AM335x Thermal Chamber Experiment

=Overview= ''This experiment is still ongoing! Additional data will be updated periodically.''

An experiment was completed to analyze the thermal behavior of the AM335x on real systems in a controlled environment. Using a thermal test chamber to manipulate external environmental conditions, such as ambient temperature and air flow, we simulated a range of temperature scenarios to model common customer use cases. By altering thermal management variables like board size, MPU frequency, and application intensity, we can create an operating thermal profile for each AM335x system tested. These thermal profiles are intended to serve as example use cases for users to evaluate during their thermal design process. Users can utilize the collected data to interpolate an approximation of their system’s thermal behavior, and estimate the risk of exceeding the maximum operating junction temperature in their system—and consequently, determine the need for improved thermal management.

The example AM335x systems chosen for this experiment do not contain any complex cooling systems; they are both heat sink-less and fan-less. Both boards were enclosed and populated with worst-case hot silicon to account for the additional leakage power incurred at high temperatures. The systems were placed inside a thermal chamber to simulate a spectrum of ambient temperatures with no air flow. For each temperature test point, an application was run while monitoring the power consumption and case/board temperature. Starting at 25°C, this test procedure continued until the maximum operating junction temperature was exceeded.

In this experiment, two evaluation boards were tested to represent two types of systems with contrasting thermal efficiencies:


 * Rev 1.5A General Purpose EVM baseboard
 * 6.69” x 6.69”
 * 6 layers (4 signal, 2 power)
 * Components are placed in spacious design to fit in mini-ITX form factor
 * BeagleBone Black

Having over 6 times the board area, the EVM baseboard represents a more thermally effective system solution than the BeagleBone. This means more copper and board space for spreading and dissipating heat. The EVM also has a better component placement in terms of thermal management, as the major sources of heat (processor, PMIC, Ethernet PHY, etc.) are placed spaciously apart on the PCB. The extra room also allows select traces to be wider on the EVM.

Both test systems were enclosed in ABS plastic cases to remove any airflow from the system. The EVM was enclosed in a B&amp;W Type 10 Outdoor case, and the BeagleBone was enclosed in a Twin Industries B10-7100 ABS project box. These enclosures further reduce the thermal efficiency of the test systems, as heat must be dissipated from the processor/PCB through the internal ambient air, through the ABS plastic enclosure, and through the external ambient air of the thermal chamber. One goal of this experiment was to create a thermally-poor system (within reason) running an MPU-intensive application, with the hopes of achieving a worst-case to give users an extreme example for assessment; here, this is the enclosed BeagleBone.

=Setup=

Electrical

 * Rev1.5A GP EVM Baseboard; power/serial cables, SD card
 * BeagleBone Black; power/FTDI cables, microSD card
 * PMDC, with required daughterboards/cables
 * Omega TC-08 8 channel thermocouple, thermal grease, thermal stickers
 * Keithley 2420 Source Meter
 * Keithley 2701 Multimeter
 * Vötsch Industrietechnik VT 4002 Thermal Chamber

Mechanical

 * B&W Type 10 Outdoor Case
 * Twin Industries B10-7100 ABS Project Box
 * Drill, screws, spacers for mounting test boards inside enclosures

Software

 * Linux environment to run power measurement scripts
 * Perl with SerialPort module for PMDC automation
 * Python with pexpect module for Keithley 2701 automation
 * Terminal emulator (Teraterm) for interfacing with test boards
 * Omega Logging Software for capturing thermocouple data

=Preparation= The high-level procedures for preparing the experiment are as follows:
 * 1) Attach thermocouples to test board
 * 2) Mount board into enclosure
 * 3) Verify serial and power measurement connections
 * 4) Verify test script “dhrystone_temp_xxxHz.sh”
 * 5) Place enclosure into thermal chamber

Power Measurement Equipment

 * Keithley Source Meter
 * Edit voltage (5V) and current (1A) range settings
 * Connect to BBB and eyeball current measurement
 * PMDC
 * Connect all three USB ports to Linux machine
 * Run “AM335xEVM_PMDC_linux.pl”
 * Wait until power has somewhat stabilized at a certain point
 * If all rails are not captured, touch the pins on daughterboard connected to older PMDC
 * Ctrl-C to exit
 * Keithley 2701 Multimeter
 * Connect cables to EVM supplies
 * Connect straight-through RS232 cable from DMM to PC
 * Test by running keithley.py and verifying power

Temperature Measurement Equipment

 * Thermal Chamber
 * Turn on switch on the back of the battery
 * Turn switch on chamber
 * Allow system to boot up
 * Press ‘←’ until T/C at the bottom is lit, then adjust desired ambient temp with ‘↑’ and ‘↓’
 * Chamber will heat up, overshoot, undershoot, until stabilized
 * To power down, reverse power up sequence
 * Hold ON button on front of battery for 5 seconds to turn off completely
 * Thermocouple
 * Apply a dab (no more than 2mm x 2mm) of thermal grease to desired measurement point
 * Insert end of thermocouple into the thermal grease
 * Cover with a thermal sticker
 * Secure thermocouple cable to the board with stickers to minimize the heat sinking nature of the thermocouple. For case temperature measurements, the wire should be dressed along the diagonal of the package, down to the PCB surface, and over a minimum distanced of 25mm before lifting from the PCB.
 * Connect USB cable into computer and Omega thermocouple
 * Open Omega Logging Software to start monitoring the temperature
 * Bandgap register
 * Data is output in serial terminal while running the Dhrystone script
 * Record 2 hex characters in [31-16] (should range from 00 to 7F)
 * Convert hex to degree in Excel lookup table

=Test= All tests were performed in the TI-Dallas ARM MPU Lab, Dallas TX, USA (430ft/131m altitude). The following describes the general test procedure that was performed:
 * 1) Set desired ambient temperature on the thermal chamber
 * 2) Wait until ambient is stabilized and case is stabilized
 * 3) Record power consumption, case, board, ambient temperatures, bandgap register data
 * 4) Change variable (ambient temperature, MPU frequency, use case, SmartReflex, test board)
 * 5) Repeat steps 1-3 for each data point

When performing temperature sensor testing, we need to wait for the chip and board to stabilize thermally once at the correct temperature.

=Data and Results=

The following data was collected after running several use cases on 4 different systems: BeagleBone Black, EVM baseboard, enclosed BeagleBone Black, and enclosed EVM baseboard. For each system, power and temperature measurements were captured over combinations of 2 applications and 4 operating performance points (OPP), which are fixed voltage and frequency targets.

The applications selected were “OS Idle” and “Dhrystone.” OS Idle is the use case where the AM335x processor is idling at the Linux Matrix GUI, waiting for a command. Here, the processor is active, but consumes minimal power. Dhrystone, on the other hand, is the use case of running the Dhrystone benchmark continuously. This application was selected to represent a computationally-intensive use case, as it exhausts 100% of the MPU and consumes a significant amount of power. Both of these applications are available in the Linux EZSDK for Sitara ARM Microprocessors.

Certainly, more power-hungry applications exist, but they are beyond the scope of this experiment. Dhrystone was chosen because of its effect on the MPU power domain. As the MPU frequency is scaled up to 1GHz, thermal issues become more prevalent with the increased processor power consumption. Your system may be significantly different than the ones described in this experiment! Your application may consume more power than Dhrystone, and your system may be worse at dissipating the generated heat. Please be cautious when comparing the following results and conclusions with your system.

System Comparison at 1GHz, Dhrystone
Note: the results of this test are currently under speculation.   As expected, the enclosed BeagleBone performed the worst and the more thermally-efficient EVM performed the best. The EVM was able to withstand an ambient temperature of 60°C (140°F) before exceeding the 90°C maximum operating junction temperature. However, the enclosed BeagleBone exceeded this at just over 35°C (95°F) ambient.

EVM
Running the Dhrystone application at 1GHz causes the junction temperature to exceed 105°C at an ambient temperature of 75°C. By scaling down the MPU frequency to 300MHz, the maximum operating ambient temperature is extended to 85°C. In the OS Idle application, the ambient temperatures can exceed 85°C before the junction temperature gets too hot.

SmartReflex is an active power management technique which optimizes voltage based on silicon process. Because these test systems use a worst-case hot silicon process, we can use SmartReflex to reduce active power—and as a byproduct, reduce heat. In this test, we investigate the thermal management effect with SmartReflex enabled at different frequencies. On the EVM baseboard with a worst-case silicon unit, SmartReflex reduces the operating junction temperature by 1-3°C, depending on frequency.

The thermal effect of MPU frequency is significant. At an ambient room temperature of 25°C, the AM335x junction temperature heats up ~10°C while running the Dhrystone application at 300MHz. Operating at 1GHz, the junction temperature increase from ambient is ~22°C! This effect increases as ambient temperature increases.

=Conclusion= ''This experiment is still ongoing! Additional data will be updated periodically.''