrete,
In the dead frying tired substrate,
Which is “crunch!" – not at all a little,
Which holds a healing memory,
Where nothing is left to dour,
In the map,
Which smeared the coin.
1. Quantum fairy tales.
Now there will be some quantum physics for housewives. Moreover, the "housewife" in this chapter, I will be dear reader. Before this investigation, I had never steamed on the question of why the components were burning. The manufacturer in the documentation said 0 …+85 – “This is a commercial grade”, so we use. In the flow of electronics development on such things, it is too late to focus on too late.
At first there was a particle. And the particle was an electron. And it was impossible to see it, because the dimensions are less than the length of the light wave. Smell, feel and bite too – Mr. Geisenberg “forbade” the principle of uncertainty. It is difficult to even realize, because if you want to catch it, then it, suddenly, is not a particle, but a wave, therefore it can be at several points at once. Scientists came up with what to do with this outrage (check the pun: “AN-SIME”) and described his behavior with a wave function. This is a kind of mathematical Abracardra, which takes into account the coordinates in space and time, issuing at the exit The density of probability the stay of our little quantum friends. Wave functions are obtained by the solution of the Schrödinger equation. An even more incomprehensible mathematical cadabra, the essence of which, I believe, is in the search for energy balance. There is a special charm of fundamental physics: many of its sections are quite often based on several simple (only in the record) equations that humanity can comprehend whole eternity)) it is clear, it is impossible to find Schrödinger’s decisions without the Aztskih Matan, but to guess the character, it is quite … Gentlemen Bor and Rutherford have posted that there are orbits in the atoms. So they “rotate” in a circle with some frequency. And for periodicals in matan, trigonometric functions are responsible. Which, after some substitutions, degenerate to a conditional set of numbers called a discrete spectrum (which refers us to the fact that electrons do not “rotate” in a circle, but rather are in its space with a certain probability). Or, also, the levels of permitted energies. They are multiple of the number of bar. The first rule of the number of the bar – nothing can be less than the number of the bar.
Of course, “Feynman lectures in physics” are best disposed of a physical confusion. The great maestro, juggling delightly simple concepts, interprets very complex processes. In the picture above – experience from the third volume (chapter 37, p.2.). Bullets – electrons, 1.2 – control holes in the walls, P1, P2, P12 – probabilities. This interpretation does not remind you anything?
The same scene from the second series of Ch2019, where Legasov explains for the operation of the reactor, comparing atoms with bullets.
The next question is in this semantic ladder: “And what about the electrons in the substance?"The" zone theory "is engaged in the response to it. She says that a solid body is a crystal in which there are a lot of atoms, and, respectively, so many electrons that all their levels of permitted energies merge into continuous zones. “Conductivity zone” is … hmm … I do not understand the abstract part of the concept. Too complicated fison. In a certain approximation, this is also an energy range, allowing the charge carriers to skip from orbital to orbit. In the “Valence Zone”, everything is a little simpler – the electrons are “tied” to atoms, they are normal. The whole story is interested in the sake of the picture below.
It is already more and less clear on it. In metals, these two zones overlap, which allows you to transmit an electric signal with minimal efforts. For a long time, I imagined a copper wire with a kind of “pipe for electrons”. Type, they, like water, fly under the influence of voltage from one end to the other. Similar analogies provide all “simple” textbooks for electricians. In fact, everything turned out the other way around. In this pipe, the Kulon’s uglys stand in a row grabbing each other’s hands, and when the potential appears, they transmit the “handshake” to the neighbor “wave”. And the electron that entered this chain from the power side pushes another electron from the load. Somewhere here, conductivity appears with the speed of light, but the carriers themselves are lazy and inactive. Something like Moscow roads: a huge number of people come to their work, but, at the same time, everyone is in traffic jams. MCED – Magic of quantum electrodynamics.
At the moment when I understood this, many thoughts in my head took their places. For example, high -frequency phenomena (CSV, reinforcement, coordination of impedances, antenn theory) settled in a circle). Want a small check? Without looking at Wikipedia, try to guess what will happen if you connect a piece of wire to the signal generator and give a single impulse? And what looks like a ruler, the edge of which was pressed with a finger to the school desk and made "Trunk"?
Everything is also clear with dielectrics: by default, electrons are so good with atoms that you need to apply tangible power to raise the signaling "revolution". Can plastic be carried out? Of course, give a couple of millions of volts and get delightful “figures of Lichtenberg”.
Semiconductors – are located in the middle, between insulators and conductors. Yours, captain obviousness)) although the concept itself is no longer so transparent. As follows from the text above, the zones of valency and conductivity are not crossed, but the energy gap between them is quite small. This means that, by default: our Silicon Friend behaves almost like an insulator. It is worth throwing a little energy thrush and it will be moved to the side of the conductors. Salt is how to throw.
Photoresistors are a clean semiconductor under a glass. Sprinkling photons can be almost linearly reduce its resistance, t.e. Measure light.
Thermistors are the same, only the resistance changes depending on the temperature. Moreover, in different versions, the dependence may have a direct or reverse slope.
Imprisoned gloom – this is the bricks of modern electronics. They earned their “half-” because they conduct current only one way. Suppose valency of silicon is 4, which means it can form 4 connections. Whatever the crystal is bored, a five. In this situation, 4 connections and a “dangling” electron can form, which will become the carrier of the charge. In this case, they talk about the semiconductor of “n-type” (Negative). If the neighbor turns out to be trivalent, then three connections and a “hole” from the electron are formed.
In quantum physics, “Key drawing” also canows. The holes equally or late will drag an electron to themselves, but outwardly it looks like conductivity, on the contrary or “p-type”.
2. The plastic world won. Mosfet was stronger.
In this part of the performance, we still reach winstler.uk/ Mosfet transistors. This is the main switch in the structure of any microcircuit. It consists of a doped substrate (Substrate) – which is the crystal of the impurities. Two wires are stuck in it, one drain stock, the second source of “Source”. In the usual state – this is just a brick, does nothing, he is purple on everything. Energy shows here are thrown through the “Gate” shutter isolated by the oxyl. Since there is no direct contact with the substrate, the current needs a very small. Speaking closer to science, an electric field on the shutter, we inform the charge in the substrate inducing into a conductive state. That is, the system, generally speaking, works as a capacitor, although the concept can be traced the same as that of a vacuum triode in my article about Outer Worlds. A small signal on the shutter can be steamed with horse capacities between the drainist.
A little thing, a little thing and a square, that’s a silicon-metall-oxide semiconductor field transistor (CMOP = Mosfet-Metal Oxide Structure Field-Effect Transistor)
All this physical physics was needed for eyeliner by the beginning of the investigation. The first request in Google on the degradation of semiconductors brought to a wonderful article of Russian scientists on the effects of temperature on the MosFET structure.
Let me steal a piece of paragraph from the motivational part in a quote. I won’t say better.
With prolonged storage of MOS, integrated microcircuits (IC) accumulation of defects and, as a result, degradation of electrical characteristics, such as threshold voltage, steepness, etc. D. One of these defects are superficial conditions (PS), which are formed in the underlying area of the semiconductor.
One of the main mechanisms of the formation of PS is a rupture of stressed valence ties according to the reaction of the type Si – SI + H + → Si • + Si, where h + is a free hole, si • – three -coordinating silicon with one dangling connection. When the SI -Si -Si -Si -SI stress is broken, the energy level occurs in the prohibited zone of the semiconductor, which can be captured by the charge, which leads to the degradation of the IMC MOS.
Behind the scenes of this quote, it is said that actually semiconductors are very tender guys: radiation, temperature and other hard operating conditions spoil them. If Crystal receives a micro -mode, or rather, the silicon atom loses in touch on it, then it will not be exchanged by electrons, as conceived. And as we recall from the physical fairy tale above, the electron entering the “prohibited zone” is not permissible. This means a) We have an electronic parasite where it is not necessary, and b) in the place where it should sit, the “hole” remained. Concentration begins to shift somewhere. As far as I understand the article, not concentration itself, but the general boundaries of the “Valentine” and “conductivity” zones. Outwardly, this manifests itself as a change in the gear characteristics of the transistor, t.e. Its parameters "float". In the terminology of Mosfet, one of the key values is the threshold voltage (V Threshold). This is the minimum amount of volts of potential, which needs to be hung in the shutter so that the transistor substrate began to drive (or would not finish). T.e. Mosfet would begin to “get touched” (or vice versa, approach the “cut -off”).
So, gentlemen, the authors of the article conducted a classic experiment on some domestic microcircuit of logic K1526LP13, which has many connected Mosfet transistors on the crystal.
The temperature t = 140 c was used as an accelerating factor. Tests were carried out in a thermostat in passive electric mode. After exposure at elevated temperatures during a certain period of time, the samples were cooled for 24 hours and the intensity characteristic at room temperature was measured at room temperature.
UIT delta – this is a threshold voltage. Starting 0.654 volts – turn into 1.093 in 1000 hours.
On the basis of these experiments, they deduced a certain “obsolescent coefficient” (this is my interpretation, it officially sounds like “PS formation rate”) with a dimension of 1/h. For a test chip, it turned out about 0.00037 1/h. This may mean that after one and a half thousand hours the voltage required to open a transistor on the crystal will double. What can this lead to? Well, the rest of the scheme is not a fact that it knows about the overheated chip, and switching errors may climb. How to determine the logical levels of “0” and “1” when all the parameters went upward?
3. Mathematical Zadonons.
Generally speaking, the value of 1/h is quite abstract. A logical question of how something analytical can be pulled out of it? Here mathematics of the ratio come to the rescue. It seems to me appropriate to quote the conclusion of the article (this is the time for today, I swear):
In the work, a physical model of the aging process of a pray gun and a calculation-experimental way has been presented with an energy characteristic of the structure (energy level of intense ties), which determines the kinetics of the process of accumulation of PS. The obtained values of the energy levels of intense connections from MOPs and bipolar ICs are almost the same (−1.18 and −1.14 EV), and it can be assumed that the level of energy of intense ties for any IMS lies within these limits.
If “for any”, then you can try to do the next trick:
Take the formula for the calculation of the aging rate depending on the temperature (from the article). It is most of the physical constants. And take temperature data for 140 degrees Celsius from the practical part in the neighboring chapter (article).
Calculate the relationship of aging speeds. At the same time, long tail tails will be reduced, and the result will show us how much the difference in temperature increases the wear of the chip.
It may sound difficult, but this is a completely elementary algebra of secondary school. As they say: see the picture – follow your hands.
W140, W130, W85, W95, W75 – the speed of "wear" at different temperatures
T140, T130, T85, T95, T75 – temperatures themselves. They are transferred to Kelvina so that the dimension converges.
Ratio – ratio of aging speeds.
Guided by the consideration above, first we compare the rate of formation of defects at 130C and 140C, and we get the acceleration of crystal degradation by 2.33 times. (See the ratio of Ratio1.) Perhaps you heard an empirical radio amateur expression: "Improving the operating temperature by 10 degrees reduces the life of the microcircuits by 2 times". It seems that this formula is beating well with him. But, of course, the video cards do not work at such temperatures, so the next formula for W85 will actually resurrect the calculations to the values of the maximum working heating of the GDDR6 RAM allowed by the manufacturer. From this value we put off 10 degrees up – these are conditions under which the memory will still work, but the processor is obliged to apply “lowering” sanctions to survive the system. Then we set aside 10 degrees down. Well, what is 75C – almost ordinary work under load in some RDR2. Now let’s take an additional look at the warranty agreement from NVIDIA.
In this case, the video card can be considered as an indivisible product. This means that the manufacturer guarantees 3 years of work of each component, at a maximum commercial temperature of 85C. By introducing our calculated coefficients, we get that somewhere in a parallel universe, at a temperature of 95C: the warranty should act something like: 3 / 2.941 = 1.03 years. And in another parallel universe at 75C: 3 * 3.085> 9 years!! Fiction? Yes, not really, in the sense that in 5 years, iron will be morally obsolete.
In my personal experience (it is certainly not representative) – not a single video card has burned down. And this is a line of: MX440, 8400GS, HD6870, Divine furnace GTX580, GTX1080. And GTX580 – it was really a stove, with a normal operating temperature 90+. I also had the experience of buying two GTX1080 after mining from a trustee – and everyone works as new (albeit resold, because it did not work out with SLI). Therefore, I want to listen to other people’s stories. If it’s not difficult for you: tell us where you bought used video cards and what impressions of use. And if suddenly, there are owners of small farms, it will simply be superinsense to find out the cut of wear on the first hand. If not sorry, of course.
4. Conclusion. A few words about the results.
Separately focusing on: from above – not absolute numbers. They do not take into account many other factors and do not claim accuracy. By making such conclusions, we simply move from abstractions, such as: “Nuu, warm the video card” to empiric: “At least for a year I lasted in horse conditions”.
The conclusions from above (both strangers and mine) can be fair for any semiconductors: percent, memory, power transistors, t.e. for everything on Silicon structure. Only depending on the specifics of the chip, you need to consider a lot of additional conditions.
I want to note that the source plan of the article was 4 times longer. Let me show.
Formally good scientific sources in overheating were already 4 PDF-Nika. I read all of them, in some places has dripped a new terminology, excavated the standards of III … I left it for later. Occam principle. The material is so overloaded, 16,000 words are superfluous. Therefore – to the imminent meetings. See you on the expanses of iron sections Stopgame.ru.
The best comments
Wait, wait. You were mixed straight into a pile of horses)) I agree with everything, but here you need to separate flies from cutlets.
1) external causes of breakdown – yes, bad passivation, poor -quality heat sink. Dust in all cracks, drying thermal paste, cheap x5R dielectrics – and t.D. A system with negative feedback (by temperature) is brought to the edge, the connection becomes positive and it will already work itself to the coal. I left this for some future article, simply because to print so much-the buttons from the claws will fall out))
2) the internal causes of the departure of the mosfet – thermal or avalanche breakdown. But, again, for a freshly closed energy transistor, you need to pump horseback. Even in ancient IRFP was under Joule. Stories about how to blow up transistors on a board is a volume of material on another article. I’m not sure, even that relevant for sg. Then it smacks with some habrome))
3) Of course, horse energies may not be needed if Fet is already tortured. If the substrate has accumulated defects in the Chinese basement, the case remains small – to stick it into a dusty system on the floor and run RDR on maximums)) Actually, about this “thermal memory” material and turned out.
4) In addition, GDDR is the capacitors matrix. She needs to consider individual effects, such as reducing the charge time. Pulls already on the 4th article))
5) some more manufacturers – give Boundary Scan and the ability to climb through service ports in a flaw regimen. And in principle – if you know where to look at the board, you can draw runaway conclusions in particularly neglected cases. But this is already the 5th article.
Pity)) 16000 words – this is rather a understanding. Here the question already arises – is it necessary to all this in such a volume and immediately))
And in general, the instinct did not fail me: the day, consider, passed – there are no views, the blog will most likely remain in the boiler. This clearly indicates that the topic has not entered, you have to shift back to the games, and I will postpone the philosophy of iron))
Special thanks for such a detailed answer. You can find out where you work? Service officer, FAE, designer?))
With each article, everything is more interesting and interesting. Author, do not throw this business, you are very good to write – informative and not boring.
Therefore, I look forward to new materials)
I work a little in a small "computer" service in my free time from the rest of the work.
All this confusion is that the reader may have the impression that the degradation of Mosfetes is a serious problem. However, this is a fairly rare breakdown and is most often found in “errors” of design in economy class devices, where the components are simply selected “impending” to the requirements.
It would be necessary to expand the topic, but it is very extensive. Crystal degradation (including in mosfta) has been discussed for decades. The problem is that at least few people deny the presence of a phenomenon in themselves, but its real influence is in question. Practice is not in laboratory conditions, when individual transistors are tormented, but on lively, with full -fledged devices does not show anything at all. It seems that there are such breakdowns, but they are disappeared rare.
Stories about how to blow up transistors on a board is a volume of material on another article. I’m not sure, even that relevant for sg. Then it smacks with some habrome))
If with a demonstration, they will appreciate everywhere. Beautiful, after all. )
And as semiconductors, in general, they relate to temperatures?
I quickly answer the question: semiconductors as a whole are sources of these temperatures when an electric current passes through them. Internal resistance converts electrical energy to thermal, your cap.
In fact, the article is interesting and it was pleasant to recall some terms from radio engineering and electronics. It’s a pity I have already borne everything and I can’t calculate anything from the article.)
We must clarify that if someone had a video card and he thought that here it is, my dream fish! The "aged" mosfet is the cause of all troubles – then this is very hard. The death of the mosfet, as a rule, a consequence of another breakdown. The breakdown of the diode, the loss of capacitors’ capacity, critical overheating, in the end, from which the Mosfet will die immediately. Most likely, the reason will be comprehensive-high pulsations from PSU, increased heating-more tender components will suffer from this, and then violations in the work will affect and more serious in some ways. With the most important engineering transistor miracle – a processor, obvious degradation is also a very rare thing. The same famous terrible "dump", that’s exactly what the dump is. In the processor, the direct connection of the legs (balls) with the board is literally violated. At the same time, as a rule, the stone itself remains quite alive.
From my observations, although, for example, in my use all the time, only the same notorious GTX 580 died, but the service lies a whole box with “dumpy” vidyuhs. However, the living vidyuhs that were able to serve, with a tightly dusty radiators, with dead fans and dried thermal package – many times larger. For example, on the shelf lies quite a living NX8600GTS-T2D256EZ-HD. The joke of this dinosaur is that this is not the coldest vidyuha with passive cooling, which has been “raped” for years with toys in a stuffy body with a bunch of dust.I strongly suspect that the main factor here is in the initial defects of the crystal, strapping and their soldering. Well, incorrect operating conditions of the destruction accelerate the process. But this is sad. You can’t guess how much the video card will still work out. I had unsuccessful purchases and cards from under mining, but, again, these are isolated cases from a dozen or two cards and in my case this issue was resolved painlessly. But, the fact remains a fact, buying B | The card has a lottery. And not only miner – any cards. By the way, among the miner, the cards of the served. Sometimes even with completely new fans. So this is far from the worst option.One thing can be said for sure – hot cards will die much more often. Even if the chip itself is able to hold huge temperatures, then no one canceled the temperature expansion when heated. A large temperature amplitude to the benefit of devices clearly does not go.
Well, you can still study the ROHS directive. In particular, the application of the occasion. I strongly suspect that this was very serious about the durability of electronics.