For decades, the semiconductor industry followed a predictable cadence: shrink the physical dimensions of silicon field-effect transistors, and integrated circuits would automatically become faster, cheaper, and more power-efficient. This dual promise of Moore's Law (the doubling of transistor density) and Dennard scaling (maintaining constant power density) formed the bedrock of modern computing.
However, as feature sizes approach the atomic scale, this scaling paradigm hits hard physical, thermal, and economic walls. As Matsuoka et al. observe, the cessation of these foundational scaling laws has disrupted traditional industry trajectories. Understanding why complementary metal-oxide-semiconductor (CMOS) scaling is reaching its limits requires looking at the underlying physics of sub-nanometer devices.
The Thermal and Leakage Walls
In a classical transistor, the gate acts as an electrostatic switch controlling the flow of electrons through a conductive channel. To maintain tight control over this channel as its length shortens, the insulating gate oxide layer must also become thinner. At atomic dimensions (approaching the sub-5 nm regime), this layer becomes so thin that classical physics no longer describes electron behavior.
Instead, quantum mechanics dominates. Electrons begin to spontaneously tunnel through the physical potential barrier of the gate oxide and across the channel even when the transistor is explicitly turned "off" (Krzakala et al.; von Lerber et al.).
Classical Behavior (Thick Gate Oxide):
[Gate] ---> Electrostatic Field Controls Channel
==============================================
[Source] ---------- Electrons -----------> [Drain]
Quantum Behavior (Atomic-Scale Gate Oxide):
[Gate]
| \--- Quantum Tunneling Leakage
==============================================
[Source] ---- Tunnels Through Barrier ----> [Drain]
This quantum tunneling causes a massive spike in leakage current (Bishnoi et al.). Instead of consuming power primarily during switching operations (dynamic power), transistors continuously bleed energy while idle (static power). The resulting heat dissipation creates a thermal ceiling; chips cannot be packed more densely or clocked faster without generating destructive thermal loads (Krzakala et al.; Zhang et al.).
Diminishing Returns of Advanced Geometries
To combat leakage and maintain electrostatic control, the semiconductor industry transitioned from planar transistors to complex three-dimensional architectures, such as FinFETs and Gate-All-Around (GAA) nanosheets. However, as highlighted by Chen, these advanced geometries face severely diminishing returns as they approach the 1 nm process node.
When channel lengths become extremely short, short-channel effects take over. The drain voltage begins to heavily influence the channel source, lowering the potential barrier and making it difficult to turn the device completely off. Consequently, reducing the operating voltage required for switching via electric fields becomes increasingly difficult (Domingo et al.). This creates a fundamental divergence between actual planar transistor density and historical scaling predictions, signaling that two-dimensional miniaturization is no longer sufficient (Chen).
The Interconnect and Complexity Bottlenecks
The scaling crisis is not confined to the transistors themselves; the metallic wires connecting them (interconnects) present an equally daunting physical barrier. As circuits shrink, the cross-sectional area of these copper or tungsten interconnects decreases, which exponentially increases their electrical resistance.
This leads to severe resistive losses and signal propagation delays (Kazanskiy et al.). Signals take longer to travel across the chip, and parasitic interference between closely packed lines degrades signal integrity (Zhang et al.).
Furthermore, this explosion in component density and non-ideal physical behavior causes a scalability crisis in Electronic Design Automation (EDA) tools. Legacy simulation methods struggle to model the highly complex, non-linear interactions of timing, congestion, and reliability at advanced nodes, making the design and verification process computationally prohibitive (Knechtel et al.).
The Post-CMOS Paradigm Shift
Because traditional dimensional shrinking ("More Moore") is hitting physical and economic limits (Sinanoglu et al.), the research community is shifting focus toward alternative materials and architectural paradigms to sustain computing growth:
- Alternative Materials: Researchers are exploring atomically thin 2D materials like Indium Oxide ($In_2O_3$) (Xu et al.), black phosphorus (Ye et al.), and MXenes (Subedi et al.) to achieve higher carrier mobility and superior electrostatic control at atomic thicknesses.
- Volumetric Integration: Shifting from 2D planar layouts to 2.5D and 3D heterogeneous integration allows designers to stack components vertically, increasing functional density without relying on dimensional scaling (Chakrabarti et al.; Sinanoglu et al.).
- Alternative Substrates: Emerging paradigms seek to bypass electronics entirely. Integrated photonics utilize light instead of electrons to eliminate resistive losses and propagation delays (Kathirvelu et al.; Kazanskiy et al.), while memristive architectures merge memory and logic to break the von Neumann bottleneck (Ye et al.; Mukhopadhyay et al.).
As Pan notes, these emerging nanoelectronic and nanomagnetic paradigms require entirely new design guidelines and abstractions, as current EDA frameworks remain tightly coupled to the fading rules of classical CMOS scaling.
Key Foundational Work
Theoretical Studies of Mixed-Valence Transition Metal Complexes for Molecular Computing (2002) — contributed 1 concept elements to this direction.
Electrical Conduction in 10 nm Thin Polysilicon Wires from 4 to 400 K and Their Operation for Hybrid Memory (2006) — contributed 1 concept elements to this direction.