Beyond the 10 Ångström Barrier: The New Era of 3D Transistors and Workload-Aware Silicon

Semiconductor scaling is moving beyond the transistor as the industry embraces advanced packaging, photonics, and CFET technology to extend CMOS scaling beyond 10 Å. These innovations are essential for meeting the power and performance demands of the AI era.

Share
Beyond the 10 Ångström Barrier: The New Era of 3D Transistors and Workload-Aware Silicon

The semiconductor industry is entering a "post-transistor" era where traditional Moore’s Law scaling is no longer sufficient. At recent industry summits, the conversation has shifted toward a holistic view of chip design, integrating advanced packaging, silicon photonics, and new architectural paradigms like Complementary Field-Effect Transistors (CFET). These technologies are designed to extend CMOS scaling beyond the 10 Ångström (1 nm) barrier, a threshold once thought to be the physical limit of silicon.

CFET technology, in particular, offers a way to stack n-type and p-type transistors vertically rather than side-by-side. This configuration delivers higher performance through larger effective channel widths and improved electrostatics while significantly reducing the footprint on the die. However, this verticality introduces new challenges in power delivery and thermal management. Engineers are now turning to stacked-die multi-chip architectures and integrated voltage regulators to manage current density and ensure stability across these dense 3D structures.

Furthermore, the industry is moving toward "PVTW" characterization—extending silicon testing from standard Pressure, Voltage, and Temperature (PVT) to include real-world Workload (W). This shift acknowledges that two identical chips may behave differently under the intense, erratic demands of AI processing. By combining richer metrology with explainable AI, manufacturers can optimize these next-generation chips for specific high-performance computing tasks, ensuring that the hardware can keep pace with the exponential growth of software requirements.


Source: Semiconductor Engineering