91³Ô¹ÏÍø

The use of artificial intelligence (AI) in devices we use to make everyday decisions is making us more productive and, in some cases, keeping us safe. The use of generative AI in automotive systems is rampant and expected to grow even more. According to , ¡°AI in the automotive market is set to record a robust CAGR of 55% during 2023 and 2033. The market is anticipated to cross a value of US$ 744.39 billion by 2033.¡± Today, AI workloads are deployed in the cloud to meet the consumer demands for personalized, immediate, private, and reliable experiences in the car. In the consumer world however, the need is to help process these workloads at an even faster pace. For that reason, data processing is now taking place closer to the source, complementing the capabilities of the Cloud. In the Automotive industry, AI-enabled automotive devices and systems are dramatically transforming the way SoCs are designed, making high-quality and reliable die-to-die and chip-to-chip connectivity non-negotiable.  

The automotive industry is adopting a new electronic/electric (EE) architecture where a centralized compute module executes multiple applications such as ADAS and in-vehicle infotainment (IVI). With the advent of EVs and more advanced features in the car, the new centralized zonal architecture will help minimize complexity, maximize scalability, and facilitate faster decision-making time. This new architecture is demanding a new set of SoCs on advanced process technologies with very high performance. More traditional monolithic SoCs for single functions like ADAS are giving way to multi-die designs where various dies are connected in a single package and placed in a system to perform a function in the car. While such multi-die designs are gaining adoption, semiconductor companies must remain cost conscious as these ADAS SoCs will be manufactured at high volumes for a myriad of safety levels.  One example is the automated driving central compute system. The system can include modules for the sensor interface, safety management, memory control and interfaces, and dies for CPU, GPU, and AI accelerator which are then connected via a die-to-die interface such as the Universal Chiplet Interconnect Express (UCIe). Figure 1 illustrates how semiconductor companies can develop SoCs for such systems using multi-die designs. For a base ADAS or IVI SoC the requirement might just be the CPU die for a level 2 functional safety. A GPU die can be added to the base CPU die for a base ADAS or premium IVI function at a level 2+ driving automation. To allow more compute power for AI workloads, an NPU die can be added to the base CPU or the base CPU and GPU dies for level 3/3+ functional safety. None of these scalable scenarios are possible without a solution for die-to-die connectivity. 

The industry has come together to define, develop, and deploy the UCIe standard, a universal interconnect at the package-level. In a recent ¡°updates to the standard with additional enhancements for automotive usages ¨C such as predictive failure analysis and health monitoring ¨C and enabling lower-cost packaging implementations.¡± Figure 1 shows three use cases for UCIe. The first use case is for low-latency and coherency where two Network on a Chip (NoC) are connected via UCIe. This use case is mainly for applications requiring ADAS computing power. The second automotive use case is when memory and IO are split into two separate dies and are then connected to the compute die via CXL and UCIe streaming protocols. The third automotive use case is very similar to what is seen in HPC applications where a companion AI accelerator die is connected to the main CPU die via UCIe. 

OEMs are attracting consumers by providing the utmost in cockpit experience with high-resolution, 4K, pillar-to-pillar displays. Multi-Stream Transport (MTR) enables a daisy-chained display topology using a single port, which consists of a single GPU, one DP TX controller, and PHY, to display images on multiple screens in the car. This revision clarifies the components involved and maintains the original meaning. This daisy-chained set up simplifies the display wiring in the car. Figure 4 illustrates how connectivity in the SoC can enable multi-display environments in the car. Row 1: Multiple image sources from the application processor are fed into the daisy-chained display set up via the DisplayPort (DP) MTR interface. Row 2: Multiple image sources from the application processor is fed to the daisy-chained display set up but also to the left or right mirrors, all via the DP MTR interface. Row 3: The same set up in row 2 can be executed via the MIPI DSI or embedded DP MTR interfaces, depending on display size and power requirements. 

An alternate use case is USB/DP. A single USB port can be used for silicon lifecycle management, sentry mode, test, debug, and firmware download.  USB can be used to avoid the need for very large numbers of test pings, speed up test by exceeding GPIO test pin data rates, repeat manufacturing test in-system and in-field, access PVT monitors, and debug.

The transformation to software-defined vehicles marks a significant shift in the automotive industry, bringing together highly integrated systems and AI to create safer and more efficient vehicles while addressing sophisticated user needs and vendor serviceability. New trends in the automotive industry are presenting opportunities for innovations in ADAS and IVI SoC designs. Centralized zonal architecture, multi-die design, daisy-chained displays, and integration of ADAS/IVI functions in a single SoC are among some of the key trends that the automotive industry is tracking. Synopsys is at the forefront of automotive SoC innovations with a portfolio of silicon-proven automotive IP for the highest levels of functional safety, security, quality, and reliability. The IP portfolio is developed and assessed specifically for ISO 26262 random hardware faults and ASIL D systematic. To minimize cybersecurity risks, Synopsys is developing IP products as per the ISO/SAE 21434 standard to provide automotive SoC developers a safe, reliable, and future proof solution.