Dissecting the New Fab Plans

Article By : Don Scansen

TSMC's expansion into Arizona might be much bigger than realized; Intel also has fab news; and IBM is creating a path to the future.

IBM (particularly the marketing team) deserves a healthy dose of congratulations for its announcement this week of a 2nm test chip. The news was picked up far and wide, reported everywhere from The Verge to Fox News and in between.

IBM compared the improvements of its 2 nm technology to 7 nm technology power and performance characteristics claiming 45% higher performance and 75% reduction in power consumption.

IBM 2nm nanosheet gate-all-around transistors (source: IBM)

IBM also made the now obligatory biological entity comparison by explaining the transistor was the size of two DNA strands.

I pulled the DNA reference from this CNN story. With apologies for piling on CNN, the press handling of the announcement was interesting since IBM will not be producing product wafers on this technology. According to CNN, “Most computer chips powering devices today use 10-nanometer or 7-nanometer process technology, with some manufacturers producing 5-nanometer chips.”

While it may be true that 5 nm technology is currently somewhat limited, high volume products have been on the market since the Apple A14 application processor launched in the iPhone 12.

State-of-the-art production semiconductor technology TSMC 5 nm finFET (source: semiconDr)

To put more perspective on this, 5 nm semiconductor technology is now in volume production at two foundries. TSMC produces the A14 and the mega-launch Apple M1 chip used in their non-phone products. Samsung foundry is also manufacturing on the 5 nm node for the flagship Qualcomm Snapdragon 888 processor.

It’s understandable for the IBM marketing team to want to compare its latest test chip from a development process to the current production state-of-the art, but judging the 2 nm process against trailing edge mature technologies is a stretch. With that peeve out of the way…

Duelling dollars

Considering the current chip shortage, too little attention was paid to reports about high-volume wafer production facilities. The most tantalizing was Reuters’ claim that TSMC will potentially add five additional wafer fabs to its Arizona operations.

The relative tax incentive advantages may favor Taiwan, but Jim McGregor recently listed five key reasons that any semiconductor manufacturer might be attracted to Arizona:

  • available land
  • infrastructure (including water and electricity supply)
  • talent pool
  • lack of natural disasters, and
  • tax incentives.

The rumors certainly fit with these desert attractions.

What we know for sure is that TSMC announced plans to spend an eye-watering $100 billion to expand production capacity over the next three years. With chip shortages making mainstream news headlines, few will quibble with any perceived one-upmanship in this since TSMC’s $100 billion announcement came hot on the heels of Intel’s $20 billion.

A significant chunk of that $100 billion might be funneled to Arizona. The Reuters report cited three unnamed sources claiming as many as five new TSMC fabs were planned for Arizona in addition to the officially announced one.

The Reuters rumor adds some weight to the earlier report from Anton Shilov that Arizona would be a “megafab-class” site with six plants operating there. Shilov mentioned seemingly credible (although unconfirmed) reports from Taiwan-based UDN that TSMC would be building six factories in the US.

TSMC pushed a little snow down the mountain when it announced plans to build a plant in Arizona just about one year ago.

BRINGING IT BACK

My opinion was and is that the plant announcement was a significant step toward a shift in thinking and location of semiconductor manufacturing on the map of the globe. Many share this opinion, but TSMC’s original Arizona plans had a few detractors suggesting that the plant would be obsolete by the time construction begins.

When the originally announced TSMC Arizona site starts producing chips in 2024 or 2025, the relative importance of the 5 nm wafers could be debated. If the plant originally comes online with 5 nm production, it will still play a big role in wafer supply, and 5 nm will still be an important part of the mix. The “obsolescence” of process technology will also be mitigated with the dissaggregation of function into chiplets produced on the technology most suited to the design and the economics of the particular function. More square millimeters of silicon in even the most advanced products will need to be on the cutting edge process.

Let’s assume the reports of TSMC expansion in Arizona are true. Given the scale of TSMC operations and the mood inside the hallowed halls of US government driving an appetite for spending on domesticating semiconductor production, it seems like a safe bet. It’s pretty obvious that TSMC would exceed its initial plans for expanding production into the US.

US President Joe Biden supports semiconductor wafer manufacturing on US soil (source: AP)

Going from one plant to a total of six, we can expect the Arizona manufacturing to converge with leading edge process technology by the time the last plant starts up. The global semiconductor landscape might start to look a lot different over the next decade.

Another safe bet is that any significant expansion of TSMC’s presence in the US would include advanced assembly and packaging.

In a recent conversation with Deca Technologies CTO Craig Bishop, the US lacks advanced assembly and packaging facilities with high volume capability offering contract services. Deca develops advanced packaging technologies, but these are all deployed to sites in Asia. Anticipating future chip supply issues and geographic concentrations of semiconductor production requires thinking through the entire component production including both wafer fab and packaging. It seems like the US is finally rethinking its domestic “fab-lite” policy.

A company that does operate advanced manufacturing on US soil is Intel. With a bold announcement of moving aggressively into the contract manufacturing space, Intel is also likely to offer a one-stop shop for fabless design companies.

Speaking of Intel, they were not standing still in terms of expanding their announcements of expanding production. And it was specifically related to advanced packaging as $3.5 billion was announced to expand their IC packaging facility in Rio Rancho, New Mexico.

The back-and-forth of announcements between Intel and TSMC is a lot of fun and drives a lot of optimism for the American economy. But it will be more interesting to see the two companies taking this duelling to the next step once shovels hit dirt.

Keep it cold

Although the planned expansions of semiconductor manufacturing in the US are very positive, there is a dark side to it. A big part of the expanding recognition of the strategic importance of semiconductors is the precarious geographic position of more than a half of world’s production.

There is concern for Taiwan and its chip output with China’s ambitions becoming clearer.

This article was originally published on EE Times.

Join the Conversation

  1. bob Whitcombe says:

    Given a Fab is a sealed box with a lot of equipment in it, and the equipment is what determines what process is used – I doubt the facility will be obsolete that quickly. Just waiting for extreme UV light to scale to an intensity level that is cost effective for large wafer chip manufacture will take a few more years.