Resurrections of Moore's Law

Introduction

Co-founder of Intel Corp, in 1965, predicted in a paper, that

The number of transistors in a dense integrated circuit (IC) doubles about every 12 Months.

Moore’s law has been adopted by the semiconductor industry and led it for more than a decade now. It has been adjusted depending on technological advancement. It is the law of economics thought to lead the semiconductor industry. We shall explore when the law failed and how it was resurrected by Industry.

Importance of Moore’s Law and roadmaps

Getting finished semiconductor products into the market is a complex business. Research and development is just one aspect of it. The major portion is the supply chain, manufacturing and IP contracts. There might be a CPU processor in a Research laboratory able to cross the 5 GHz clock frequency but getting it into the market can be impractical right now. The reason is, it has to be tested, auxiliary hardware has to be manufactured, and the foundaries should have the manufacturing technology to manufacture it to mass scale with the least defects in them. Having roadmaps helps orchestrate state-of-the-art hardware manufacturing.

Moore’s law has been used as a roadmap by the semiconductor industry to this date and achieved significant success in doing so. Since innovation in semiconductor hardware acts as enabling technological innovation in other fields, Moore’s law is in the news every year. The industry calls it alive while scientists call it dead.

The First Ressurection

In 1975, Moore had to correct the prediction of doubling transistors from 12 months to 18 months and then to 2 years.

The Second Ressurection

Around 2005, Intel was working on a single-core CPU called Tejas and Jayhawk. They were supposed to work at a clock frequency greater than 5GHz. Intel had to shut down the project because managing heat dissipation become impossible. Intel then kept Moore’s law alive by adopting multi-core architecture.

The Third Ressurection

In 2015, ITRS(International Technology Roadmap for Semiconductors), the body that gave a roadmap for keeping Moore’s law alive declared it dead. The invention of FinFET(Fin field-effect transistor) at a gate length of 14nm saved Moore’s law. However, a new body, rebooting computing at IEEE was formed in order to make a new Moore’s law that will help the semiconductor industry work in tandem.

Several news outlets marked its death.

https://arstechnica.com/information-technology/2016/02/moores-law-really-is-dead-this-time/

The Fourth Ressurection

In 2022, post-pandemic, the supply-chain issue was dominating in semiconductor industry. Though smartphone sales are increasing, the demand for consumer-grade computers is going down. There was a huge demand for GPUs and ASIC chips for mining crypto-currencies. But it has gone down after countries banned it to meet their carbon-neutral agenda. The falling price of crypto made the illegal miners leave the market as well. The semiconductor industry has to decide if the innovation will be in a new kind of computing device that can be coupled with technological advancement in other fields such as metaverse, web3, IoT, 5G, remote work. As of now, the semiconductor leaders should have started manufacturing in 3nm gate length. Samsung and TSMC have announced to ship chips, and Nvidia to ship 2nm chips in 2023. All major chip manufacturing companies are publishing new articles of Moore’s law to be alive for at least the next 4-6 years.

Conclucion

Moore’s law will still be used as a roadmap. I am not expecting any remarkable technology(in power and clock frequency) to hit the market any time soon. The innovation will be in ISA(Instruction Set Architecture). With the wide adoption of RISC-V, some performance and new applications can be seen. The demand for AI/ML will fuel the major demand for semiconductor chips. New Chiplet( A chiplet is a tiny integrated circuit (IC) that contains a well-defined subset of functionality) architectures will be flushed into the market for custom demands. Innovation will head towards packaging. More 3D packaged chiplets in the market.

The current focus of Intel is to increase the overall performance of computation instead of pushing processors to the limits. Fix the supply chain issue with long-term contracts, better synergy in EDA tools and foundries and adopt open standards in instruction sets. Von Neumann’s architecture is another bottleneck that should be optimized/replaced. The XPUs, FPGAs and ASICs will rise in the market adaptation.

Moore’s law is a brilliant economic idea that drives the semiconductor industry, It shall never die, it will be resurrected to guide the world into better computation technologies. The interesting part will be somewhere in the 2030’s when we will hit the bottleneck of physics at the atomic level.

Reference

https://www.tsmc.com/english/news-events/blog-article-20190814

https://rebootingcomputing.ieee.org/

https://rebootingcomputing.ieee.org/images/files/pdf/rc_irds.pdf

https://arstechnica.com/information-technology/2016/02/moores-law-really-is-dead-this-time/

Photo by Vishnu Mohanan on Unsplash