Today's designs require more in-depth analysis of the thermal budget and closed-loop temperature control is fast becoming an important feature of electronic systems. Failing to take these issues into account can result in more than burnt fingers, warns Uwe Kopp of DCS (Data Conversion Systems).
Two facts in building electronic systems force system designers to think more and more about the temperature budget. One chief driver is that cost and size of the PCB and enclosures have to be reduced resulting in less air flow to cool semiconductors which are both power hungry and at the same time sensitive. The other factor is the need for ever higher performance from processors, DSPs and microcontrollers.
Moore's law of doubling processor power is having the side effect that power dissipation is raising proportionally. These trends cause the temperature in enclosures to reach higher values affecting reliability and even functionality of the whole design. One key protection element is the temperature-sensing device.
The first step after the architecture of the functional design of boards and enclosures should be thermal modelling. The resulting colour plots show clearly where the bottlenecks in the design are and whether active temperature sensing and control are necessary.
Modelling will also confirm if simpler passive methods are sufficient. After identifying the critical areas in the application, a protection strategy has to be defined. There are various alternatives available to protect, enhance or ensure extended operation.
Up until a few years ago copper traces or metal blocks used as heat sinks were the means to protect heat-dissipating devices. This method is still fashionable if heat sink weight is no problem and especially where other techniques such as fan cooling or reduction of power supply voltage and / or processor speed are not viable. From a heat management point of view a lot can be learned from computer manufacturers.
Protection mechanisms that have been used in desktop and laptop computers have found their way into other industries from game manufacturers to automotive applications. The different ways of protecting an electronic system against overheating or operating it in its optimal environmental temperature are:
- Passive protection by convection
- On/off protection through thermostats
- Active temperature sensor protection
One important fact to note here is that silicon temperature sensors can work from -55°C to 150°C when housed in plastic packages. For more advanced applications, over 150°C and up to 180°C, the sensors can still be used in die form. For most electronic applications this method should be sufficient. Outside of this range, passive sensors like Negative Temperature Coefficient thermistors (NTCs) or thermocouples should be used.
POWER DE-RATING, FREE AIR CONVECTION AND HEAT-SINKING
The most economic way to protect electronic circuitry is good power de-rating, free air convection and heat-sinking. For uncomplicated systems using minimum power hungry devices where space and weight are not an issue this method works fine. To ensure the reliability levels guaranteed by the IC manufacturers are met in the customer's design, the junction temperature of the chips themselves must not exceed the values given in the datasheets. Normally these temperatures are around 150°C.
All power device datasheets also give formulas to calculate the size of a heat sink through power de-rating. The formula considers the thermal resistance of the different elements through which the heat dissipates, such as chip to package, package to heat sink and finally convection to air.
To optimise heat flow, PCBs should be mounted horizontally with the ICs which are dissipating the greatest heat mounted on the upper side edge. At a system, or sub-system, design level these same simple rules apply, the hot device should be mounted so its heat can rise and find a route out of the system.
Another simple and efficient method for protecting electronic devices from thermal runaway are silicon thermostats. They can be used to bring the power devices in question into standby mode, disable the power supply or start a fan. The LM26 and LM27 series of silicon temperature sensors are very cost-efficient implementations of this concept. They come in various trip point options from 23°C up to 145°C to ensure flexible usage in a wide variety of applications. Also the reset point, or hysteresis, is selectable to be 2°C or 10°C to take care of the different thermal lag times and thermal masses that are inherent in electronic systems.
The sensor is paired closely with the power device and the temperature can be accurately estimated. The temperature trip point has to be evaluated empirically. As a rule of thumb the air temperature at the sensor is about 15°C lower than on the package of the monitored device. This type of measurement can only give a rough estimate but will protect efficiently. Due to the inherent inaccuracy of this measurement method the sensor itself does not have to be very accurate. Temperature accuracies between +/-2°C and +/-4°C are sufficient.
DIGITAL TEMPERATURE SENSORS
Digital temperature sensors integrate a diode-sensing element, an Analogue to Digital Converter (ADC), registers and an interface block. They communicate with the microprocessor through digital interfaces such as SPI, SMBus or I2C. They are available with ADC resolutions from 8Bits to 16Bits. Higher resolution devices can be trimmed more accurately at the end of line by the manufacturer.
Accuracies for an 8Bit device are around +/-4°C and can be as good as 0.33°C at room temperature for 16Bit devices. The smaller the temperature range a sensor has to cover the better the accuracy for this window can be defined.
Accuracy is needed for various reasons. One such example is where system parameters change rapidly. The closer operation is to the limits without risking reliability the better. Assume that an MCU is used that operates up to 125°C before functioning abnormally. Using a temperature sensor with +/-4°C accuracy, an alarm to switch off the processor should be set at 117°C. A better sensor with +/-1°C accuracy gives headroom of up to 123°C.
The robust data output stream means sensors can be placed 20cm to 30cm from the MCU even in relatively noisy environments. No additional circuitry or pins on the controller are needed. Together with other digital interface devices the sensor can operate on the same bus. Analogue sensors on the other hand need a dedicated microcontroller with an integrated AD converter and would have to be placed very closely to the MCU as the analogue output signal that carries the temperature information is susceptible to noise.
Digital sensors such as the LM75B can handle up to 300mV noise on the bus without hanging up or delivering false data. Analogue sensors typically output 10mV/°C and consequently give erroneous data if 10mV of noise is introduced. Filtering can help, but it does not eliminate this inherent problem.
Even with a very accurate digital temperature sensor one never acquires the exact temperature of the chip being protected. Several manufacturers of 'hot' devices like FPGAs, MCUs, video processors or power FETs have implemented a feature used on Intel processors.
DIODE TEMPERATURE MONITORS
Through this built-in diode, the device temperature can be monitored very closely and accurately. To accommodate many different remote diode characteristics, the user can select diode non-ideality factors and can also compensate for trace resistance. The latter is important as a trace resistance of just 1Ω can introduce a temperature error of 1°C because these sensors measure microvolt signals.
Remote diode sensing chips are available with up to three remote inputs to measure different spots in a system. The interface to all of them is SMBus from the PC world. However this should not be a stumbling block to using them in a different application. SMBus is very similar in functionality to I2C and actually more robust because a timeout function is already implemented that prevents one user on the bus from locking the whole system.
Rather than stopping operation when the temperature rises, many of the remote sensing devices described so far allow the control of DC fans. Depending on the application these devices may well be sufficient to manage the thermal budget and are relatively easy to implement at the system level.
Most fan controllers are not simply on-off devices, but use Pulse Width Modulation (PWM) to control DC fan speed. Unfortunately many of the PWM outputs operate at frequencies below 20kHz resulting in audible noise when spinning up and down. To avoid noise audible by human ears the PWM switching has to take place either at very high or very low frequencies. To overcome this, devices implementing free programmable PWM switching frequencies at up to 180kHz in steps are available, such as National's remote diode sensor LM63. If resonance frequencies are encountered with either the enclosure, the fan itself or other system components is in this upper frequency band then the device also can be used with any PWM frequency between 22Hz and 703Hz to tune the noise out of the system.
In many systems the best acoustic fan noise performance is achieved when the PWM versus temperature transfer function curve is parabolic in shape. Moreover, the temperature sensor allows the user to program a transfer function individually into a lookup table to match system needs.