Cars are equipped with a great many electronic devices, and as cars become more intelligent and gradually become fully automated, the number of these electronic devices will continue to increase.
The average number of electronic control units (ECUs) in automobiles now exceeds 40, with some luxury cars integrating over 100 independent computing units. With the advent of digital driving, advanced driver assistance systems (ADAS), and autonomous vehicles (AVs), the demand for high-quality human-machine interfaces (HMIs) and autonomous driving functions is growing exponentially. This increased demand for power consumption urgently needs to be addressed, as automotive systems have strict budget limits for the power consumption of electronic components.
Increased power consumption requirements
Based on experience, the processing power between each level in the driving level chart developed by the Society of Automotive Engineers (SAE) can increase tenfold. Level 0 indicates no driver assistance features, while Level 5 indicates the car can achieve autonomous driving in any environment. Currently, most new cars already possess Level 2 driving capabilities, but achieving fully autonomous driving is still a long way off.
Image: Driving grades established by the Society of Automotive Engineers (SAE)
Many parts of automotive systems that traditionally did not use electronic components are now incorporating electronics, posing an increasing challenge to the power consumption design of vehicles. Major subsystems in modern vehicles include the electric motor management unit, multimedia system, heating system, ventilation and air conditioning (HVAC) facilities, chassis electrification, lighting (exterior and interior), and future-oriented applications such as autonomous driving (AV), all of which contribute to increased power consumption in automobiles.
Looking at the car's dashboard itself, it's powered by a 12V battery, but it doesn't have unlimited power generation capabilities. We've also recently seen some large SoC chips with power consumption ranging from 100-300W, which means these chips won't be used in car dashboard systems anytime soon.
One of the main issues is temperature. Cars produced for the global market must be able to operate normally in a range of -40°C to 150°C, from the frigid North and South Poles to the scorching heat of the Sahara Desert. The temperature inside the car and on the dashboard can be much higher than the ambient temperature, and the components also generate heat continuously during operation. If the power consumption is too high and exceeds the temperature tolerance limit of the components, it will cause unreliability or even failure.
Many automakers are developing semi-autonomous vehicles with electronic systems consuming over 100W. This is fine for system prototypes, but putting them into mass-produced vehicles may require adding expensive water-cooling systems, which will affect the reliability of the car and increase costs.
Addressing future power consumption needs
Let's examine SoCs, especially GPUs, where there are various ways to reduce power consumption. Over the past decade, with the advancement of electronics, the complexity of power management has increased dramatically. As a result, power management has had to be applied to larger SoC projects. The biggest impact is to look at this problem at the architectural level and solve it at the highest level—which is what PowerVR has been doing for the mobile market for many years.
Currently, various technologies can be applied to SoCs and chip cores, such as clock gates, power gates, save and restore features, and Dynamic Voltage and Frequency Scaling (DVFS). Each part within an SoC device is divided into different power islands, which can be turned on when in use and off when not in use. When executing instructions, advanced devices can activate the relevant parts of the SoC, and power islands can also be set on different voltage rails, adjusted according to performance and power requirements. They also time at a slower rate to reduce power consumption when not fully loaded.
One of the challenges faced by power architects is the lack of specific standards on how to handle power management. As a result, a SoC consists of multiple cores from different vendors, which means it can have a range of different power management approaches that must be addressed at the architecture level.
In addition to the design techniques mentioned above, there are many other ways to reduce power consumption in GPUs. One of them is Tiled Deferred Rendering (TBDR), which uses the device's local memory to store tile content. This reduces the bandwidth of external memory, thereby reducing power consumption.
We also have a range of compression techniques to reduce memory bandwidth. For example, PowerVR uses three compression techniques: texture, image, and parameter compression. These help reduce memory usage bandwidth and power consumption.
As true-color textures consume a significant amount of memory in the ground state, texture compression becomes essential. PowerVR has its own texture compression algorithm, which we call PVRTC. This technology can significantly reduce texture size, thereby reducing memory bandwidth.
Summarize
With the significant increase in computing power required for Level 4 and Level 5 autonomous driving, coupled with the expansion of ECUs within vehicles and power budget constraints, power consumption is becoming a critical issue for SoC designers. However, with the support of advanced technologies and analytics tools, these complex power management problems can be solved efficiently.
Disclaimer: This article is a reprint. If there are any copyright issues, please contact us promptly for deletion (QQ: 2737591964). We apologize for any inconvenience.
