As the amount of power dissipated by personal computer's CPU and other heat sources has increased, brushless DC fan control has progressed through several levels of improvement to keep temperatures within safe limits. Starting with simple two-wire fans with only power and ground wires, the three-wire fan added a tachometer (tach) signal for feedback, and the four-wire fan added an additional PWM signal for improved control.

At the same time, the temperature and control capability has improved from changes to both internal and external circuitry. In some cases, the control has addressed the increased noise that PC fans can generate, a problem that is getting more attention in environments such as densely occupied offices and quiet living rooms. In fact, several transitions have occurred in fan control to help PC designers and fan suppliers cope with PC heat and noise issues.

Sources of heat in PCs
In addition to the CPU, the memory modules, the memory controller, the graphics controller, hard drives and internal power supplies all create heat within the enclosure. Using recent multi-core designs, CPUs have had a temporary reduction in operating temperatures but CPU temperatures will inevitably increase in the future. In fact, the amount of heat inside the box is gradually and continually increasing from the combination of all heat sources.

With new applications such as high-end entertainment systems, the physical environment changes from the desktop to the living room and the tolerance for fan noise drops even more. Users have been increasingly sensitized to the amount of fan noise in an office or desktop computing environment but certainly don't want to hear fan noise from an entertainment system located across the room. However, delivering MPEG4 video to a display takes a lot of performance and subsequently generates a substantial amount of heat. PCs put into the entertainment center form factor present greater cooling challenges because both the boxes/enclosures and the fans are smaller. For engineers who have to address the temperature, cooling and motion control noise problems at the same time, the challenge becomes quite daunting.

Removing heat from PCs
The ability to remove heat by a fan and heatsink combination is determined by the mass of the heatsink and the thermal resistance from the case of the device out to the ambient. The rate of heat flow is proportional to the difference in temperature. Therefore, a larger difference in temperature, such as a cool inlet and a hot heatsink. provides the highest amount of heat flow. As the ambient temperature increases, thjs difference ("delta-T" or T), and therefore the effectiveness, both decrease.

The switching control of the PC's fan motor is fairly conventional pulse-width-modulation (PWM) technology for a single-phase, non-reversing motor. While the control of the motor is rather straightforward, the system aspects of cooling can be quite complex. In PC applications, the motor has a two-, three-, or four-wire connector from the processor. The fan connector provides both an interface and a barrier.

On the CPU side, the system looks at all the accessible temperatures and applies preset algorithms to minimize the temperatures with the cooling resources that are available. On the other side of the connector, the fan has typically been a dumb peripheral but recently has achieved greater involvement in solving both thermal and noise problems. The motor control aspect can be confined to the fan itself for certain tasks, but overall, the fan is a resource to the computing system that determines how fast the fan should be run based on measured temperatures.

Initially, two-wire fans with simple on-or-off control were used for cooling. One of the immediate improvements was the addition of a third wire to provide a tachometer or sense signal. In many systems the input was used merely to monitor the fan's rotation for safety purposes. It provided a rough indication of rotational speed, but did not provide improved performance. A fourth wire, a control signal, was added to provide improved performance and better control. The transition from three to four-wire fans has occurred within the last three to four years, driven primarily by acoustic requirements as opposed to improved cooling. With a four-wire fan, a constant voltage is always applied and the PWM input is used to control the speed of the motor. Figure 1 shows the four-wire fan connector and a typical PC fan.

Figure 1: PC fans have progressed from two wires (power and ground) to three wires, which add a tach signal, and to four wires with an added PWM signal. The Intel-defined four-wire fan identifies pin 3 as sense and pin 4 as control. (Fan picture courtesy Intel.)
(Click to enlarge image)

Today, all fans used on mother boards are three-wire or four-wire fans. However, in certain form factors, two-wire fans are still used for the lowest cost in applications such as Video Graphics Array (VGA) cards. Similar to the initial use of a two-wire fan in PCs, there is no speed control and the fan is on 100% of the time.

The system runs more efficiently with a variable-speed fan. Many techniques have been applied to achieve variable speed, such as varying the voltage to the fan. Simple voltage variation does not provide a very large dynamic range because starting the fan requires more than the minimum voltage. Once the fan is started, the voltage can be reduced. The fan may initially take 50 percent of the rated voltage, but then only 40 or perhaps 35 percent, instead of 20 percent.

Additionally, the response to voltage is not linear. An open-loop fan control does not perform well at low speeds. A 20 percent PWM command with an open-loop fan control will probably result in 35 to 40 percent RPM. However, this low-speed-range operation is required for minimum acoustic noise. Without the linear response, the entire low-speed performance of the fan (the ability to run slowly and maintain adequate cooling and control at all times) is less than optimum. Figure 2 shows an existing open-loop fan control.

Figure 2: Temperature data is used to control a PC fan motor with a 4-pin connector. In this approach, the motion control is implemented on the PC side of the interface. (Source: Intel)
(Click to enlarge image)

Closed-loop motion control
In any motion-control system, closing the loop guarantees a linear relationship between the commanded speed and the actual speed. For improved cooling in PCs, the newest technique includes closed-loop speed control with thermal temperature compensation, Figure 3. The control is implemented on the fan side of the connector, not the PC side.

Figure 3: Closed-loop motion control of a PC fan using both RPM and fan-inlet temperature for feedback.
(Click to enlarge image)

The penalty of a fan running faster than necessary is increased acoustic noise. With the current physical configurations of using control on the PC side of the fan connector, it is very difficult to sense the fan inlet temperature and have the fan speed control IC on the mother board take advantage of this temperature reading.

Another over-speed situation occurs at start-up. While the PC is booting up, by default, the fans operate at 100% (maximum acoustic noise) even though the cooling requirement is minimal. For safety purposes, without a signal to drive the PWM, it is commanded to go to 100 percent, in case its control circuitry never wakes up. Autonomous circuitry in the fan, not tied to the boot-up process, avoids the initial turn-on surge, power consumption and noise.

Using local intelligence, based on sensing the temperature at the inlet of the fan with a sensor that resides in the fan's hub, the RPM can be decreased when the temperature is low. By delegating this task to the fan, the system acoustic response is optimized. With the classical servo-system approach that exists in the closed-loop control, feedback and command, the loop parameters are adjusted for an intentionally slow response without incurring a significant penalty from a cooling perspective, since cooling operates at a rather slow time constant. This dual-parameter control, based on RPM and local temperature, provides safe operating temperature and reduced acoustic noise.

For computing systems with several fans, such as servers, fan control during boot-up has another implication. When a half dozen or even a dozen fans are turned on at the same time, and each takes from one to two amps to startup, a significant amount of power is required. A power supply with the ability to handle this concurrent startup is much larger and more expensive than it has to be for steady-state operation alone, since that inrush current level only occurs at startup.

With a minimum amount of heat to remove at start-up, server companies have requested that the rate of change of speed should be limited when a fan is started from a stop condition. Taking between three to five seconds to get to full-starting speed (100 percent) would be desirable; however, the system is not always capable of providing this level of control. Building the capability into the fan provides an improvement in overall system design. With the amount of control in a closed-loop fan motor control, this additional start-up ramping circuitry is minimal but provides a valuable benefit for server designers.

Future improvements for fans
The term smart fan is frequently used to describe a newer-generation fan, but the smart aspect really is more of a future possibility for fan control. What makes a fan smart is communication, not just the control technology. In addition to microcontrollers, the control technology can be accomplished by hard logic and even analog circuitry. Today, in a four-wire fan, two wires provide the communication interface. One wire is dedicated to telling the fan how fast to go, using a PWM fixed-frequency, variable duty-cycle signal.

Coming from the fan, there is another one-bit channel: the tach signal, with variable-frequency and fixed duty cycle. From a communication perspective, these are two very inefficient channels. They cannot tell temperature in the fan's location, determine the fan's health, or provide life-enhancing functions for the system to improve reliability.

Thermal management and fan operation in the future needs to be based on a communication backbone and have the capability to understand the cooling resources, temperature sensors and communicate between them. At the box/system level, this allows the fan-motor control to perform the most effective cooling, as well as acoustic management for the minimum amount of power consumption, by cooling the chassis based on the end-customer's requirements.

A single-wire communication channel which can carry any kind of data in either direction and which takes a minimum amount of overhead could easily replace today's two wires, and use protocols that are already in place on the mother board. With the introduction of the Simple Serial Transport (SST) bus in 2006, the architecture shown in Figure 4 is a good speculation of a potential future communications for fan motors in the PC.

Figure 4: A backbone structure for controlling temperature and acoustic noise in PCs.
(Click to enlarge image)

The backbone and network technology exists today. There are not a lot of PC OEMs currently using the backbone on the network but that could change in the future.

One final consideration for PC motion control is common to most motor control applications. In a server environment or any office environment that has several computers, the energy cost savings by running fans at the lowest speed will increasingly become more important. As entities, such as the U.S. Department of Energy, continue to put restrictions on energy consumption and ratings on acceptable energy consuming electronic products, the need for reduced fan operation for reduced acoustic noise will also provide an energy savings to users.