NLOG-283 | Semiconductor And Beyond Newsletter | The Semiconductor Laws

Chetan Arvind Patil
December 21, 2025

The Semiconductor Laws

The semiconductor industry has progressed with purpose, not by chance. For over five decades, consistent empirical laws and scaling principles have defined how chips are designed, manufactured, powered, and evaluated economically. These principles have served as the backbone of the industry's evolution.

These laws became operating assumptions for technology roadmaps, capital planning, and product strategy. Foundries, fabless companies, equipment suppliers, and system designers aligned their expectations around them.

For a long time, these laws reinforced one another, creating a virtuous cycle of performance improvement, cost reduction, and integration.

Today, many of these laws are strained, bent, or partially broken, altering the pace and direction of innovation, affecting economic models, and forcing engineers and executives to reconsider design, manufacturing, and long-term investment decisions.

Yet these laws still define the constraints within which modern semiconductor systems must operate. Understanding them remains essential to understanding why the industry looks the way it does today.

Moore’s Law And Density Driven Scaling

Moore’s Law states that the number of transistors on an integrated circuit doubles about every two years. This single observation shaped the semiconductor industry more than any other principle. Higher transistor density enabled more functionality per chip, lower cost per function, and predictable performance improvement across generations.

For decades, density scaling translated directly into economic advantage. Each technology node delivered smaller transistors, higher yields at scale, and expanding design freedom.

As a result, Moore’s Law became less about physics and more about coordinated efforts in design, manufacturing, and supply chains.

However, while transistor density continues to improve, the relationship between scaling and cost has weakened. Advanced nodes now demand disproportionate capital, process complexity, and energy. Moore’s Law still exists, but its economic certainty no longer does.

Dennard Scaling And The Power Balance

Technology advanced rapidly, in part due to Dennard Scaling, which stated that as transistors shrink, voltage and current decrease proportionally, keeping power density constant.

This proportional scaling enabled higher frequencies without exceeding thermal limits, which sustained performance improvements.

This balance ended when voltage scaling stalled. Rising leakage currents and slower voltage reductions sharply increased power density. Frequency scaling plateaued, and thermal limits became key design constraints.

Dennard Scaling’s breakdown caused a structural shift. Faster clocks could no longer drive performance growth. Power became the main constraint, reshaping architectures, floorplans, and roadmaps.

Kim’s Law And 3D Scaling

Kim’s Law describes an empirical trend seen in three-dimensional integrated circuits. The number of stacked layers doubles about every two years. First articulated by Professor Joungho Kim at KAIST’s TeraLab, it extends semiconductor scaling beyond planar transistor density. It now includes vertical integration, which has become crucial as traditional Moore-style scaling slows.

In memory technologies such as High Bandwidth Memory and advanced 3D NAND, increasing the number of stacked layers directly improves bandwidth and capacity. This is achieved without shrinking individual transistors. Early architectures started with a few stacked dies. Now, there are dozens of layers, making vertical scaling the main driver of system performance.

Kim’s Law reflects a shift driven by necessity. AI and high-performance computing workloads now demand more memory bandwidth and density. Planar scaling alone is no longer sufficient. Vertical stacking, enabled by through-silicon vias, interposers, and advanced packaging, is now essential for continued system-level scaling. It defines a new axis of progress alongside Moore’s and Dennard’s laws.

Koomey’s Law And Energy Efficiency

As frequency scaling slowed, efficiency improvements continued. Koomey’s Law states that computations per joule approximately doubled every 1.5 years. This reflected advances in microarchitecture, power management, process optimization, and workload specialization.

Efficiency gains enabled progress even when raw performance gains slowed. They justified the rise of accelerators, domain-specific architectures, and heterogeneous compute models.

However, efficiency improvements themselves are becoming harder to sustain as data movement and system-level losses dominate.

The focus has shifted from compute efficiency alone to system efficiency across memory, interconnect, and packaging.

Rent’s Rule And Interconnect Pressure

Rent’s Rule relates the number of external connections of a logic block to its internal complexity. As integration increased, interconnect became a limiting factor.

As a result, routing congestion, signal integrity, latency, and power associated with data movement grew faster than logic capability.

Consequently, this rule explains why scaling challenges increasingly appear outside the transistor. Wires, not gates, dominate delay and energy. Packaging, not lithography alone, defines performance limits at advanced nodes.

Therefore, Rent’s Rule is a primary reason for the industry’s focus on chiplets, advanced packaging, and 3D integration as solutions to overcome interconnect limitations.

Pollack’s Rule And Diminishing Returns

Pollack’s Rule states that performance increases roughly with the square root of complexity. Doubling complexity does not double performance.

This insight exposed the inefficiency of ever-larger monolithic cores.

As complexity grew, returns diminished while power and verification costs exploded. Consequently, the industry responded by shifting toward multicore designs, specialization, and parallelism.

Because of this, Pollack’s Rule remains central to modern architecture decisions. It reinforces the move away from brute-force scaling toward balanced, workload-aware designs.

Power Wall And Memory Wall

The Power Wall reflects the inability to increase performance without exceeding power and thermal limits.

The Memory Wall highlights the growing gap between processor speed and memory access latency.

As a result of these limitations, system design underwent significant changes. Cache hierarchies deepened, memory technologies diversified, and near-memory and in-memory computing concepts emerged. Performance became increasingly constrained by data access rather than compute.

Ultimately, these interrelated walls explain why system architecture and packaging are now as critical as silicon process technology.

Closing Perspective

The semiconductor laws explain not only how the industry grew, but why it now looks fundamentally different from its past.

Scaling is no longer automatic. It must be engineered deliberately across architecture, manufacturing, and systems.

Those who understand where these laws still apply, and where they no longer do, will be best positioned to navigate the next phase of semiconductor innovation.

CONNECT

Whether you are a student with the goal to enter semiconductor industry (or even academia) or a semiconductor professional or someone looking to learn more about the ins and outs of the semiconductor industry, please do reach out to me.

Let us together explore the world of semiconductor and the endless opportunities:

And, do explore the 300+ semiconductor-focused blogs on my website.

NLOG-283 | Semiconductor And Beyond Newsletter | The Semiconductor Laws

The Semiconductor Laws

Subscribe Now To Continue Reading

Reply

Keep Reading

Chetan Arvind Patil

Home