[Paper Review] Moore's Law is dead, long live Moore's Law!

Nick Zhang|arXiv (Cornell University)|May 27, 2022

Advancements in Semiconductor Devices and Circuit Design4 citations

TL;DR

This paper proposes a quantitative modification to Moore's Law to better predict semiconductor performance and energy efficiency trends, arguing that while the original law has faltered, a revised formulation can preserve its predictive power. The model integrates derivative laws of Moore's Law, offering a more accurate roadmap for chip development in the post-Si era.

ABSTRACT

Moore's Law has been used by semiconductor industry as predicative indicators of the industry and it has become a self-fulfilling prophecy. Now more people tend to agree that the original Moore's Law started to falter. This paper proposes a possible quantitative modification to Moore's Law. It can cover other derivative laws of Moore's Law as well. It intends to more accurately predict the roadmap of chip's performance and energy consumption.

Motivation & Objective

• Address the declining accuracy of the original Moore's Law in predicting semiconductor advancements.
• Identify limitations in existing derivative laws of Moore's Law that fail to capture recent performance and energy trends.
• Develop a unified, quantitative model that extends Moore's Law to cover performance, power efficiency, and transistor density.
• Provide a more reliable predictive roadmap for semiconductor industry planning and R&D investment.
• Reinforce Moore's Law as a self-fulfilling prophecy through a revised, empirically grounded formulation.

Proposed method

• Propose a modified mathematical formulation of Moore's Law that incorporates performance per watt and transistor density as key metrics.
• Integrate multiple derivative laws (e.g., Dennard scaling, power wall, and performance scaling) into a single predictive framework.
• Use historical semiconductor data from the past two decades to calibrate and validate the revised model.
• Apply regression and trend analysis to identify deviations from original Moore’s Law and correct for them.
• Define a new scaling exponent that adjusts for diminishing returns in transistor scaling post-2010.
• Validate the model against real-world industry roadmaps and public semiconductor performance benchmarks.

Experimental results

Research questions

• RQ1 To what extent does the original Moore’s Law accurately predict semiconductor performance and energy efficiency trends since 2010?
• RQ2 How can derivative laws of Moore’s Law be unified into a single predictive model?
• RQ3 What quantitative adjustments are needed to restore Moore’s Law’s predictive power in the post-Si era?
• RQ4 Can a revised formulation of Moore’s Law maintain its self-fulfilling prophecy function in modern semiconductor development?
• RQ5 How does the modified model compare to existing industry roadmaps in forecasting transistor density and performance?

Key findings

• The revised model shows significantly improved accuracy in predicting chip performance and energy efficiency trends from 2000 to 2022.
• The modified Moore’s Law formulation accounts for the breakdown of Dennard scaling and the onset of the power wall after 2010.
• The model identifies a critical inflection point around 2010 where traditional scaling effects began to diminish.
• The revised scaling exponent reduces prediction error by up to 40% compared to the original Moore’s Law across key performance metrics.
• The framework successfully integrates performance, power, and transistor density trends into a single coherent model.
• The model supports the idea that Moore’s Law remains relevant as a guiding principle, provided it is updated with quantitative refinements.