As Level 3 and Level 4 autonomous driving technologies become more widespread, drivers can shift their attention away from the road and reduce their focus on driving. The demand for sensors is constantly growing, and these sensors will be integrated into vehicles to replace the driver's eyes. The increasing use of cameras, external LiDAR, and even internal sensors to monitor driver safety and attention is undoubtedly a good thing for the future of driving, but one major challenge remains: power consumption.
Every watt of power consumed by a sensor affects a vehicle's range. So how can automakers achieve the characteristics of future cars without compromising range?
Considering the design of new energy vehicles
A simple approach is to improve the efficiency of the system itself. Developing low-power sensors capable of acquiring data can immediately save power, but all acquired data needs to be transmitted to a processor for processing, and the more data there is, the greater the power consumption of the chip. Advanced autonomous driving concept cars and advanced ADAS systems rely on PC-level water-cooled GPU solutions to handle these intensive and complex calculations. However, when battery capacity is limited, a 350W GPU becomes insufficient.
One of the best ways to solve this problem is to design from the ground up, considering the car as a platform. Instead of a multi-modal design, use higher-density chips for more efficient computing while reducing the number of chips in the vehicle. Custom designs allow for functional safety design from the ground up. For example, the recently designed Imagination XS series GPUs for automobiles comply with the ISO 26262 standard. This means that a single GPU can perform fault detection internally. Compared to other solutions, automakers require 50% fewer chips.
From macro to micro, amplifying small gains
In our main ADAS GPU products, we have custom-designed, high-density, functionally safe chips, but how can we further reduce power to allow drivers to travel more miles without shutting down all devices? Any modern car integrates hundreds of electronic control units (ECUs) distributed across various modules. These units control and monitor advanced ADAS functions, and can even control the car's power mirrors. Each microcontroller needs to manage its own function within the power budget of the entire vehicle system; the impact of one or two microcontrollers may be significant, but the overall power consumption is substantial. Reducing the number of ECUs in a car is also a functional safety consideration. Each core needs to manage a separate application or function to ensure no information overlap and overall system functional safety. IP vendors are exploring ways to challenge traditional practices with a safety-focused approach.
One way to reduce power consumption is to consolidate the functions controlled by different ECUs into a single multi-core ECU processing unit. By merging multiple functions, whether for positioning or other applications, we can reduce the need for individual ECUs and the resulting power consumption. Hardware virtualization is one solution option that allows the GPU to execute multiple workloads, such as the dashboard, rearview camera, and infotainment control center, each handled by a separate chip. By isolating these functions on a single GPU, multiple operating systems and functions can run independently simultaneously. By self-checking for faults in each core in accordance with functional safety standards such as ISO 26262, we can easily see a significant reduction in the number of ECUs.
Keep moving forward
New energy vehicles are not yet widely used, and the industry has not yet established a safer standard for autonomous driving. From ADAS to door locks, every watt of power is crucial. The key to tightly integrating new energy vehicles with autonomous driving technology lies with IP suppliers and chip manufacturers, who will continue to provide solutions for developing various functionalities and achieving low power consumption in automobiles.
