Intel’s Foundry Roadmap Describes the Angstrom Put up-Nanometer Period


  • Intel’s graph of the past and future.

  • Intel’s new naming scheme is more in line with the other foundries.


  • RibbonFET (or a wraparound gate transistor) allows you to stack channels vertically, which takes up less space.

  • Describing Intel: “The image on the left shows a design with power and signal cables intermingled on top of the wafer. The image on the right shows the new PowerVia technology, Intel’s unique industry first implementation of a rear power network.”


Earlier this year, Intel got a new CEO and launched a new business plan that would open its foundries to other chip design firms, just like TSMC and Samsung Semiconductor. At today’s “Intel Accelerated” event, the company drafted a roadmap for its future as a for-hire foundry. In addition to the future of ever smaller process nodes, the company also announced that it has won Qualcomm, one of the world’s largest chip designers, as a future foundry customer.

In the course of entering the foundry market, Intel will name its process nodes more like its competitors. The process node numbers used for chips like “5nm” started out as a measure of transistor size, but eventually marketers got them and companies started cheating their numbers to look more advanced. Intel says its new naming scheme better fits the way TSMC and Samsung talk about their foundry technologies. Gone are the days of “Intel 10nm Enhanced Super Fin” – instead the node is called “Intel 7”. It should have a density comparable to that of the 7 nm nodes from TSMC and Samsung and be ready for series production in the first quarter of 2022 (TSMC and Samsung are currently delivering “5 nm” products). “Intel 4” – previously called “7nm” by Intel – is now to correspond to TSMC and Samsung’s 4-nm nodes and will begin manufacturing products from 2023.

If you’re wondering what if we run out of “nm” numbers, Intel’s selling point for this is the “angstrom” era, a unit of measurement that is a tenth of a nanometer. In 2024, the company plans to ramp up the “Intel 20A” process node (which is a “2nm” equivalent, but Intel previously called this node “5nm” but remember that these are marketing metrics, not actual units of measure). At the beginning of 2025 the company is working on “Intel 18A”.

The name change to “Intel 20A” instead of “2nm” appears to be in part because this process node will include some important architecture changes for Intel’s chips. The company has used FinFET transistors for years, but for the Intel 20A, the company will be moving to a gate-all-round (GAA) design it calls “RibbonFET”. FinFETs would scale the channel current capacity by adding more fins and therefore more horizontal space. ATM designs, however, allow chip manufacturers to stack multiple channels on top of each other, making current capacity a vertical problem and increasing chip density. Intel 20A will also introduce “PowerVias”, a new chip design method that uses power from the back of the chip. This design would put the power layer on the bottom of the chip, then the transistors, and then the communication lines. In traditional chip design, the transistors are placed at the bottom, and higher signal and power layers must mix to reach the transistor layer.


If Intel actually manages to stick to its roadmap, Qualcomm should be an interested customer. Expressing interest in the 20A node, President and CEO Cristiano Amon said, “Qualcomm is excited about the breakthrough RibbonFET and PowerVia technologies coming in the Intel 20A [Intel Foundry Services] that will help the US fabless industry get their products to an onshore manufacturing site. “

Today Qualcomm makes many chips and is a customer of both TSMC and Samsung. The two companies regularly compete for every new design in Qualcomm’s lineup, with industry reports often describing a down-to-earth horse race in which one beats the other. Whether or not Intel will be in most of these Foundry battles depends on whether it can catch up with TSMC and Samsung. At least now, Qualcomm offers Intel a place in the race, instead of smartphone triviality.

Offer image from Intel