Speaking at the Auto EV Tech Vision Summit 2025, Namrta Sharma, Technical Director at Aritrak Technologies, highlighted how chiplet architectures are emerging as a crucial enabler for the next generation of automotive semiconductors. As vehicles transition toward software-defined platforms, Sharma emphasized that semiconductor design must evolve to meet unprecedented computational demands while balancing cost, scalability, and time-to-market.
Setting the context, Sharma described the ongoing transformation in the automotive sector. “We should all agree that we are actually going through a metamorphosis,” she said, adding that the automotive industry is currently experiencing “a big transformation of the century.” Vehicles are no longer defined solely by mechanical and electrical components. Instead, the modern car is increasingly becoming a software-driven platform.
Computational Requirements at Rise in Automotives
“The car is no longer just mechanical and electrical,” Sharma noted. “The car is a software-defined vehicle. So, the role of the semiconductors in the car is also changing. It is no longer just adding to some features. It actually defines the car. It is the core intelligence of the car.” This shift is dramatically expanding the semiconductor requirements inside vehicles. Modern automotive electronics must support electrification, battery management, in-vehicle infotainment, constant connectivity, and advanced driver assistance systems (ADAS), while also laying the groundwork for autonomous driving.
Sharma highlighted how the computational requirements behind these capabilities have grown exponentially. “If you see the numbers for the compute, Level-2 ADAS required just 10 TOPS, which is 10 trillion operations per second, which was just a decade ago,” she explained. “But now the requirement is about 1,000 TOPS for full autonomy. So it is like a 100x gap.”
Limitations at Play
Meeting such performance requirements using conventional monolithic chip design is becoming increasingly difficult. Integrating CPUs, GPUs, communication circuits, and power management components into a single system-on-chip (SoC) results in extremely large semiconductor dies. However, manufacturing such large chips is constrained by physical and economic limits. “There is a limit. It is called the reticle limit,” Sharma explained. “That is the biggest size that we can manufacture in a foundry, and it is about 850 mm² as of now.”
Large monolithic chips also face yield challenges. A defect in even a small portion of a large die can render the entire chip unusable, leading to significant losses in manufacturing yield and rising costs. To address these issues, the semiconductor industry has increasingly turned to chiplet architectures. “The solution was simple,” Sharma said. “Just cut this big die into multiple dies. These are all small functioning blocks, and these are called chiplets.”
Chiplets: A Yield Optimisation Solution
By breaking down a large chip into smaller modular dies, chiplet architectures help overcome reticle limitations and improve yield. If a defect occurs, only the affected chiplet is discarded rather than the entire system. This modular approach has already gained traction in high-performance computing systems and is now finding relevance in automotive electronics.
Sharma also pointed out that chiplets enable a shift toward heterogeneous integration. Instead of manufacturing every component using the most advanced—and expensive—process node, different chiplets can be fabricated using technology nodes optimized for their specific functions. “For example, the CPU needs the highest compute, so it is in the latest technology node,” she explained. “But the other things, like sensors or memory engines, need not be.”
Chiplets Optimised Time to Market
Beyond cost and yield advantages, chiplets significantly improve time-to-market. Sharma emphasized that modular architectures allow semiconductor companies to reuse proven components and focus their efforts on product differentiation. “You need not design the whole chip, you need not design the whole SoC,” she said. “You can just design your differentiating chiplet.” This flexibility also allows manufacturers to scale systems quickly for different vehicle segments.
Performance levels can be adjusted simply by replacing or modifying individual chiplets, enabling rapid customization without redesigning the entire architecture. However, Sharma cautioned that the transition to chiplet-based systems introduces new challenges. With multiple chiplets integrated into a single package, design complexity shifts from the chip level to the system level. This requires advanced electronic design automation (EDA) tools capable of co-optimizing silicon, packaging, and interconnect technologies.
Testing & Standards
With this, testing and validation also become more complex. The success of chiplet integration depends on ensuring that each component integrated into the system is a “known good die.” As Sharma noted, this requires new testing methodologies and infrastructure capable of validating chiplets both individually and within the larger system.
Another key factor in the long-term success of chiplets is the development of industry-wide standards. Sharma highlighted the importance of emerging interconnect standards such as Universal Chiplet Interconnect Express (UCIe), which aim to enable interoperability between chiplets from different vendors. As the ecosystem evolves, Sharma believes collaboration across the semiconductor value chain will play a critical role. Foundries, design houses, EDA companies, substrate providers, and industry consortia are already working together to establish the standards and infrastructure needed to support chiplet-based systems.
Conclusion
Summarizing her key message, Sharma emphasized that chiplet architectures are not just about cost optimization. Instead, they represent a fundamental shift in how semiconductor systems are designed for rapidly evolving markets like automotive. “Chiplet heterogeneous integration provides not only the cost benefit,” she concluded, “it also provides the speed of execution—the speed to make changes fast and react to innovation.”
As automotive electronics continue to grow in complexity and performance requirements, chiplets may well become the architectural foundation enabling the next wave of innovation in software-defined vehicles.
