Abstract: In October 2023, Fan Chen, an Associate Editor at IEEE Circuits and Systems Magazine, had the privilege of interviewing Dr. Nicky Lu, an icon in IC design and the semiconductor industry. This engaging interview covers Dr. Lu’s educational journey, illustrious career, perspectives on research, innovation, entrepreneurship, and his forward-looking vision, along with invaluable advice for the younger generation. We’re confident our CAS readers will find this conversation both enlightening and enjoyable!
Introduction
Dr. Nicky Lu, a seminal figure in the semiconductor industry and founder of Etron Technology, reflects on a 46-year career that transitioned the world from vacuum tubes to 3D transistors. The interview outlines his pivotal role in inventing 3D trench DRAM and pioneering "Heterogeneous Integration"—a paradigm shift he calls "Silicon 4.0" that aims to sustain the spirit of Moore’s Law through volumetric density rather than simple planar scaling.
The Semiconductor Density Gap
The semiconductor industry is currently facing a "density gap." While Moore’s Law historically dictated a doubling of transistor density with each node, recent advancements in process technology have begun to deviate from this trend. As we approach the physical limits of 1nm nodes, the industry faces challenges in heat dissipation, economic viability, and the slowing of traditional planar scaling.
As shown in Table 1, quantitative comparisons reveal that at the 2nm node, current industry leaders achieve approximately 333 MTr/mm², whereas a strict adherence to Moore’s Law would require roughly 1848 MTr/mm². Lu argues that while we are failing to meet Moore’s Law in terms of planar area, we can "catch up" by measuring peak transistor densities by volume, shifting the metric to MTr/mm³.
The Silicon 4.0 Framework
To overcome these hurdles, Dr. Lu proposes shifting focus from "Line Scaling" (Silicon 1.0) to "Heterogeneous Integration" (Silicon 4.0). In this era, diverse components—logic, memory, RF, and even non-silicon materials—are integrated into a single, high-performance system-in-package.
Heterogeneous Integration (HI), a term coined by Lu in 2004, is presented as the defining era of Silicon 4.0. It allows the industry to bypass the limitations of a single monolithic die by stacking specialized "Known Good Dice" (KGD). This "Function x Value Scaling" methodology integrates MEMS, optical devices, and non-silicon materials (such as GaAs and GaN) into silicon-centric systems.
Technical Methodologies and Historical Impact
Throughout his career, Dr. Lu employed several key technical strategies to drive innovation:
- 3D Trench-Capacitor Structure: Invented by Lu at IBM, this used the silicon substrate as a counter-electrode, enabling the world’s first 4Mb DRAM on 8-inch wafers.
- Known Good Die (KGD): A novel testing methodology Lu developed to conduct burn-in tests on individual dice without packaging. This allowed for the reliable integration of multiple dice into a single package, a precursor to modern chiplet architectures used in early mobile products and miniature hard drives.
- Flexible Foundry Rules: During Taiwan’s "National Submicron Project," Lu implemented design rules that allowed for forward migration to new process nodes while reusing existing equipment, significantly reducing development time and cost.
Future Outlook and Limitations
Despite the promise of Silicon 4.0, significant challenges remain. Thermal management is the primary bottleneck for stacked 3D dice configurations. Looking ahead, Lu predicts a major "computing power boom" around 2030 driven by the convergence of AI and quantum computing. This will require entirely new materials and a total rethink of transistor architecture beyond current FinFET or Gate-All-Around (GAA) structures as the industry marches toward the 1nm frontier. Lu identifies the pursuit of room-temperature superconductors as a long-sought-after phenomenon that AI-driven research may finally unlock.
Summary updated