Cooling electronic systems has always been an important design consideration. As electronic systems continue to shrink and become more powerful, heat dissipation isn't just important—it's a critical design issue. Today's higher levels of IC performance and integration, coupled with smaller packaging and tighter form factors, are producing more heat in a given area than ever.
Making matters worse, electronics systems are being pushed to take on appearances that are more aesthetic and blend in with their surroundings. Unfortunately, aesthetics often occurs at the expense of natural air intakes/exhausts, so many of these systems will contain one or more fans to provide adequate cooling—frequently at the expense of acoustics.
Employing fans is contrary to the whole concept of silent PCs, unobtrusive office or home equipment, and electrical-only systems. That's especially true today because people are obsessed with minimizing noise. Minimizing acoustic pollution isn't just becoming a social requirement, but environmental and legal policy too.
Europe takes the lead in this area, as it tends to have much more stringent acoustic requirements than the U.S. Of course, this forces American companies to define and design better, quieter products for European markets, ultimately benefiting consumers on a global scale. Award standards such as the European Blue Angel Award solidify the effort by setting defined noise targets.
Adding a fan: When adding a cooling fan, very carefully select your vendor. Fan noise is related to blade pitch, number of blades, fan size, airflow volume, and rotational speed. Other factors include the physical construction, bearing type (e.g., ball or hydrodynamic bearing), and the internal electronic commutation circuitry. Also, due to resonance with the enclosure, a fan fitted into the system will make more noise than if it was operated in free air.
To simulate free air, the fan is suspended from a 150-mm piece of string, and measurements are taken with a microphone situated 50 cm from the fan. This typically gives best-case acoustic data (Fig. 1). In contrast, when that same fan is placed into a notebook chassis, the noise increases by up to 10 dB simply because the fan is inside the enclosure (Fig. 2). Clearly, the fan is noisiest at high speeds. To minimize acoustic noise, only run the fan at higher speeds when necessary.
Intelligent fans or intelligent control? Rather than specifying a very complex fan with myriad features, the key to using fans intelligently is to specify the most basic fan and make it act intelligently by adding an intelligent fan controller. With this approach, any 2- or 3-wire fan from any vendor can be made intelligent, allowing easy and inexpensive multiple sourcing of fans.
Measuring system temperature: Base the amount of cooling on system temperature. Cooling a system that's not hot will waste power and cause unnecessary noise. If the system requires cooling, but it's not extremely hot, the amount of cooling can be reduced. Slowing down the fan will minimize the cooling effect, but save power and reduce system noise.
Some fans incorporate a thermistor to measure the temperature at the fan so its speed can be increased as the air around it heats up. Due to the thermal lag associated with heating the air, these fans tend to have slow responses to changes in system temperature. Also, the fan is often located remotely from the temperature hot spot, so it might not sufficiently cool the system even if it's adequately rated. In many applications, this type of fan isn't very useful.
The ultimate measurement in temperature is to directly measure the die temperature of the hottest ICs. This technique eliminates inaccuracies and thermal lags associated with reading a temperature through a heatsink. For a number of years, CPU manufacturers have integrated a thermal diode into their products to facilitate measuring the die temperature. More recently, this thermal diode has been incorporated into chip sets and graphics controllers.
Because the technique used to determine temperature only requires a semiconductor junction, a simple transistor such as a 2N3904 will suffice. It can easily be placed in remote system hotspots to measure the temperature. Multiple thermal sensors, carefully located in different thermal zones, permit truer system profiling—leading to a distributed, more intelligent cooling approach. By employing an ADM1023 thermal-diode-monitor (TDM) IC, users can measure temperature remotely to an accuracy of within 1°C (Fig. 3).
Temperature-based fan-speed control: Combining accurate monitoring of the die temperature with a cooling fan provides the basis for intelligent fan control. Reducing the fan speed as the temperature drops will minimize the system acoustic noise. This can be accomplished with an integrated temperature-monitor and fan-controller IC. The best solution is to use an automatic-fan-speed control loop or closed-loop fan control. This offers the big advantage of operating completely independent of software once the control loop is configured. In the un-fortunate event of some system failure (either hardware or software), thermal management will still work.
Two distinct techniques of controlling fan speed exist: linear control and pulse-width modulation (PWM). Each has its advantages and disadvantages.
Linear, or voltage control, varies the voltage applied to the fan to adjust its speed. A typical 12-V fan will have an effective operating-voltage range of about 7 to 12 V. The lowest voltage at which the fan will operate is 7 V, but this may be the lowest voltage at which the fan will run once already spinning. For a stalled fan, the actual starting voltage re-quired to overcome inertia may be higher. It can only be found through experimentation.
For a typical 5-V fan, the operating-voltage range shrinks to about 4 to 5 V. Some fans won't spin at all below 4 V, while they spin at almost full speed above 4 V. Therefore, there's little or no adjustment range when using linear control to drive a 5-V fan. Another distinct disadvantage of linear control is the physical space necessary for the gain-setting resistors and op amp required to drive the fan.
PWM control drives the fan by adjusting the duty cycle of an applied square wave. Thus, the full voltage swing is always applied to the fan—0 to 5 V for a 5-V fan, or 0 to 12 V for a 12-V fan. Most fans will run reliably with a 33% to 100% duty cycle, so the control range can be wider than with linear control. With PWM control, the fan drive circuitry is much simpler, requiring only a single FET. Additionally, switching the fan on and off reduces the average power consumption and improves the efficiency.
Automatic-fan-speed control: The automatic-fan-speed control function turns the fan on at a predefined temperature point and automatically adjusts its speed. Operation of the automatic-fan-speed control loop is illustrated in Figure 4. The loop is programmed by defining the temperature parameters TMIN and TRANGE.
TMIN is the temperature at which the fan switches on and runs at minimum speed with a 33% duty cycle. When the measured temperature rises above TMIN, the fan will switch on and run at full speed for 2 seconds before dropping back to minimum speed. This ensures the reliability of fan startup every time.
TRANGE defines the temperature range over which the fan automatically varies in speed. The temperature at which the fan will reach full speed is TMAX:T MAX = T MIN + T RANGE
Built into the loop, 5°C hysteresis prevents the fan from cycling on and off when the measured temperature is near TMIN.
The most tangible benefit of the control loop is that the fan operates at its optimum speed for any given temperature, greatly reducing current consumption and system acoustic noise. The acoustic behavior of the system im-proves with automatic-fan-speed control (Fig. 5). Here, with TMIN = 40°C and TRANGE = 40°C, the fan runs full-speed at 80°C. Notably, fan noise is well below the 36-dB target value for most of the temperature profile.
Filtered automatic-fan-speed control: The automatic-fan-speed control loop is easy to understand and configure. Yet, additional filtering is sometimes needed to enhance system acoustic behavior. For example, large temperature transients may occur when a program runs on a computer. The CPU temperature rises sharply and peaks while the program loads. Afterwards, it drops back to its stable state.
Clearly, the fan shouldn't speed up for a short time and then slow down again. That would create very unsettling dynamic acoustics for the user. Studies of the human ear reveal that the brain adapts to a constant noise or tone far more effectively than to a sound that's shifting in frequency or intensity. For instance, automobile alarms warble to catch your attention. Unfortunately, a cycling fan will seize your attention too.
Filtered automatic-fan-speed control mode effectively smooths the fan's response to changes in temperature. With this approach, the fan ramps up to its new speed instead of trying to jump there instantaneously. This filtering effect is much more pleasing acoustically, because the fan changes speed gracefully, eliminating cycling due to fast temperature transients.
Filtering is achieved by programming the ramp rate and the number of temperature updates per second. Both factors influence how fast the fan responds to changes in system temperature (Fig. 6).
Measuring fan speed: Unfortunately, PWM control corrupts the fan's tach-ometer (TACH) signal, increasing the difficulty of measuring fan speed. The ADM1030/31 chip, both an intelligent fan controller and a remote temperature (thermal-diode) monitor, overcomes this problem. (The ADM1030 is a single-channel device for controlling one fan and one remote temperature, while the dual-channel ADM1031 can control two fans and two remote temperatures at the same time.) It accurately measures the speed of both 2- and 3-wire fans being driven under PWM control. The controller chip integrates high-accuracy, 1°C temperature measurement with automatic-fan-speed control. Its PWM drive is suitable for use with almost any fan.
For both driving and measuring the speed of a 3-wire fan, the chip's external circuitry is very simple (Fig. 7). The interface to the fan consists of a single n-channel MOSFET. The PWM output controls fan speed by varying the duty cycle used to switch the FET on and off. The fan's TACH signal feeds the chip's TACH/AIN (analog input) pin. Fan speed is determined by measuring the period of the fan revolution. An on-chip oscillator is gated into an internal counter, which is initialized on the rising edge of one PWM output and counts for two TACH periods.
A slight modification to the drive circuitry lets the controller/monitor IC drive and accurately measure the speed of 2-wire fans (Fig. 8). A single bit in the chip's configuration register sets the device to measure the speed of a 2-wire or 3-wire fan (i.e., that bit defines whether fan input is digital or analog). The only additional components re-quired are a sensing resistor, RSENSE, and a 0.01-µF capacitor.
The ability to accurately measure the speed of a 2-wire fan opens up interesting new possibilities. In addition to detecting fan failure, the circuit can use rpm control to accurately set fan speed. So now, fans that don't have a TACH indicator (the third wire) can still run at exact speeds to counteract wide (±20%) manufacturing variations in fan speed.
The ADM1030/31 also provides intelligent fan-failure detection. If a fan fails, an interrupt is generated. When this happens, the chip automatically tries to restart the fan. If the fan fails to restart after five consecutive attempts, a FAN_FAULT output is asserted to indicate a catastrophic fan failure in the system. A second fan (on the ADM1031) will automatically spin at full speed to compensate for the loss of airflow from the failed fan. If the error condition subsides (like a dust blockage at the air inlet clears up) or the faulty fan is replaced, the second fan will automatically return to its normal speed, and the FAN_FAULT condition self-clears.
This level of built-in intelligence is a major improvement over past methods of fan-failure detection. Previously, the CPU would have had to read the measured fan speed from a register and decide in software how to react. This meant delays inherent with establishing communication over the systems management bus (SMBus), reading and interpreting the data, and writing to the SMBus device to spin the second fan faster. Now the fan itself becomes intelligent, able to operate independently of software control.
Control of rpm: Fan speed can typically have production tolerances of ±20%. If the design requires a 5-V fan that can push 4 cfm (cubic feet per minute) of air at 5000 rpm, a fan that nominally runs at 6250 rpm must be specified to ensure that all of the fans can deliver the required airflow. If these fans are run at 5 V, they will spin at 5000- to 7000-rpm speeds. Therefore, many of the fans will run too fast and make too much noise.
This is where rpm control becomes useful. It works by monitoring the TACH speed signal from a standard 3-wire fan. A typical fan usually produces 2 pulses per revolution, which gate an internal oscillator on the controller to provide an accurate measure of fan speed. With the ability to measure fan speed comes the ability to accurately control it.
The RPM mode operates by fine-tuning the PWM output to give exactly the necessary fan speed. A fan speed, say 5000 rpm, is programmed into the device, and the controller measures it. If the fan is running too slow, the PWM output duty cycle increases until the set fan speed is reached. Conversely, if the fan is spinning too fast, the PWM output duty cycle decreases until the set speed is achieved.
This mode accounts for fan-speed variations and guarantees that different fans with a wide tolerance distribution will all run at the same speed. As a result, they will provide exactly the same airflow and roughly the same level of noise. Moreover, fans tend to spin faster as they age. The RPM mode eliminates this problem by automatically reducing the PWM duty cycle as the fan ages, maintaining the set fan speed. This increases fan lifetime and mean time to failure (MTTF).
Putting it all together: The way to integrate intelligent fans into your system is depicted in Figure 9. The controller/monitor chip connects to whichever microcontroller is used in your application. It can be programmed over a dedicated 2-wire serial SMBus, via an I2C bus, or be "bit-banged" in software using two spare I/O pins. Simple, inexpensive 2- or 3-wire fans are deployed.
The flexibility of the chip means that the same circuit is implemented no matter the mode of operation. With the dual-channel ADM1031 device, you can even mix modes. For example, Fan 1 can be operated under automatic-fan-speed control, while Fan 2 is driven in filtered mode. This capability makes it easy to experiment with and reconfigure system thermal and acoustic designs.