Printable View
Hey we all know that computers run better when cool, but what is happening to the hardware when hot that causes instabilities?
I realize that is you get too hot you simply burn up the transistors...I guess I mean warm versus cool, not blazing.
------------------
[SGC Moderate Member]
Higher temperatures increase the resistance of metals. This increases the RC time constant of circuits, decreasing the speed at witch they can function.
As heat goes up in a resistor, the resistance value also goes up, thus raising the amount of power dissapaited over it, thus heating up the resistor even more....etc. When the resistance goes up too far, very little current gets passed through, and the voltage dropped over the resistor goes up too, meaning that less current and less voltage is available to components further into the circuit. This causes the transistors to behave in unpredictable ways, since the actual voltage or current into it is either not known to any degree of accuracy, or the voltage or current is insufficient to "turn on" the transistor. When things like this happen, bad things aren't too far behind.Quote:
Originally posted by hobbes2112:
Hey we all know that computers run better when cool, but what is happening to the hardware when hot that causes instabilities?
I realize that is you get too hot you simply burn up the transistors...I guess I mean warm versus cool, not blazing.
------------------
Raven667
=================================
Windows 2000 Professional
AMD K6/2-400 @ 450MHz
FIC VA-503+ mobo
2 x 128 MB PC133 RAM
30 GB 5400rpm Fugitsu HD
3Dfx Voodoo 3 2000 AGP @ 175MHz(main vid card)
TVM AS-5S 15" Primary Moniter
SiS 6326 PCI (secondary vid card)
PowerMate/MultiSync II 14" Secondary Moniter (old 386 POS)
D-Link 10/BaseT NIC ISA
Soundblaster 16 ISA
=================================
????? how much metal do you think is actually in your processor?? In reality it is mostly silicon (doped and non-doped), with very small traces (granted, in a metal, but see previous post for the real problemQuote:
Originally posted by Old Fogy:
Higher temperatures increase the resistance of metals. This increases the RC time constant of circuits, decreasing the speed at witch they can function.
------------------
Raven667
=================================
Windows 2000 Professional
AMD K6/2-400 @ 450MHz
FIC VA-503+ mobo
2 x 128 MB PC133 RAM
30 GB 5400rpm Fugitsu HD
3Dfx Voodoo 3 2000 AGP @ 175MHz(main vid card)
TVM AS-5S 15" Primary Moniter
SiS 6326 PCI (secondary vid card)
PowerMate/MultiSync II 14" Secondary Moniter (old 386 POS)
D-Link 10/BaseT NIC ISA
Soundblaster 16 ISA
=================================
Two points ...Quote:
Originally posted by Raven667:
????? how much metal do you think is actually in your processor?? In reality it is mostly silicon (doped and non-doped), with very small traces (granted, in a metal, but see previous post for the real problem
1) Actually, there is rather a lot of metal in your typical IC - about 5 or 6 layers of it!! Metal interconnect is now the dominant factor when accounting for delays in VLSI chips.
2) How is you explanation any different from Old Fogy's ? You both state that resistance goes up with temperature. How can one be the "real" reason, and the other not - they are both the same!
Also, I am draging stuff up from my memory that hasn't seen the light of day for about 10 years - but doesn't majority charge mobility fit in here somewhere ? Or is it minority charge lifetime ? I thought there were some transistor parameters that were severly affected by heat - but then, maybe that's bi-polars I am thinking about, not FETs. Hmmmmm .... I feel some investigation coming on ......
Quote:
Originally posted by SlartyB:
Two points ...
1) Actually, there is rather a lot of metal in your typical IC - about 5 or 6 layers of it!! Metal interconnect is now the dominant factor when accounting for delays in VLSI chips.
2) How is you explanation any different from Old Fogy's ? You both state that resistance goes up with temperature. How can one be the "real" reason, and the other not - they are both the same!
Also, I am draging stuff up from my memory that hasn't seen the light of day for about 10 years - but doesn't majority charge mobility fit in here somewhere ? Or is it minority charge lifetime ? I thought there were some transistor parameters that were severly affected by heat - but then, maybe that's bi-polars I am thinking about, not FETs. Hmmmmm .... I feel some investigation coming on ......
I think this is closer to the truth , but like you I am having a lot of trouble summoning up details. BJT's are very heat sensitive, but that isn't a problem here. I don't think the temperatures differences we are talking about would have a huge impact on wire resistance, so I think it is related to charge mobility. I just can't remember the exact mechanism.
------------------
Life is complex. It has real and imaginary parts.
I just looked it up - The channel current Ids = k.(Vgs - Vt)^2, but k is temperature dependant. This means that despite an increase in carrier mobility due to the increase in temperature, the channel current Ids actually *DECREASES* as temperature goes up. In a CMOS circuit, this will have the effect of slowing down transitions because the capacitance of the interconnect will not charge or discharge as quickly.Quote:
Originally posted by Moridin:
I think this is closer to the truth , but like you I am having a lot of trouble summoning up details. BJT's are very heat sensitive, but that isn't a problem here. I don't think the temperatures differences we are talking about would have a huge impact on wire resistance, so I think it is related to charge mobility. I just can't remember the exact mechanism.
The beneficial effect of all this is that FET's do not suffer from thermal runaway like bi-polar transistors do. Increasing temperature reduces the current flow through the transistor, which in turn decreases the power disipation - unlike bi-polars, whose operation relies on charge mobility and wll increase their current - and power dissipation - until they destroy themselves.
However, overclocking increases power disipation because the rate at which the circuitry switches increases - which is why you should *always* cool your parts if you are going to overclock.
Oh - and one more effect that is not really on-topic, but is related. One way you can speed up the transistors is to run them at a higher voltage. This not only increases speed, but also power dissipation, but it also has a much more damaging long-term effect : migration. Yes, believe it or not, the very atoms that your chip is made of do not stay in one place, they wander off. The rate at which they wander increases with temperature and applied voltage (due to the intense electric fields). This can reduce the life of an IC to just a few years - though I guess by then it would be obsolete anyway.
Sorry for rambling on a bit there ... but I thought it was relavant. https://www.sharkyforums.com/images/.../2005/06/6.gif
Ahhh, there we go...
I already had know the bit about simple resistance, but that didn't seem like the whole answer.
Quote:
Originally posted by SlartyB:
I just looked it up - The channel current Ids = k.(Vgs - Vt)^2, but k is temperature dependant. This means that despite an increase in carrier mobility due to the increase in temperature, the channel current Ids actually *DECREASES* as temperature goes up. In a CMOS circuit, this will have the effect of slowing down transitions because the capacitance of the interconnect will not charge or discharge as quickly.
The beneficial effect of all this is that FET's do not suffer from thermal runaway like bi-polar transistors do. Increasing temperature reduces the current flow through the transistor, which in turn decreases the power disipation - unlike bi-polars, whose operation relies on charge mobility and wll increase their current - and power dissipation - until they destroy themselves.
However, overclocking increases power disipation because the rate at which the circuitry switches increases - which is why you should *always* cool your parts if you are going to overclock.
Oh - and one more effect that is not really on-topic, but is related. One way you can speed up the transistors is to run them at a higher voltage. This not only increases speed, but also power dissipation, but it also has a much more damaging long-term effect : migration. Yes, believe it or not, the very atoms that your chip is made of do not stay in one place, they wander off. The rate at which they wander increases with temperature and applied voltage (due to the intense electric fields). This can reduce the life of an IC to just a few years - though I guess by then it would be obsolete anyway.
Sorry for rambling on a bit there ... but I thought it was relavant. https://www.sharkyforums.com/images/.../2005/06/6.gif
------------------
[SGC Moderate Member]
The primary reason heat is bad is it effects chip (primarily metallization) reliability. The metal reliablity is decreased due to a phenomenon known as electromigration. This occurs in high current density metal lines and sometimes in contacts or vias. Electromigration causes short and opens - neither of which is good! Electromigration is highly temperature acclerated. This is another reason Copper metallization is superior to Alumninum. Copper has much better electromigration resistance.
Temperature can also accelerate wearout of the gate oxide - another very bad thing.
Heat also increases resistance and reduces the channel mobility. These definitely slow the chip down, but usually chips are designed to 'worst case' operating points so timing can be maintained even at high junction temps (within reason). Usually max Tj is 90C, 120C or 140C (depending on the chip designer and process groundrules). Leakages also increase at high temp, this can cause big problem for DRAM and SRAM. Rule of thumb is leakage doubles every 10degC.
I once worked on a chip project that ran at liquid N2 temp (77K) - that thing flew, but was rather impractical.
Someone mentioned the amount of metal on an IC. There is actually quite a bit. It is not unusual to have >10meters of metal on a chip that is about the size of your thumbnail.
Cheers.