Automatic fan control: the trend of high-speed chip cooling technology

Abstract: Cooling fans are an important component in thermal management of high-power chips (such as CPUs, FPGAs, and GPUs) and systems. Unfortunately, they can sometimes introduce audible noise that is annoying to the user. By measuring the temperature and adjusting the fan speed accordingly, the fan speed (and noise level) can be minimized when temperatures are low, but increased to prevent chip damage in the worst case. This article discusses two techniques for automatically controlling the speed of cooling fans.

High-speed chips tend to get hotter. The faster they run, the hotter they get. New generations of high-speed digital chips use smaller process geometries that allow supply voltages to be reduced, which helps somewhat, but transistor counts are increasing faster than supply voltages can be reduced, so power levels are still rising.

When die temperatures rise, performance suffers. Parameters change, maximum operating frequencies drop, and timings are out of specification. From the user's perspective, when these things happen, the product no longer works properly. Therefore, the primary reason for cooling high-speed chips is to maintain good performance over the longest possible operating time and over the widest range of environmental conditions. The maximum allowable temperature of a high-speed chip while meeting parameter specifications depends on the process and chip design method (how "close to the edge" the chip is operated), as well as other factors. Typical maximum chip temperatures range from +90°C to +130°C.

Excessive chip temperatures can cause catastrophic damage to the chip when operated beyond the point where performance begins to degrade. Maximum chip temperatures are usually well above +120°C and are determined by factors such as process, packaging, and time spent at high temperatures. Therefore, high-speed chips need to be cooled to prevent them from reaching temperatures where performance degrades and permanent damage occurs. High

-speed chips rarely use a single cooling technique. In fact, a combination of techniques is typically required to ensure high performance and continued reliability. Heat sinks, heat pipes, fans, and clock throttling are the most common cooling methods for high-speed chips. The last two, fans and clock throttling, can help solve the thermal problem, but they introduce problems of their own.

Fans can significantly reduce the temperature of high-speed chips, but they can also generate a lot of audible noise. The noise of cooling fans running at full speed is annoying to many consumers and is becoming a concern for government agencies, as is the long-standing effect of noise in the workplace. Adjusting fan speed based on temperature can significantly reduce fan noise; fans can run slowly (quietly) when temperatures are low and speed up as temperatures rise.

Clock throttling—reducing clock speed to reduce power consumption—works by reducing system performance. When the clock is throttled, the system continues to operate, but at a reduced speed. Obviously, in high-performance systems, throttling should only be done when absolutely necessary, when the temperature reaches a point where it would stop operating.

Controlling fan speed or clock throttling based on temperature requires first measuring the temperature of the high-speed chip. This can be done by placing a temperature sensor close to the target chip—directly next to or sometimes underneath, or on a heat sink. The temperature measured in this way corresponds to the temperature of the high-speed chip, but it will be significantly lower (as much as 30°C), and the difference between the measured temperature and the chip temperature will increase as power consumption increases. Therefore, the temperature of the board or heat sink must be correlated with the temperature of the high-speed chip.

For many high-speed chips, there is a better solution. Many CPUs, graphics chips, FPGAs, and other high-speed ICs contain a "thermal diode," which is actually a bipolar transistor connected as a diode, located on the die. Connecting a remote diode temperature sensor to this thermal diode allows a very accurate direct measurement of the high-speed chip temperature. This not only circumvents the large temperature gradients encountered when measuring temperature outside the IC package, but also eliminates the problem of very long thermal time constants, increasing the response speed to changes in die temperature from seconds to minutes.

The need for fan control forces the designer to make several key choices. The first choice is the method of regulating the fan speed. The common method of regulating the speed of a brushless DC fan is to adjust the fan's supply voltage. This method works well with supply voltages as low as 40% of the rated value. But there is a drawback. If a linear regulator is used to change the supply voltage, the efficiency is very low. Using a switching power supply can achieve better efficiency, but it increases cost and component count.

Another popular fan speed control technique is to drive the fan with a low-frequency PWM signal, typically in the 30Hz range, and adjust the fan speed by adjusting the duty cycle of the signal. This solution is inexpensive because only a single small switch is used. Since the transistor is used as the switch, the efficiency is very high. However, the disadvantage of this method is that the fan can be somewhat noisy, which is caused by the pulsed power supply. The fast edges of the PWM waveform cause the mechanical structure of the fan to move (somewhat like a poorly designed speaker), thus generating audible noise.

Another design choice for fan control is whether to measure fan speed as part of the control strategy. In addition to power and ground, many fans have a third wire that provides a "tachometer" signal to the fan control circuit. The tachometer output produces a specific number of pulses (e.g., two pulses) for each revolution of the fan. Some fan control circuits use this tachometer signal as feedback to adjust the fan voltage or PWM duty cycle to achieve the desired RPM. A simpler approach is to ignore any tachometer signal and only adjust the fan drive to speed up or slow down. This approach provides less precise speed control, but is less expensive and at least eliminates one feedback loop, simplifying the control system.

In some systems, it is also important to limit the rate of change of the fan speed. This is especially important when the system is in close proximity to the user. In some cases, it is acceptable to simply turn the fan on and off or to change the speed immediately when the temperature changes. However, when the user is nearby, the sudden change in fan noise can be particularly noticeable and annoying. Limiting the rate of change of the fan drive signal to a certain limit (e.g., 1% per second) will minimize the audible effect of fan control. The fan speed still changes, but it is not particularly noticeable.

Another important design factor is the fan control strategy. Typically, the fan is turned off below a certain temperature threshold and starts to spin at a low speed (e.g., 40% of full speed) above the threshold. As the temperature rises, the fan drive increases linearly with temperature, up to 100% drive. The optimal slope depends on the system requirements. A larger slope will give a more stable chip temperature to some extent, but the fan speed will vary more when the power consumption varies over time. If the goal is high performance, the starting temperature and slope should be chosen so that the fan reaches full speed before the chip temperature is high enough to start clock throttling.

Fan control circuits can be implemented in a variety of ways. A variety of remote temperature sensors with up to five measurement channels can directly sense the temperature of high-speed chips and transmit temperature data to a microprocessor. Fan speed regulators with multiple fan tachometer monitoring channels can provide reliable control of fan RPM or supply voltage and can accept commands from an external microcontroller. To reduce cost and simplify design, ICs that include temperature measurement and automatic fan control in a single package are available. The sensor/controller also typically includes overtemperature detection, which can be used for clock throttling or system shutdown, thus protecting the high-speed chip from catastrophic damage due to overheating.

Two examples of such ICs, one with a DC drive and the other with a PWM drive, are shown in Figures 1 and 2. The IC in Figure 1 remotely senses temperature and controls fan speed based on temperature. The chip generates a DC supply voltage through an internal power transistor. The IC in Figure 2 has similar functionality, but drives the fan with a PWM waveform through an external transistor. Both have complete thermal fault monitoring and overtemperature outputs that can be used to shut down the system if the high-speed chip gets too hot.


Figure 1. Linear (DC output) temperature sensor and automatic fan speed controller. Automatically controls fan speed based on the temperature of the high-speed chip. Tachometer feedback from the fan allows the fan controller to directly adjust the fan speed. The system shutdown output prevents the high-speed chip from reaching catastrophic temperatures.


Figure 2. Temperature sensor and automatic fan speed controller with PWM output. Fan speed is automatically controlled based on temperature. Clock throttling and system shutdown outputs prevent high-speed chips from reaching destructive temperatures. Connecting the CRIT0 and CRIT1 pins to power or ground selects a default shutdown temperature threshold, ensuring safety even when system software is suspended.