In the realm of automotive technology, the fusion of silicon and software has become the new driving force, with silicon compute performance emerging as the new horsepower. However, the demand for exponential performance cannot be met by monolithic die alone. This has led to the rise of chiplets, heralding a paradigm shift in automotive compute.

While performance and architectures parallel those of High-Performance Computing (HPC), the unique mission profile of automotive applications poses distinct challenges such as heat, vibration, safety, and reliability (QnR). Collaboration and trust across the industry are imperative to overcome these challenges and foster an open ecosystem.

Evolution of Automotive Computing:

The journey towards automotive chiplets began with the humble microcontroller unit (MCU). Over time, the demand for more sophisticated functionalities led to the development of new software-defined vehicles SDV architectures.

SDV separates hardware from software and these centralised compute architectures required new advanced 2D and 2.5D multi-chip packages (MCPs). They combined multiple CPU, Comms, Memory, and Accelerator die into a single package to achieve the compute needed for Advanced Driver Assistance Systems (ADAS) / next generation In-Vehicle Infotainment (IVI) and Autonomous Driving systems.

The Rise of Chiplets:

Chiplets offer a multitude of benefits, empowering designers to push beyond the limits of traditional monolithic designs:

• Flexibility and Scalability: Chiplets transcend reticle limitations, offering scalable solutions with improved thermal management.
• Cost and Yield Optimisation: By leveraging smaller die sizes and older process nodes, chiplets reduce design costs and enhance yield.
• Early Market Entry: Chiplets provide a strategic advantage, enabling early market entry, prolonged product life cycles and a higher profit margin.

Challenges and Considerations:

However, the transition from monolithic SoCs to chiplets presents several challenges:

• Standardisation and Ecosystem Development: Establishing standardised interfaces and protocols is crucial for fostering an open chiplet ecosystem.
• Performance Optimisation: Ensuring performance across the chiplet system and managing data flow efficiently are essential.
• Return on Investment (RoI): Uncertainties regarding RoI necessitate a thorough evaluation of non-recurring engineering (NRE) costs versus full factory costs (FFC).
• Market Adoption: The size of the automotive market for chiplets is small and alignment with other industries are key considerations for market uptake and improved RoI.

The Road Ahead:

Despite the promising potential of automotive chiplets, the journey towards widespread adoption is just beginning. Various industry stakeholders, from standardisation bodies to leading companies, are actively engaged in advancing chiplet technology. Some of these include;

• UCIe Auto Working Group
• Open Compute Project (OCP)
• IMEC ACA’s initiatives
• ASRA (Japanese Auto Makers)
• ACC (Chinese Chiplet Standard) and
• Industry players like Intel, Renesas, TI, Arm, Bosch, Denso…

Collaboration, alignment of interests, and a shared vision for the future across these companies and standard bodies are imperative to drive progress and overcome obstacles.

Conclusion:

While significant strides have been made in automotive chiplet development, we are still in the nascent stages of this transformative journey. Achieving success requires alignment across the automotive supply chain, identification of gaps, and joint efforts towards standardisation and collaboration.

 In the short term, automotive chiplets may predominantly emerge from vertically integrated companies, but true openness and standardisation will take time to materialise, promising a future of enhanced performance, efficiency, and innovation in automotive compute.

---

Copyright shall at all times remain vested in the Author. No part of the work shall be used, reproduced, stored in a retrieval system, or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise, without the Author’s express written consent.