ADAS and vehicle automation: Learning from others’ experience can help improve our own

Automakers can benefit greatly from data centers when designing scalable systems and applications that require low latency and high energy efficiency.

The inclusion of Advanced Driver Assistance Systems (ADAS) features has become an important aspect of automotive design to improve safety and ease of use. Manufacturers are looking to build cars with higher levels of automation and ultimately fully autonomous driving (AD).

ADAS and AD, coupled with rising user expectations for infotainment and personalization, mean that cars are evolving into mobile data centers. As a result, communication between the key hardware elements (ICs, boards, or modules) required for software-defined vehicles (SDVs) is essential for successful operation. In fact, some cars today already contain more than 100 million lines of code, and Straits Research predicts that the automotive software market will reach nearly $58 billion by 2030, growing at a compound annual growth rate of 14.8%.

Software complexity and real-time processing of large amounts of data from various vision system sensors (such as cameras, radar, lidar, and ultrasonic) presents challenges. For example, Figure 1 shows that the traditional communication infrastructure and standards used in the automotive industry have reached their limits. Ethernet and controller area network (CAN) buses will still have a place in future automotive architectures, but must be supplemented to meet the needs of high-performance computing platforms (HPC) to embed artificial intelligence (AI) and machine learning (ML) in ADAS and AD.

Figure 1 – Cars are becoming data centers on wheels, as ADAS must process large amounts of data from different types of sensors in real time.

PCIe® Technology

Peripheral Component Interconnect Express (PCIe) technology was introduced in 2003 to meet the needs of the computing industry. Today, PCIe is being deployed in aerospace and automotive to implement safety-critical applications in firmware that must comply with DO-254 standards.

PCIe is a point-to-point bidirectional bus. As a hybrid serial bus, it can be implemented in a single channel or 2, 4, 8, or 16 parallel channels to achieve greater bandwidth. In addition, the performance of each generation of PCIe is constantly improving. Figure 2 shows the evolution of PCIe.

Figure 2 - PCIe® performance evolution

PCIe has been used in some automotive applications for a while now, and it has been around since Gen 4.0. With the performance increase of Gen 6.0, the data transfer rate is 64 GTps, and if 16 lanes are used, the total bandwidth can reach 128 GBps. Now is a great time to adopt PCIe, and it is worth noting that PCIe also provides backward compatibility.

Cars are becoming data centers on wheels. Let's consider why PCIe is used in land-based data centers under such a premise.

High performance, low power consumption

A data center consists of one or more servers and peripherals, including storage devices, network components, and I/O to support HPC in the cloud. PCIe is used in today's high-performance processors and is an ideal bus for establishing low-latency, high-speed connections between servers and peripherals.

For example, the non-volatile memory standard (NVMe) is specifically designed to work with flash memory through the PCIe interface. PCIe-based NVMe solid-state drives (SSDs) offer faster read/write speeds than SSDs using the SATA interface. Indeed, all storage systems, whether SSDs or mechanical hard drives, cannot provide the performance required for complex AI and ML applications.

Providing low latency between applications running in servers via PCIe directly helps improve cloud performance. This means that PCIe will be embedded in components other than processors and NVMe SSDs. It also provides a gateway between the cloud and the systems that access the cloud, along with many components. It is important to note that while cars themselves are becoming mobile data centers, they will also become nodes that move between "smart cities."

From the perspective of power consumption, NVMe is also favored in data centers. For example, the U.S. Department of Energy estimates that a large data center (with tens of thousands of devices) requires more than 100 MW of electricity, which is enough to power 80,000 homes. Compared with SATA SSDs of the same size, NVMe SSDs consume less than one-third of the power.

In the automotive sector, power consumption is undoubtedly an important factor, especially for electric vehicles (EVs), which directly affects the driving range. In fact, automotive engineers (especially EV designers) are increasingly concerned about size, weight and power consumption (SWaP). This is not surprising, considering that future ADAS implementations may require up to 1 kW of power and require liquid cooling systems for thermal management.

But equally, there are opportunities to learn from other sectors. The aerospace industry has been designing products to meet stringent SWaP and cost (SwaP-C) requirements for decades, and liquid-cooled line-replaceable units (LRUs), such as power supplies, have been used in some military platforms for more than a decade.

Where to start?

Data centers have been taking advantage of PCIe hardware for years as they look to optimize systems for different workloads. They are also adept at developing interconnect systems that use different protocols; for example, using PCIe for communications that are not time-critical (like Ethernet for geographically dispersed systems).

In an automotive environment, those “not time-critical” communications include telemetry and lighting control between sensors. They do not necessarily require PCIe, but PCIe is needed for short-range, higher-data-volume communications between ICs that perform real-time processing and are only a few centimeters apart. Therefore, an optimized ADAS/AD system will likely need to include Ethernet, CAN, SerDes, and PCIe.

Unlike Ethernet, there is no specific automotive PCIe standard, but this has not limited its use in automotive applications in recent years. Similarly, the lack of an aerospace PCIe standard has not prevented large aerospace/defense companies (who are always striving for SWaP-C advantages) from using the protocol in safety-critical applications.

As solutions must be optimized for interoperability and scalability, PCIe is also gradually becoming the preferred computer interconnect solution in the automotive industry, providing ultra-low latency and low-power bandwidth scalability for CPUs and dedicated accelerator devices. Although there is no specific automotive PCIe standard, semiconductor suppliers are working hard to make PCIe further shine in the harsh automotive environment.

Figure 3 - PCIe® switch chip with low latency, low power consumption and excellent connectivity

For example, in 2022, Microchip launched the industry's first automotive-grade PCIe switch chips that meet Gen 4 requirements. These switch chips, called Switchtec™ PFX, PSX and PAX, provide the high-speed interconnect required for distributed, real-time, safety-critical data processing in ADAS architectures. In addition to these switch chips, the company also provides other PCIe-based hardware, including NVMe controllers, NVRAM drivers, retimers, redrivers and timing solutions, as well as flash-based FPGAs and SoCs.

Finally, another thing the automotive industry must consider is that data centers view capital expenditures as a way to invest in future annuities. Until now, most automotive OEMs have viewed capital expenditures as having a one-time payback (at the time of purchase), which works well for hardware. Granted, most OEMs charge for software updates from time to time, but the business model for SDVs needs to be rethought entirely. It is no longer appropriate to focus solely on the bill of materials cost of hardware.

Summarize

To increase the level of automation in cars, the car needs to become a high-performance computing "data center on wheels" to process large amounts of data from various sensors. Fortunately, HPC has matured and become the core of high-frequency trading (HFT) and cloud AI/ML applications. Proven hardware architectures and communication protocols such as PCIe are available. This means that automakers can benefit greatly from HPC implementations in data centers.

Because AWS, Google, and other cloud service providers have been developing and optimizing their HPC platforms for many years, much of the hardware and software is readily available when needed. Automakers can take advantage of these existing HPC architectures instead of reinventing the wheel and building solutions from scratch.