Technical challenges in developing next-generation ADAS and autonomous vehicles

Scalable high-performance computing architecture to handle massive data growth

These increasingly advanced ADAS and autonomous driving systems being introduced to the market will gradually assist drivers in taking their eyes and mind off the road ahead. However, achieving this goal first requires adjustments to computing architectures. Despite the growing number of Electronic Control Units (ECUs) in vehicles, traditional automotive computing architectures consolidate processing tasks for each specific function within a single computing element, which integrates data from dedicated sensors or inputs. This approach is challenging even for lower-level ADAS systems. In higher-level autonomous systems, the cost, data rates, and power dissipation of the required processing make reliance on a single processing element impractical.

Handling the massive increase in data necessitates a revision in computing architectures that distributes workloads across multiple cores within one or more processors. This architecture forms a central compute cluster, effectively creating a "data center on wheels." In this data-center model, specialized processing elements perform image analysis on camera data streams. Separate processing cores or CPUs can fuse camera data with radar or lidar data, using the “fusion” of data from different sensor types to better understand the vehicle's surroundings. Other cores can then focus on "policy" processing, determining the required actions based on the outcomes of sensor analysis.

Learning from decades of data center development, the fundamental elements of a scalable High-Performance Computing (HPC) architecture have been identified. These include HPC System-on-Chips (SoCs) with specialized image processing (inference and acceleration) engines, and high-bandwidth, low-latency PCIe® interconnects for data transfer between SoCs. Additionally, in-vehicle Ethernet network connectivity links to the HPC cluster, transmitting sensor data flows and control signals to other major vehicle systems. A safety processing element manages the operation of the HPC system.

Ethernet as the dominant technology for in-vehicle network backbones

In both highway and urban environments, any latency or delay in data transmission can be fatal. To rapidly and efficiently transmit data from sensors to edge processors, and then to or between central computing units, the vehicle's data network must be its most critical system.

Ethernet has become the mainstream technology for in-vehicle network backbones for most global OEMs. As the global data center standard, Ethernet offers a large supplier base, cost-effective and scalable bandwidth options, and various adaptations well-suited for the automotive industry. For mainstream network requirements, Single-Pair Ethernet (BASE-T1) standards at 100M and 1G speeds meet most demands and are widely deployed globally. BASE-T1 reduces cable count, offering weight savings and the potential for lower costs in the Ethernet physical layer.

The emergence of 10BASE-T1S technology provides a lower-bandwidth, cost-attractive alternative to CAN networking. The 10BASE-T1S standard supports multi-drop functionality, simplifying wiring to sensors across different vehicle zones. It also has the potential to eliminate ECUs previously used for processing and routing CAN messages, simplifying the design of zonal network controller nodes. These nodes aggregate data flows onto the backbone at various points within the vehicle, akin to highway entrances.

Serializer-Deserializer (SerDes) technology offers a high-bandwidth, cost-effective solution for continuous, predominantly unidirectional data transmission requirements. Cameras with raw data interfaces utilize SerDes to enable real-time data delivery necessary for high-performance image recognition. Using a SerDes interface eliminates the need for preprocessing on the camera, allowing raw data to be sent directly to the ADAS SoC, where optimized video engines can process the pixel-complete data more efficiently. This enhances overall system performance and avoids the cost of redundant preprocessing hardware at the camera side. In practice, almost all ADAS-targeted SoC processors integrate multiple camera interfaces and image processing engines, making data preprocessing on the camera an unnecessary resource drain.

For applications with only Ethernet connectivity or where pixel-complete data is not required (e.g., backup cameras), Ethernet-based cameras can be more cost-effective. Previously, SerDes solutions were often proprietary, locking OEM designs into specific vendors with associated hardware and software constraints. Now, with the Motion Link standard from the Automotive SerDes Alliance (ASA-ML), scalable and interoperable products are emerging, providing OEMs with a superior and flexible long-term ecosystem that enhances camera data management efficiency and system compatibility.

PCIe is the standard for high-bandwidth inter-CPU communication in data centers. PCIe offers extremely low latency, which is crucial for processing safety-critical data. PCIe achieves scalable bandwidth by simply increasing the number of lanes (an Rx/Tx pair) within any given "port," allowing connections to be easily optimized based on actual bandwidth requirements. Furthermore, the PCIe protocol is supported by virtually all high-performance processors, a significant advantage when selecting SoCs from different vendors for different tasks. Although there is no specific automotive PCIe standard, its superior latency and low processing overhead have led to its widespread adoption in automotive applications, with automotive-qualified chips already available. A common four-lane Gen4 connection used in current ADAS SoCs offers a bandwidth of 64 GT/s (roughly equivalent to 64 Gbps), making it highly suitable for high-speed sharing of image data.

Software-defined vehicles as the core of future automotive applications

Software development has become central to virtually all vehicle functions, extending beyond just developing AI systems to interpret the vehicle's environment and respond to emergencies. In early vehicle platforms, software was primarily used to implement unique, advanced functions like Anti-lock Braking Systems (ABS). With the evolution of infotainment and human-machine interfaces, application software, similar to that found in smartphones, began running on application processors within vehicles. Today, vehicle chassis, powertrain, safety features, and infotainment are increasingly implemented through software. This has given rise to the concept of the Software-Defined Vehicle (SDV), enabling vehicles to be upgraded throughout their lifecycle, enhancing existing functions and adding new ones.

Traditionally, software upgrades for legacy, function-specific systems like engine management were performed during service visits via a data cable. Now, an increasing number of manufacturers are adopting Over-the-Air (OTA) updates. OTA updates from OEMs can include bug fixes, new feature additions, or performance enhancements such as improvements for high-altitude driving or off-road handling.

From a software perspective, an SDV can now be considered a mobile data center. New functions can be added years after the vehicle leaves the production line, potentially reducing depreciation rates. Enhanced maintenance features could be sold as a package, enabling real-time monitoring of parts wear and tear, and scheduling maintenance based on actual usage rather than just mileage driven.

However, automotive companies with limited experience in implementing basic data center technologies must invest in developing new skills. Network management has become a specialized functional area not previously required for standalone systems. Unique to automotive, functional safety processes like ISO 26262, combined with data center hardware and software, influence the development and implementation of electronic and software systems.

Comprehensive product portfolio for ADAS and autonomous driving applications

Microchip offers a wide range of product lines to meet the demands of ADAS and autonomous driving applications. The main technology platforms in this field include Functional Safety, Embedded Security, Touch and Gesture, and Ethernet Technology, forming a quite complete portfolio.

The robustness, reliability, and security of end products are becoming increasingly important. Within the Functional Safety product portfolio, Microchip supports products that are compliant with functional safety standards or are functional safety-ready, such as Microcontrollers (MCUs) and Digital Signal Controllers (DSCs) including AVR® and PIC® MCUs, dsPIC® DSCs, SAM, and PIC32 MCUs, as well as Field-Programmable Gate Arrays (FPGAs) and SoCs. These products incorporate the latest hardware security features and supporting documentation to help achieve ISO 26262, IEC 61508, and IEC 60730 safety certifications. Some of Microchip's devices are already functionally safety certified, meaning they were developed following ISO 26262 compliant processes, are designed per AEC-Q100 standards, and possess professional hardware security features.

For the Embedded Security platform, Microchip provides a comprehensive portfolio of security ICs, along with security-focused MCUs, Microprocessors (MPUs), and FPGAs. It also offers software libraries, enhanced protocols, development kits, training, and other resources to help customers quickly start building secure solutions.

Within the Touch and Gesture application platform, Microchip offers MCUs with touch functionality, capacitive touch controllers, and 3D gesture controllers, enabling the replacement of mechanical buttons with touch or gesture-controlled interfaces to enhance the user interaction experience of end products. Microchip provides capacitive touch solutions for various applications. Its touch products include turnkey capacitive touch controllers, touch libraries for implementing touch sensing on most PIC®, AVR®, and SAM MCUs, as well as dsPIC33C DSCs, and single-chip solutions for adding gesture recognition to almost any product.

Microchip also offers flexible Ethernet solutions to add robust and reliable high-speed communication to embedded designs. Microchip's standalone devices, as well as Ethernet-enabled MCUs and MPUs, facilitate easy implementation of Ethernet in applications. Microchip provides timing solutions for its Ethernet products to help achieve higher reliability and lower power consumption in designs, meeting AEC-Q100 requirements for automotive networking applications. Microchip's high-performance Ethernet transceivers (PHYs) significantly reduce footprint, power consumption, and cost, offering 10BASE-T, 10BASE-T1S, 100BASE-TX, 100BASE-T1, and 1000BASE-T PHYs. Microchip's Ethernet bridges enable flexible Ethernet connectivity to host processors via USB, High-Speed Inter-Chip (HSIC), PCI™, or PCI Express® (PCIe®) interfaces, thereby shortening development time.

Conclusion

The development of next-generation ADAS and autonomous vehicles faces technical challenges including changes in computing architectures, data network transmission speed and interference immunity, and functional safety regulations for software-defined vehicles. Only through collaborative breakthroughs in hardware design, software algorithms, vehicular network infrastructure, and regulatory standards can autonomous driving truly progress towards large-scale commercialization while ensuring safety. This will pave the way for more efficient, safe, and sustainable development paths for intelligent transportation and future mobility. The numerous solutions offered by Microchip for ADAS and autonomous driving applications will assist customers in overcoming these challenges and developing the next generation of ADAS and autonomous vehicles.