Energy Efficient Thermal Solutions
Multichannel Temperature Sensors
The ever-increasing power supply requirements and board densities in computing systems require efficient thermal management solutions. Factors such as cost, accuracy and system size help determine the type of temperature sensor applicable to manage the thermal load of the computing system.
Silicon-based temperature sensors use a fundamental property of bipolar transistors to determine temperature. Temperature may be calculated by measuring a change in voltage when two different currents are implemented. Figure 1 and Equation 1 illustrate the concept of a silicon-based temperature measurement.
Figure 1: Silicon-based temperature measurement
Equation 1: Silicon-based temperature measurement
Where T = temperature in Kelvin
ΔVBE = change in diode base emitter voltage
k = Boltzman’s constant
q = electron charge
n = diode ideality factor
IC1 and IC2 = currents with n:1 ratio
The EMC18xx is ideal for low-cost motherboards to monitor the ambient temperature at several locations. These may include the chassis, add-in card sockets, and surrounding components as shown in Figure 2.
As temperatures increase due to high levels of processor load, whether it is the Central Processing Unit (CPU), Graphics Processing Unit (GPU), or memory modules, the system may throttle the clocking frequency to mitigate the temperature rise in the system. The ±1°C maximum accuracy of the EMC18xx eliminates the concern of premature throttling or system shutdown due to increasing system temperature.
- Maximum error of ±1°C from −20°C to +105°C
- Programmable temperature limits
- Configurable ALERT function as a comparator mode or as a system interrupt mode
Figure 2: System diagram
For higher-density applications, such as high-end system boards for laptops, server racks and tablets, a remote diode temperature sensor such as the EMC1814, may be used for increased flexibility and reduced component count.
The Anti-Parallel Diode (APD) feature of the EMC1814 allows for a single device to monitor several zones at once, leveraging a stand-alone diode connected transistor or the discrete diode-connected transistor of a CPU, GPU, or an Application-Specific Integrated Circuit (ASIC) processor.
It is important to understand the processor geometry in today’s computing products. A substrate diode is often used as the temperature sensor. Decreasing geometries degrade the accuracy of the sensor. To ensure accuracy, the EMC1814 employs automatic beta detection before every conversion. Microchip’s patented frequency hopping technique mitigates noise coupling into the input traces from switching noise sources including backlight inverters, Switch Mode Power Supply (SMPS) and other sources of Electromagnetic Interference (EMI).
The EMC1814’s Resistance Error Correction (REC) mitigates temperature error from long Printed Circuit Board (PCB) traces, cabling and interconnects resistances. Without REC, every 1Ω of resistance in the measurement path would have added an approximate +0.7°C error to the temperature measurement.
The key features of the EMC1814 are:
- Maximum error of ±1.5°C from −20°C to 105°C
- Programmable temperature limits
- Configurable ALERT and THERM functions for system interrupts
Fan Controllers
Temperature sensors may also work in conjunction with fan controllers ensuring a greater depth of thermal management. Microchip offers fan controllers with programmable features allowing for flexible solutions for modification to servers, Liquid Crystal Display (LCD) projectors, workstations and networking equipment racks.
Audible noise from the fan spinning up or changing Revolutions Per Minute (RPM) too abruptly may dampen the end-user experience. The spin-up routine (Figure 3) of Microchip’s fan drivers, such as the EMC2303, mitigates audible noise during the initial start of the fan. During normal operation, programmable ramp-rate control may be implemented to reduce audible fan noise when the speed of the fan is required to change.
Figure 3: Spin-up routine
In servers and network racks, the EMC2303 allows for multiple zones to be cooled. When a system detects a thermal condition from its array of temperature sensors, the CPU will drive the fan to increase airflow volume to lower the temperature.
Typically +12V fans are implemented in the server space requiring an interface circuit to isolate supply voltages as illustrated in Figure 4. The isolation circuit uses only two components, a Field-Effect Transistor (FET) and a Schottky diode.
Figure 4: 12 V isolation circuit for fan drive
The key features of the EMC2305 are:
- Closed-loop, Proportional-Integral-Derivative (PID) control
- High-frequency Pulse Width Modulation (PWM) (26 kHz) to reduce acoustical noise
- Stalled fan and aging fan detection
Cloud computing solutions for modern data centers further increase the need for thermal management. High-end servers rely on a distributed network of sensors and fans throughout the system. This network allows individual peripherals to be monitored for increased granular control for greater energy efficiency. Implementing APD, REC, automatic beta detection, spin-up and ramp-rate control in one solution while supporting five temperature zones (four external, one internal) to be monitored. Products such as the EMC2106 combine temperature sensing and fan control in a 28-pin Quad Flat No Leads (QFN) package.
Combined with closed-loop fan control, the EMC2106 drives up to two fans to provide additional airflow in the server rack. Figure 5 shows the block diagram of a blade server.
Other features of the EMC2106 are:
- Maximum error of ±2°C from 0°C to 125°C
- Stalled fan and aging fan detection
Figure 5: Blade server