In Korean

Large language models (LLMs), the artificial intelligence systems behind products such as ChatGPT and Bard, could soon revolutionize our interactions with autonomous vehicles.

LLMs explore to merge text with visual data and perform tasks like image classification, text-to-image retrieval and visual question answering. Therefore we can understand the basic LLM knowledges & the potential of LLMs far beyond text and discuss what we need to consider at the effective LLM SoC design through this session.

Automotive electronics design is both demanding from market and challenging from technology perspective. Solving both safety and reliability challenges with reasonable and competitive PPA is key to success. This presentation talks about FuSa implementation challenges and how Cadence solve the problems.

In Korean
Introducing the Helium SW interface on Automotive – digital twin and cloud development environment

As infotainment systems evolve, they must handle diverse workloads, including audio, voice, speech, AI, and vision. This necessitates a flexible and scalable Digital Signal Processor (DSP) solution that can serve as an offload engine for the main application processor. Using a single DSP for various workloads presents a cost-effective solution for high-performance, low-power processing, aligning perfectly with Electric Vehicle (EV) requirements.

The latest DSP introduces several features that significantly enhance performance, including Double Precision Floating Point Unit (FPU): This feature improves audio performance and Signal-to-Noise Ratio (SNR) by leveraging 64-bit processing over 32-bit. It enables the management of more than 32 speakers without performance degradation. Improved AI Capabilities can perform Engine/Road noise reduction, shaping the audio output to highlight voice and dialogues, and adapting the cabin soundfield, contributing to a quieter and more enjoyable driving experience. These advancements position the latest DSP as a powerful tool in the evolution of infotainment systems, delivering superior performance and versatility.

Arm and Cadence present how our collaboratively designed chiplet reference platform for ADAS systems will jumpstart collaboration and innovation in the automotive ecosystem. Our collaboration is founded on the confluence of two seemingly contradictory industry trends: the move towards high-performance centralized computation for software-defined vehicles and the disaggregation of computation into discrete chiplets that can be independently verified and certified. The centralization is driven by the increased need for efficiency and fuel economy while autonomy and electrification drive the need for high performance. Simultaneously, the industry needs to maintain and improve safety certification and product quality, which are often in conflict with aggregated complexity. A multi-chipset-based architecture including high-performance compute chiplets and lower complexity safety certified chiplets achieves both goals while enabling ecosystem members to innovate through the custom accelerator and application-specific chiplets.

Arm’s recently announced Automotive Enhanced IP portfolio provides both the high-performance and safety-certified cores necessary to give this design a firm foundation. Cadence interface IPs provide the backbone for certifiable interfaces between chiplets, and configurable Cadence Tensilica AI accelerators provide a starting point for innovation. To present each chiplet in the design as both a fully enabled tapeout-ready example and as a reference design ready for customization in Cadence design tools the ADAS system serves as a jumping-off point of innovation within the ecosystem. This flexibility enables ecosystem members to select some chiplets as early representations of products to purchase from a silicon supplier while focusing on other chiplets as the basis for their innovation. The reference implementations will also be available for interoperability testing without adopting full production licensing.

Automotive architectures continue to evolve with the current focus on software-defined vehicles and the demand for single SoCs and a common software architecture for central compute, zonal architectures, In-cabin sensing and digital cockpit. Automotive vendors are also designing for feet off to hands off to eventually eyes off. These trends are driving the need for cars to have a large number of sensors, as well as different types of sensors such as camera, radar and lidar. Driver and occupancy monitoring systems and privacy concerns are also driving the need for different cameras and radar sensors inside the car. In the past, specialized and discrete processors were used to handle image, radar and lidar data independently. However, these requirements have since evolved and the need for a singular and unified processor has become apparent. Automotive vendors are also developing dedicated SOCs for infotainment to offer multi-channel audio and voice input. Aside from the classical approaches, AI is now also becoming prevalent for radar and sensor fusion applications. In this presentation, we plan to cover Tensilica's answer for such requirements.

In the face of device scaling limitations at advanced nodes, the demand for compute performance and data transfer in hyperscale data centers and AI designs has reached unprecedented levels. As advanced systems-on-chip (SoCs) approach reticle size limits, innovative solutions to sustain Moore's law scaling and achieve performance enhancements with reduced power have become paramount. The challenges of SI/PI and Antenna/RF synthesis problems, which often necessitate electromagnetic simulation, are well-known. Each simulation run can consume several hours per candidate solution, and the design process often involves numerous evaluations using traditional evolutionary-based algorithms. These factors contribute to the computational expense and low efficiency of EM simulations, often requiring several weeks for optimization. This paper presents a convergence of electronic and system design automation, focusing on multi-physics analysis, electrical, thermal, and mechanical, from chip to systems, offering a novel approach to these challenges. A promising and practical approach to modeling and optimization for systems with limited data and computational resources, providing tangible solutions to these time-consuming and resource-intensive processes.

The automotive Industry is witnessing significant transformation with increasing focus on electrification globally. At the same time, OEMs and suppliers are facing significant challenges to reduce product development cost and timing while dealing with increasing technological complexity. This requires adopting new and innovative simulation workflows that deliver fast and reliable results at lower cost and improved efficiency. The presentation will highlight Cadence’s Fidelity platform and highlight the key advantages and differentiators it offers to automotive simulation engineers across pre-processing and simulation. This includes Millennium, Cadence’s state of the art GPU compute platform as well our Cadence’s vision to make AI/ML based solutions a reality for product development. Recent validation examples, customer testimonials and product roadmap specific to automotive applications will be highlighted.

In Korean

In the modern automotive industry, there is a growing consumer demand for AI services. Moreover, passengers often want to use different AI services individually, which necessitates clear identification of the speaker. However, installing microphones at each seat to address this issue can lead to sound interference and an increase in wiring, posing a burden on car manufacturing. This study introduces a system that selectively captures and transmits the voice of the speaker using a minimal number of microphones. We present the performance demonstration results of a chipset based on the HiFi5 system.