Recently, a colleague told me he had turned off his desktop PC for the first time in two years and couldn’t believe how quiet his office became. So I turned off mine too, and I was amazed at the contrasting silence after the rotating media and cooling fan spun down. The calming quiet was striking. So, even though we grow accustomed to the constant noise, it is still distracting! How can we reduce the noise? There are many challenges and tradeoffs.

Fans play a vital role in cooling the components that give PCs their performance, so their noise cannot be reduced at the expense of critical components overheating. Adding to this concern is the challenge to provide even higher-cooling capacity to deal with increasing power dissipation from new processors, graphics and memory products. So, if the PC cooling system is to undergo changes in reducing fan noise it must be done carefully so as not to throw the cooling out with the noise.

The fan noise challenge begins when engineers tasked with designing a PC cooling system are faced with the always-considered worst-case conditions. To provide adequate cooling under these conditions requires choosing a fan with cooling capacity above what is needed for normal conditions. So, when this fan is supplied the rated voltage, there is an assurance that the system temperature will not get too high, even under those worst-case conditions. But if only a single voltage is supplied the fan runs continuously at full speed all the time, yielding the highest cooling capacity, but along with it the highest level of noise and current consumption. So, maybe we should slow things down a bit . . .

Yes, one way to reduce fan noise is to slow the fan down. If fan speed (RPM) could be controlled in such a way that an adequate level of cooling is provided, even as conditions vary, it may be possible to minimize acoustic noise by slowing down RPM when a lower cooling level is OK: And increase RPM and cooling capacity only when needed -- i.e. worst-case conditions. That sounds simple enough, but what can the fan’s speed be based on? What other constraints or requirements exist? Let’s discuss the increasing challenges for defining a PC cooling system.

PC Health Monitoring, WfM, ACPI and TCO impact on Fan RPM control

There are many well-established initiatives to implement provisions for monitoring the "health" of critical hardware inside PCs through the use of mixed-signal ICs that communicate back to the system. For the cooling fan(s), this requires a speed signal (tachometer) with which to monitor fan RPM, so it can be compared against limits and an alert issued if limits are exceeded. A fan alert could be used to warn the system of an impending thermal problem due to the fan’s failing health so an application could save a user’s work. Or, under WfM, an IS manager could be alerted and preventive maintenance performed before more serious problems occur. Since the desire is for the fan to normally run at lower speed to reduce noise and the RPM to be variable as system conditions change, this speed signal must be available throughout the RPM range to provide the health monitoring feature. (Need #1: Tachometer signal over RPM range)

Another issue is how the user perceives health of the PC based on what is heard. For example, hearing the fan turn on and off typically alarms the user, who perceives it as a problem. The same is true if a small number of coarse, discrete, RPM steps are used. To avoid this, the fan speed control must provide barely perceptible, fine enough, changes in speed over a long enough time. (Need #2: RPM control must provide gradual changes slowly over time)

To accommodate the needs to 1) monitor fan health, 2) adjust fan RPM slowly over time, 3) adjust fan cooling capacity based on some parameter within the PC, and 4) provide OEMs with flexibility to tailor fan speed adjustment to PC design goals, etc., there must be the means to control fan RPM from the system. Since critical components in the PC, such as the CPU, are subject to malfunction if they become too hot, it would make sense to monitor their die temperature directly and adjust the fan speed accordingly. With the advent of new ICs that can accurately monitor and report CPU die temperature back to the system as a digital word, there is now a very good parameter available to control fan RPM. There is also the possibility of throttling back the CPU’s clock frequency should it become too hot. So, with ACPI-compatible BIOS there exists a capability for the system to balance fan speed/noise against CPU performance. The concept is shown in Fig. 1.

One might ask, "If the CPU gets too hot, which choice will the system make — decrease CPU clock or increase fan speed?" OEMs can implement that decision process via definition of an algorithm, or there could be provisions for the PC user to choose the priority of low noise or high performance via software. The good news is that implementation of the majority of this concept has already been demonstrated. (Need #3: Fan RPM must be controllable by system)

Another issue is the amount of power required to implement the fan RPM control. If the method used generates heat itself, current drawn from the power supply will be higher and, in the extreme, the fan will run harder to extract the additional heat (a paradoxical situation.) With the goal for TCO of reducing PC power consumption wherever possible, it is desirable for the RPM control system to be as efficient as possible. (Need #4: Fan RPM control method should be efficient)

And finally, since the price of PCs is continually falling, the cost of implementing RPM control is of major concern. Whatever method is used must be of lowest cost possible. (Need #5: Low cost)

Now that some of the major criteria for PC RPM control have been noted, the methods that have been implemented to date can be assessed.

Fan RPM Control Methods

To date two methods have emerged: 1) linear voltage regulation, and 2) PWM drive. There are advantages and disadvantages of both methods to consider.

Perhaps the simplest way to control fan RPM is to vary the dc voltage applied to the fan. If its speed was rated at 12 V, then applying only 8 V lowers the speed somewhat proportionally. But then there’s the matter of regulating the fan voltage down from the power supply voltage, and the power dissipation and resulting heat generated by a linear regulator.

A linear regulation scheme requires a control signal from an IC interfaced with the PC system, which will be true for any fan RPM control scheme implemented. However, with the linear scheme, discrete components used to regulate the voltage will likely take up space on the motherboard, and the power device will dissipate heat. Additionally, 2-phase brushless motors (BLM) used in the majority of fans require internal circuitry to electronically commutate drive to the motor phases. This circuitry has some minimum voltage requirement, which may limit how low the fan RPM can go before the circuitry begins to malfunction. The cost of the required discrete components and the associated space taken on the motherboard, heat generated from the pass device and a limited RPM range makes this method less than optimal. For systems where the additional heat generated is tolerable the method is certainly viable, but is counter to TCO goals.

A second method, believed to provide higher-power-efficiency fan drive and lower cost for a speed-control system, is pulse-width modulation (PWM), an example of which is shown in Fig. 2.

The PWM technique of varying the on/off duty-cycle of the supply voltage, via a drive transistor to control effective power delivered to the fan motor, typically provides high efficiency and has been used extensively with self-commutating dc brush motors. However, PWM drive has definite drawbacks with two-phase brushless motors, since they require internal circuitry to commutate driving the phases and also to generate the tachometer speed signal needed for fan health monitoring. When the PWM drive is in the OFF state, power to the fan’s internal circuitry is also OFF. Thus, both the commutation and the tachometer signals are disrupted. There are some commercially available IC devices touting high efficiency that implement PWM drive. However, lab evaluations reveal lower overall fan drive system efficiency than the linear method.

It was discovered that the drive method and PWM frequency implemented by these ICs caused heat generation in the fan phase windings. A mechanical clattering noise was also observed from the fan at lower RPM, which would undoubtedly shorten the fan’s life expectancy. Higher efficiencies and elimination of the clattering noise have been demonstrated by a higher PWM frequency and the addition of discrete components, but the tachometer signal is still a problem. This is disappointing, since generating a PWM control signal can be accomplished very cost-effectively and easily in digital circuitry.

So, the PWM method has not panned out to be as usable as was once believed. It can be used in systems that do not require fan health monitoring, or if it is acceptable to measure fan RPM with the PWM duty-cycle at 100 percent. Yet, if running the fan at lower RPM to reduce noise also eliminates the use of the tachometer signal for fan health monitoring, the goal of early warning for an impending fan problem is defeated. The PWM method also requires some type of drive circuitry, which likely will take up space on the motherboard. Thus, this solution is certainly not optimal, by any means.

In conclusion, the need still exists for an optimally-defined RPM control system for PC cooling fans. Hooks into the PC system via ACPI-compatible OS and BIOS have already been established with which CPU temperature can be measured, RPM control signals generated and fan health monitored using specialized mixed-signal ICs. But, a solid solution for driving the fan using the RPM control signal, which also satisfies all the criteria noted earlier, has not yet been demonstrated. I’ve been trying desperately to define one, but have been having trouble concentrating due to the noise in my office . . .