Abstract: The theoretical and practical aspects of the fundamental, ultimate, physical limits to scaling, or Moore's law, is presented.
TL;DR
Moore’s Law is often discussed as a manufacturing challenge, but this paper explores the hard physical "brick walls" dictated by thermodynamics and quantum mechanics. By accounting for quantum tunneling and the energy costs of cooling, the author predicts we will hit the ultimate scaling limits of traditional electronic devices within the next 30 to 40 years.
What Problem Does This Paper Address?
For decades, the semiconductor industry has followed Moore’s Law—the observation that the number of transistors on a chip doubles roughly every two years. However, as transistors approach the atomic scale, we run into fundamental physical constraints that no amount of engineering ingenuity can bypass.
The paper seeks to define these ultimate limits. It moves beyond simple manufacturing hurdles to ask: What is the minimum energy required to switch a bit? How small can a feature be before an electron simply tunnels through it? And does cooling a chip actually help us reach these limits, or does it make the problem worse?
Methods
The author employs a theoretical framework built on three pillars of physics:
- Thermodynamics: Using the Shannon-von Neumann-Landauer (SNL) expression to determine the minimum energy ($k_B T \ln 2$) required to irreversibly change a bit of information.
- Quantum Mechanics: Applying Heisenberg’s Uncertainty Principle to find the minimum switching time and feature size, and incorporating tunneling probability formulas to account for "leakage" through barriers.
- Heat Transfer: Using Newton’s Law of Cooling and Carnot’s theorem to calculate the total energy cost of a system, including the work required to remove heat from the device.
The model considers a bit as a particle in a potential well. To switch states, the particle must overcome or tunnel through a barrier of width a and height $E_b$, as shown in the diagram above.
Key Insights and Ideas
- The "Cooling Trap": A common assumption is that cooling a system improves performance. However, the author uses Carnot’s theorem to show that the work required to remove heat ($W$) increases as the device temperature drops. Consequently, the total energy cost ($E_{bit}^{total}$) actually rises at lower temperatures.
- The Tunneling Correction: Standard thermodynamic limits suggest we can go very small, but quantum tunneling creates a "floor." To keep states distinguishable (i.e., to keep the error rate low), the energy barrier must be raised as the feature size ($a$) decreases.
- Temperature Paradox: Interestingly, at extremely small feature sizes (sub-2nm), energy and power requirements are "better behaved" at higher temperatures. As seen in the 3D plot above, at low temperatures (blue regions), the quantum tunneling effects dominate so aggressively that the energy required to maintain a stable bit diverges rapidly.
- The Compton Limit: The author points to the Compton wavelength of an electron (~0.00243 nm) as a potential absolute physical limit for any electronic device, noting that power requirements diverge to infinity as we approach this scale.
Key Results
- Minimum Energy: At 300 K, the absolute minimum energy to process a bit is approximately 0.017 eV.
- Physical Dimensions: Based on the uncertainty principle, the minimum feature size for an electron carrier is roughly 1.5 nm.
- Switching Speed: The fundamental limit for switching time is approximately 0.04 ps.
- Power Density: The theoretical power density at these limits is roughly 3.7 MW/cm². For perspective, the surface of the Sun is about 6,000 W/cm², meaning these chips would be orders of magnitude more "intense" than a star's surface.
- The Horizon: The author estimates that we will hit these fundamental scaling limits in 30–40 years (from the 2015 publication date).
Limitations and Future Work
The paper primarily focuses on the limits of traditional charge-based CMOS-like devices. While it mentions that practical clock speeds have already saturated at ~3 GHz due to "parasitic" factors (interconnects and architecture), it does not dive deep into how 3D integration or new materials might delay the inevitable.
For future directions, the author suggests that we may need to move beyond traditional laws of computation. He points toward "locally active components" like memcapacitors and meminductors—theoretical components that could potentially allow for computing with significantly lower power consumption, though these remain "fantastic predictions" rather than current realities.
Summary updated