Sorry about that reply to the guy being helpful and suggesting an upgrade. It just wasn't neccesary.
About the reliability part:
just brew it! wrote:Orwell wrote:It's a fact that integrated circuits are only designed to run for a couple of years at their specified frequency, voltage and temperature range. This is becoming more and more of a problem the smaller transistors get. The user should expect their circuits to live shorter than this expected lifespan.
Aside from (possibly) high-end GPUs that are being used for crypto-currency mining or other heavy GPU compute applications, I call BS on this.
I have never had a CPU die of old age. Ever. And that includes a few that were subjected to some fairly serious overheating (e.g. a Phenom 9550 where a dead CPU fan went unnoticed, causing the core temp to rise to 100C for an extended period of time... that CPU never skipped a beat, and is *still* in service to this day in a home server, half a decade after its little "incident"). I also own multiple AMD Phenom II and FX CPUs which were used to run Folding@home -- so 100% load 24x7, with fan profiles tweaked to run them within a couple of degrees of their maximim specified temperature (to keep noise levels down).
I understand that CPUs are necessarily getting a bit more delicate as the process sizes shrink, but we're still nowhere near the point yet where they are "only designed to run for a couple of years". The primary buyers of CPU chips (corporate and datacenter customers) would revolt, as this is shorter than the typical enterprise desktop/server refresh cycle. I would be very surprised if current CPUs that are run completely within spec have an average lifetime of less than 5 years, and expect most would still make it to 10.
I agree with you that a high speed IC will have a lifetime of about ten years or so before it will start to fail to deliver its specified frequency. I shouldn't have used the "couple of" piece of text in my reply, because it applies to overclocked ICs. Anyway, I'll give everyone over here a mini course on CMOS reliability just for laughs.
First off, I am basing all of this information on what I am learning during my master's degree in microelectronics. I am by no means an expert in this field (yet, heh). If you want definitive answers, ask my prof.
These are the mayor factors that contribute to life expectancy of a CMOS integrated circuit and have to be simulated and accounted for during IC design:
- Dielectric Breakdown. Unfortunately, as MOSFETs scale down, the electric field (volts per unit distance) applied to specific parts of transistors tends to increase ("constant voltage scaling") instead of stay the same ("constant field scaling"). For example, the dielectric in a planar CMOS transistor is currently only a couple of atom layers thick at 45nm and below (that's why scaling below 30nm without FinFET is nearly impossible, you can't insulate stuff with even less atom layers). At the same time, a voltage difference of over a volt is commonly applied to that layer. This translates to an electric field of billions of volts per meter. This has an added effect of quantum tunneling. Simply put, the super strong electric field forces electrons into that insulating layer (yes, conducting insulators yay), sometimes trapping said electrons in the insulator, reducing insulative capacity and thereby changing the switching characteristics of the transistor, and finally short circuiting the gate to the channel. This is basically what causes flash chips to wear out due to writing. This effect is called "Time-dependent gate oxide breakdown".
- A different effect also caused by huge electric fields (lots of volts over tiny distances), is hot carrier damage. In essence, semiconductor conductance and electron velocity is modeled by calculating how often electrons collide with the base semiconductor material as they are pulled through said material due to an applied voltage. In some cases ("velocity saturation") and beyond, even stronger electric fields are produced in corners of the MOSFET structure ("pinch off"). This causes electrons to bump into the semiconductor structure so forcefully that it creates defects in said material and will create new charge carriers (by ionization), altering conductivity.
- Electromigration. This is related to the item above, but instead of breaking atoms, the electron flow will start to push the semiconductor or metal material it is flowing through around (kinetic energy transfer).
Most proper papers on this subject are behind a paywall over at IEEE Xplore (seriously, **** that), but here is a free one that is okay enough in my opinion:
https://www.mosis.com/files/faqs/tech_cmos_rel.pdfAnd here are some good books on the subject:
http://www.amazon.com/Essentials-Electr ... 0792379918 (semiconductor reliability)
http://www.amazon.com/Semiconductor-Phy ... 0073529583 (semiconductor physics, $187 lol)
Most weardown equations, or even most semiconductor physics equations depend strongly on temperature, current density (amperes per square meter), and electric fields (volts per meter). These are the things you really want to limit! And these are exactly the things that keep increasing when IC's are scaled down and/or overclocked. This also means that even if you keep the temperature low, a high voltage per meter (and consequently a higher current per square meter) will still slowly kill your chip.
Fun fact: a semiconductor like Si will perform better (speaking about conductivity) than at room temperature (300K) when heated to about 1000K and beyond. There's also a somewhat less pronounced optimum regarding said property at roughly 120K which is what extreme overclockers aim for. Better add some negative cooling in your next build! Here's a graph showing some properties of Si as a function of temperature. Sigma (the conductivity) needs to be as high as possible:
http://i.imgur.com/tw6aYHm.jpgAnyway, during IC design, one is expected to simulate and account for these kinds of problems. I haven't performed said task for fancy pantsy 32nm stuff (90nm UMC only for students
), but research has shown that a high speed IC at 45nm or 32nm
has a lifespan of about ten or maaybe fifteen years or so at the specified frequency, voltage and temperature. Semiconductor foundries will also report their findings on the durability of designs that are sent to them (in other words, AMD and Intel know!). And this is why you do not want to use this technology in cars or airplanes which should last longer. Either use more robust tech or don't push your designs too much (reduce the three parameters listed above when possible).
Of course, saying that an IC produced twenty years or so ago still works at spec is an invalid point because it is based on older more robust tech.
Aand, I've run out of time to type more.
2600K @ 4400MHz/1.3V | MSI GTX780Ti | Asus Maximus GENE IV | Samsung 850 500GB | Dell U2713HM | Boston Acoustics A26