Intelligent thermal management using mixed-signal FPGAs

Traditionally, thermistors, thermocouples or discrete temperature measurement integrated circuits (ICs) have been used to measure system temperature. However, as today’s system speed increases and the relative size of systems decreases, temperature management has become more important. If the temperature needs to be measured in multiple locations on the board, the cost of these devices can quickly add up. This, in turn, has created a pressing need to develop effective, compact and inexpensive methods of thermal management for applications ranging from high-speed computer systems, telecom network switching boxes and industrial temperature control to portable electronics, biomedical devices, motor control and automotive.

Because promptly and accurately correcting temperature is critical for many applications, today’s intelligent systems have active cooling systems and can often load balance their operation depending upon internal system conditions. Another advantage of these systems is the ability to track and measure specific device temperatures using on-board temperature diodes, or diode-connected transistors, giving an indication of system heath as an unexpected temperature rise can highlight that the component is not operating correctly. In response, the intelligent system could take corrective action and/or alert system management of the out-of-bounds condition. Among other system management tasks, today’s mixed-signal field-programmable gate arrays (FPGAs), representing one type of intelligent thermal management system, enable design engineers to accurately measure temperatures at a number of locations cheaply and easily.

EXAMINING VOLTAGE

When studying the relationship between the absolute temperature of a diode and its forward voltage when under a constant current, the forward voltage of the diode will drop approximately 2mV/C. To increase the accuracy of this measurement and eliminate diode-to-diode variances, utilize two known currents and the ratio of these measurements. Figure 1 illustrates how the diode voltage and currents are affected by temperature.

The equation for this measurement is shown below.

Equation 1 T = V * q / (n * k * ln(IH / IL)

Where:

T = Temperature in Kelvin

V = Difference in voltage across diode at high and low current

q = 1.602x10-19 Coulombs (charge of an electron)

n = 1 (ideality factor, assumed to be 1)

k = 1.38x10-23 J/K (Boltzman’s constant)

IH = High current

IL = Low current

To demonstrate, utilize Actel’s mixed-signal fusion programmable system chip (PSC) in a real-world situation. Mixed-signal FPGAs will send two known current sources, 100µA and 10µA (see Figure 2), and measure the voltage difference using its integrated analog-to-digital converter (ADC). Assume the diode is at room temperature and examine the impact on the circuit.

Equation 2 below solves Equation 1 for the voltage to be sent to the ADC, illustrating the voltages measured by the mixed-signal FPGA.

Equation 2 V = T * n * k * ln(IH / IL) / q



V = 298 * 1 * (1.38x10-23 J/K) * ln(10) / (1.602x10-19C)

V = 298 * 0.00019835 = 59 mV

At room temperature, this represents relatively small voltage. The temperature monitor circuit in a mixed-signal FPGA has a built in 12.5x amplifier, enabling the pre-measurement signal to be amplified resulting in a more accurate measurement of the temperature signal. Therefore, the above equation would need to be modified slightly to account for the voltage amplification prior to being passed to the ADC.



V = (T * n * k * ln(IH / IL) / q ) *12.5

V = ( 298 * 0.00019835) * 12.5 = 738 mV

FILTERING OUT THE NOISE

In general, temperature measurements are notoriously noisy. To adjust for this, the common practice is to average many successive measurements and work with the result. Fortunately, temperature is a relatively slow moving measurement, so that many (often up to 1,000) measurements can be taken without fear of the actual signal changing. The mixed-signal FPGA can easily average these measurements by filtering out the system-level noise. Software tools often support an easy graphical user interface (GUI) with which to set the averaging or filtering factor.

Now that the temperature of a location within the system has been determined, how should the design engineer use this information?

Most systems are designed to operate within a range of temperatures. Therefore, knowing where within these ranges the system is operating allows the system can respond accordingly. A mixed-signal FPGA can very easily monitor the averaged temperature, compare it against user-defined threshold values, and set flags accordingly. Again, because these solutions are programmable logic devices, many different thresholds can be implemented.

To illustrate the point, imagine a telecom chassis line card, where the user has defined three operating zones: Good, Warm, and Hot. This temperature reading is used as the feedback loop to control the amount of cooling required by the system.

When the averaged temperature is within the Good zone, no system changes are required. This is the ideal operating condition and the Good flag is asserted. However, internal or external forces can exert pressure on the system and increase temperatures, moving the operating zone from Good to Warm. As a result, the Good flag will deassert and the Warm flag will assert. While this zone would be defined as a valid region of operation, the system is now approaching temperatures where system damage could occur. Further, if the cooling system becomes overloaded or malfunctions, the temperature will continue to rise, eventually causing damage to the line cards within the system. An intelligent thermal management system is designed to prevent that from happening.

The system can now begin to take corrective action. The line card informs the system master of the elevated temperature. Then, the system master increases the cooling system output to counteract the temperature increase. The cooling system should increase output to bring the system back into the Good zone, causing the flags to be asserted appropriately. Additionally, the mixed-signal FPGA, acting as the system master, can record and time stamp this thermal event within the embedded flash memory for later retrieval by a service technician.

When the averaged temperature moves from Warm to Hot, the mixed-signal FPGA can take more drastic action as the system is now in serious jeopardy of damage. The line card begins shutdown procedures to prevent this damage. A signal can be sent to the system master to indicate this action. The time stamped conditions of the line card are saved to the embedded flash for future debug and failure analysis. The mixed-signal FPGA handling power management functions as well as thermal management can shut down power to the line card preventing damage.

CONCLUSION

Thermal management is becoming a more important aspect of many designs today, as faster, smaller process geometry devices proliferate into more and more diverse applications. With only external diode-connected transistors, the mixed-signal FPGA can measure the temperature on many different locations on a system board. The mixed-signal FPGA has the processing power to also filter the noisy signal for more accurate assessments. As a programmable logic device, very flexible thermal windows can be created, each with a unique system response from a simple time stamp of the thermal event to shutting the system down to avoid permanent thermal related damage. Mixed-signal FPGAs enable the easy and cost-effective integration of thermal management into a comprehensive system management solution.

Click here for Illustrations:

Figure 1, Figure 2, Figure 3