Miniaturization and portability trends in combination with increasing performance are contributing to the well know problem of heat concentration and dissipation in modern electronic appliances. The answer from the electronics industry has so far mostly consisted in trying to do better what "we know how to do". The processor industry is moving to SOI to reduce the heat dissipation per transistor, while the power supply industry is trying to squeeze the last % point of efficiency out of their regulators. And the two together are working closer than ever to device efficient management schemes to consume as little power as possible. Clearly such measures are slowing down the speed of the 'temperature raise' without taming it.
In portable electronics the issue of power dissipation is compounded by the lack of good energy sources. Eventually fuel cells will become viable, charging will yield to fueling and energy availability will no longer be a problem in portable systems.
When that happens the heat will remain the minimum common denominator, the ultimate problem to solve. Unless we do something about it sooner, that is.
Limits of passive cooling
The vast majority of heat management systems rely today on passive methods of cooling, typically based on a bulky mass of heat conducting material shaped for maximum radiating surface (heatsink), attached to the heat source. The heatsink may be complemented as necessary by forced air circulation. In cases where space is at a premium heat pipes are utilized as means to transport the heat from the hot spot to a peripheral areas where heat can be more easily disposed off, for example utilizing existing metallic structures as heatsink. While heat pipes are today state of the art in modern notebook computers, such technology is less than desirable being based on encapsulated fluids that may leak and damage the electronics. The fundamental limitation of all passive cooling methods, including heat pipes, is that they rely on a negative temperature gradient to work. In other words the heat always has to flow from the higher temperature point to a lower temperature point. It follows that the device or load to be cooled will always be at higher temperature with respect to the heat sink and the ambient. With ambient temperature varying easily from 25 to 70C and silicon failure rates proportional to the square of the silicon junction temperature, passive cooling resembles more a torture chamber for silicon rather than real refrigeration.
Active cooling is a forced means of refrigeration in which heat can be made to flow from the lower to the higher temperature spot. This is obviously the principle on which common refrigeration is based. While active cooling overcomes the 'negative temperature gradient' barrier it pays a price for it in terms of additional heat generation. Can active cooling be the solution?
The theoretical limit for efficient heat transport is achieved by the reversible heat engine obeying to the Carnot cycle.
The transport of heat by a Carnot cycle is described by the equation:
Pcool = Pload* Tc/(Th-Tc) (1)
Pcool = Power expenditure to cool with Carnot engine (Watts)
Pload = Power dissipated by the load to be cooled (Watts)
Tc= Temperature of the cooled side (K)
Th= Temperature of the hot side (K)
Accordingly, in order to transport 100W of heat from a cold surface (27C) to a hot surface (say 300C), an expenditure of power is theoretically necessary in absence of mechanical friction and other irreversibilites amounting to:
Pcool = 100W*(27 + 273)/(300-27)=109W (2)
In thermodynamic terms, this transport can be looked at as a refrigeration process or a heat pump process.
This can be described as a refrigeration process with the "Coefficient Of Performance", COP, defined as the ratio of the work required to the energy transferred for cooling (COPC), equal to 109W/100W=1.09. Or it can be seen as a heating process. In this case the cost of cooling, or 109W is effectively 'free' heat and hence the effective "coefficient of performance" (COPH) is 209W/100W =2.09.
Moving from thermodynamic to electronic terminology, let's now assume that 100W is the power generated by a chip powered by a voltage regulator (whose efficiency is 100% for simplicity) and cooled by Carnot. We have:
Pload =100W (3)
h%= 100*Pload/(Pload + Pcool) =100/209=48% (5)
where h is the efficiency, or ratio between useful power and overall power expenditure. Table 1 illustrates the relationships between these parameters and Fig. 1 illustrates the elements at play and the power flow.
Table 1. Watts required to transport 100W of power in a Carnot cycle
Figure 1. Schematic diagram of a Carnot engine cooling a 100W load
Notice that h% can also be calculated as 1/(1+COPC), still 48% for Carnot.
Adding to this the inefficiency of the voltage regulators powering the load and the engine and mechanical frictions we can conclude that active cooling at best will yield overall efficiencies in the range of 40%.
Active cooling yes or not?
Can active cooling be viable at such levels of efficiency? Of course yes! Low efficiency is not a killer per se but only when it generates heat in the wrong places, namely at the junction of silicon transistors. Other than that efficiency or lack of it, inefficiency is quite cheap.
Watts are cheap: at 8c/kWh a 100W load consumes 0.8c/h. Depending on usage patterns a CPU may not work at full speed for more than a few hours a day, making the daily cost of such feature around a few cents per day (say 3 cents a day) and a few bucks a year (say 10$). This is not an unacceptable cost.
With respect to fuel cells a similar calculation can be done. It is obvious that active cooling technology would burn methane (or whatever fuel we will end up using) at twice the rate or more of conventional technology but then again is that really a problem? So on the longer horizon we would not discount active cooling technology in portable computing either, once fuel cells become a viable technology.
The hot plate
Now that we are moving heat from the 'cold' silicon junction to a 'hot' plate for heat disposal, we have indeed turned the industry on its head: the hottest place is now the heat radiator, hotter than ambient, and the coldest place is the silicon junction. Doesn't this feel right? Would you want it the other way around ever again? Is a hot plate --perhaps as hot as 300C or more - a problem? I don't think so. We deal routinely with hot surfaces at home (kitchen, light bulbs) and on the road (motorcycles tail pipe).
Active cooling implementations
Today examples of active cooling, like thermoelectric cooling based on a Peltier array, are found in satellite receivers where lowering the temperature of the LNA allows a lower noise figure, and in fiber optic network equipment where again precision temperature control is required.
With thermoelectric cooling a voltage is applied to a Ohmic junction of two different conducting (thermocouple) or semi-conducting (P and N types) materials, and the ensuing current flow results in absorption or release of energy (heat) at the junction as the electrons cross a corresponding 'uphill' or 'downhill' potential. The intensity of heat flow is proportional to the current and the process is reversible, namely a heat source at the junction will produce a corresponding current flow.
Figure 2. Illustration of Peltier effect with V*I=Pcool.
In Fig. 2 the mechanism of an electron acquiring energy in order to overcome the opposing electric field Ec -- and hence cooling the 'cold' plate -- in crossing the N=>P junction, as well as releasing energy to in presence of a favorable electric field Eh -- and hence heating the 'hot' plate- in crossing the P=>N junction as illustrated.
Being the heat flow proportional to the current it means that any current controller in the semiconductor manufacturer product portfolio can be easily adapted to control current and hence, via a thermistor, temperature in a Peltier array.
For example Class D power amplifiers normally used to drive audio amplifiers are being applied successfully to drive Peltier arrays.
A few adventurous souls have applied Peltier cooling to their CPU's and managed to over-clock their PC's sensibly thanks to the lowered temperature. That Peltier will become mainstream in mass-market applications is dubious because of the extremely low efficiency, in the range of 5% of Carnot. While we have made a case for cheap energy earlier, an expenditure of 2kW to cool a 100W load seems to be a bit too much.
Stirling refrigerators -and variations on the theme- are mechanical systems based on compression and expansion of an inert gas by a piston. These systems yield efficiency closest to the ideal Carnot cycle and are being seriously investigated for high end CPU applications like IA64.
Because of their complexity, mechanical nature with moving parts and cost it is unlikely that this will be the technology that will shrink heatsinks and displace heat pipes for high volumes applications either.
What we need is an active cooling technology that has the efficiency of the Stirling and the 'solid state electronics' makeup of Peltier.
Our intuition tells us that an elegant, solid state, electronically inherent solution to heat management must exist if nothing else to mitigate the embarrassing dependence of the high tech solid state electronics industry from Iron Age passive heatsink technology.
A few pioneering companies have already advanced claims to this extent but they still have little to show for besides papers.
What we need now is a mindset change for our industry: enough with single minded push of efficiency gains and on with creative ways to bring to the mainstream a viable active cooling technology.