Smartphones feel the heat
Designing a premium smartphone has been a delicate balance between performance and battery life for generations. However, continued increases in performance cause growing concerns around heat.
The increased heat generated by advanced SoCs has led to thermal issues during benchmark testing of several of the latest premium smartphones using high-performance SoCs, such as testing of the HTC One M9 and LG G Flex 2 by Ars Technica. These premium smartphones use the Snapdragon 810, one of the most popular 3rd party chipsets on the market.
Despite the claims, however, no one has identified a specific issue with the Snapdragon or any other chipset that would cause the SoC and phone to overheat. To attempt to clear up this matter, Tirias Research has investigated the issues around smartphone overheating.
Design issues
The primary tool for silicon vendors to increase performance efficiency has been through the use of an increasing numbers of compute cores because increasing clock frequencies increase power even more. In premium handsets smartphone SoCs, the number of central processing unit (CPU) and graphics processing unit (GPU) cores has been doubling the past few generations to the point where we have up to 10 CPU cores and hundreds of GPU cores in the highest-end smartphone SoCs.
In addition, many smartphone SoC utilize highly specialized compute cores for image processing, video encoding and decoding, audio processing, signal processing for cellular communications, and sensor hubs. This all leads to more power sources that increase performance and can create thermal issues.
The problem is further exacerbated by the foundation of silicon technology — Moore’s Law. Each process generation results in smaller design features the further condense these compute cores and the space between the cores. This has the potential for creating hot spots when cores are operating near their peak potential so close to other cores.
To address the issue of higher power consumption and to extend battery life, SoC vendors have turned to increasingly complex designs that allow for separate power planes for different cores and increasingly minute levels of clock controls, referred to as clock gating.
The finer the power solution, the more of the SoC can be disabled when it’s not required. While this also aids in reducing thermals to some extent, it does not eliminate the potential problem because the SoC is not the only component contributing to thermal issues.
There are other components that can lead thermal issues when being driven to perform at or close to peak rates, including the power management IC (PMIC), the power amplifier, the image sensor, and the display. Many are the result of applications or operating conditions.
Calling or transmitting data from the edge of the cell network forces the power amplifier to or at peak levels; on-line gaming can drive the display, power amplifier, and SoC are high levels; and capturing and encoding 4k video can drive the image sensor, display, SoC, and memory at high-levels.
In all of these cases, the components temperature can rise to high levels causing the case or skin of the smartphone to heat-up excessively. And, as smartphone vendors push for thinner designs and more rigid materials, such as metal, the potential for thermal issues increases because these materials more readily transmit heat to the surface of the phone. A simple change or error in design or assembly can alter the thermal characteristics of a device and increase the potential for thermal issues.
There are two key technologies that prevent these premium smartphones, and most other electronic devices for that matter, from entering thermal runaway, or essentially catching fire or melting – thermal sensors and the smartphone software. The thermal sensors are hardware components positioned in areas of a chip or device to measure the temperature.
These temperatures are monitored in the system software according to specification of the various components. Typically, the silicon vendor provides recommendations and/or settings to the smartphone OEM, but it is up to the OEM to determine what that final settings will be and this may change over time as testing or customer feedback requires modifications. If the component temperature exceeds recommended levels, that system software reduces performance until acceptable temperatures are reached.
Putting the smartphones to the test
All premium smartphones face thermal design challenges, especially when using a leading-edge SoC. However, testing for the difference in designs and software settings requires testing smartphones with both the same components and those with different components.
Due to the purported thermal problems with the Snapdragon 810 in early benchmarking testing of the HTC One M9 or LG G Flex 2 handsets, we chose to focus on handsets with this chipset. For comparison purposes, we chose a handset using a competing Samsung chipset, the Exynos 7420. Note that the Snapdragon 810 is an octa-core SoC manufactured using a 20nm TSMC process and the Exynos 7420 is an octa-core SoC manufactured on Samsung’s 14nm FinFET process.
For the tests, we used the following handsets with the associated specifications. Note that with the number of design variables, it is impossible to create an exact comparison between any of the smartphones, but it does provide a basis for analysis on some of the internal components and the smartphone design.
Figure 1: Other handsets using the Snapdragon 810 SoC include the Fujitsu Arrow NX, HTC Butterfly 3, LeTV LeMax, LG Flex 2, Sharp Aquos Zeta SH-03Z, Sony Xperia Z3v, and ZTE Axon. Click image for larger version.
The tests were intended to stress the key components of the handsets to determine the thermal results and potential for thermal issues in accordance with natural usage models. Although not expected to yield the same results, a few industry benchmarks were also included to determine if there was a difference. The set-up included a thermal imaging camera capturing the image of three smartphones at a time and temperature probes located on the front and back of each smartphone. All smartphones were cooled and allowed to reach an ambient temperature before each test was conducted.
Figure 2: Test apparatus.
Test 1 — 4k video
As indicated earlier, capturing and encoding 4k video can place high demands on the image sensor, the SoC and the flash memory of a smartphone. To accomplish the test, a spinning color wheel was used to provide a consistent and changing image for each of the smartphones and a thermal image was taken at the five minute mark. Note that each phone had different settings on how much video can be captured according to its memory capacity. Five minutes was the lowest 4k video time limitation of the phones tested.
Figure 3: Thermal images and sensor data from 4k video test. Click image for larger version.
Notice that each phone demonstrated different image patterns based on the location of key components, particularly the image sensor and the SoC. The image sensor is visible by the small square or rectangular box located on each image and the SoC by the larger area displaying the most heat.
In the case of the Samsung and Xiaomi devices, the image sensor and SoC are located in the same approximate area. Once noticeable result was that the thinner smartphones, the Samsung and the Xiaomi, exhibited the highest skin temperatures and thermal images.
While all the smartphones exhibited similar hot spots, the skin temperature between devices varied by just over 8°C on the front and 4°C on the back. In addition, all of the devices operated well within normal operating limits without generating enough heat to make using the smartphones uncomfortable. However, those hot spots could lead to component fatigue and failure over time.
Test 2 — Gaming
Figure 4: Thermal images and sensor data from gaming test (Need For Speed) Click image for larger version.
In this test, the SoC is more defined in the thermal images and stands out as the primary culprit of the heat generation. Note, however, that other components can also generate heat as exhibited by the heat island in the lower edge (top to the picture) of the Xiaomi device. This could have been a result of the USB connection but it is not clear without cracking open the device, which was beyond the scope of our test.
Also visible is the drastically different heat generated by the HTC, ZTE, and Xiaomi devices (The lower set of images) despite using the same SoC. This highlights the difference in design and likely the software settings.
Test 3 — Browser benchmark
Figure 5. Thermal images and sensor data from graphics test (GfxBench T-Rex benchmark). Click image for larger version.
With the frame rate nominalized at 45 frames per second (FPS), the devices once again displayed significantly different thermal images and skin temperatures.
Additional tests were run using other system-level benchmarks include Geekbench 3 and AndEBench Pro, but none of the devices demonstrated thermal issues with those tests. This further indicates the impact of the applications software on thermal issues in smartphones.
Test results
As seen by the result, the thermal images and temperature measurements vary by smartphone even when using the same SoC and version of the OS. Once again, the factors that can impact this include: the components, the device design, the case thickness and materials, and the firmware settings. In real-world applications, however, even the version of the operating system and the applications could change the thermal characteristics of the smartphone.
Often, a handset may heat up because of multiple applications running or an applications running in the background. While the device is protected from entering a thermal runaway state, the device can become warm to the user if they are talking on it, holding it, or even have it close to them in their clothing.
On the morning of the test, I upgraded Android on my Samsung Galaxy S5. During the upgrade, the S5 became warm to the touch, but cooled quickly once the upgrade was complete. This demonstrates the challenges smartphone OEMs have in designing and programming for situations that may even be beyond their control.
And in all of the test cases, none of the phones reached a point of overheating past the OEM or silicon vendor recommendations, or to a point that would make using any of them uncomfortable.
Benchmarking drawbacks
While these tests did not simulate the exact situation, which was to put a system benchmark into a loop to run it over and over again, that caused the HTC One M9 or LG G Flex 2 to overheat during the initial benchmark tests of the devices, none of the devices came close to overheating during these tests.
This begs the question: Why did those devices overheat during the initial benchmarking? The answer could lie in the newness of the devices or the benchmarking process itself.
On one hand, OEMs often provide preproduction units to reviewers before they are completely qualified for release. As a result, not all the software and settings may have been fully tested for potential issues. On the other hand, benchmarking is about achieving the best scores.
Maybe, the OEMs disabled some of the thermal protection features by adjusting the thermal settings to a higher level or by adjusting other system operating settings to higher levels to achieve a higher benchmark score. Logic would tend to lead toward the latter reasoning, but without any official claim from the OEMs, any or all of those conditions may have led to the resulting situation. However, there are no reports of higher than normal return rates on any of the smartphones using the Snapdragon 810 chipset.
Takeaways
Just as PC vendors faced a thermal barrier when using the dreaded Intel Pentium 4 processor over a decade ago, smartphone OEMs face similar challenges with smartphone designs today and the SoC is just one factor to consider. While the two silicon solutions are drastically different in operation, the smartphone SoCs are much more complex with over 30x the number of transistors in roughly a 15% smaller die on average.
Unfortunately, other mobile and small form factor solutions face similar design challenges going forward, especially as performance requirements increase, package requirements shrink, and the power-saving benefits of Moore’s Law wane.
About the author
Jim McGregor is the principal analyst for Tirias Research – www.tiriasresearch.com
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
