3D Inspection Solutions

Intel Looks to Regain Semiconductor Chip Leadership from TSMC; Separates Manufacturing and Fabless Units

Could fabless be fabulous for Intel?

For the first time in more than 50 years, Intel is freeing its manufacturing unit from its fabless operations. This week, Intel detailed the steps it is taking to make that possible.

The company has set a target date of early 2024 to grant independence to its foundry business, which will get its own accounting books. In the past, Intel dedicated its factories exclusively to its own chips, while a few forays into manufacturing chips for outside customers failed.

This was the right strategy for the time and we were highly successful. The semiconductor industry and computing … has evolved rapidly and we need to adjust our business operations in response.

Premium 3D Inspection Solutions Sponsor | Koh Young

David Zinsner, Intel Chief Financial Officer

The goal of separating fabless and manufacturing operations is to serve markets that include the cloud, edge, and AI, which are emerging. While the company’s fabs were previously focused on serving the PC and server markets, emerging market dynamics guided the company’s strategy to quickly diversify into making chips for the new markets.

Intel officials detailed how the new fabs, which will be set free to make chips for non-Intel customers, will help make its chip manufacturing and design more efficient. Intel’s chip design business units will become customers of the fabs, just like any other customer seeking fab capacity.

The competition for capacity will help Intel’s fabless units become nimble and efficient. It will not be like the past when Intel’s factories were dedicated to making its own chips and manufacturing capacity was guaranteed. Going forward, Intel’s chip design units could lose out internal fab capacity to third parties, who may be willing to pay top dollar for guaranteed capacity.

It will encourage business units to move to the next node faster as they won’t be burdened by the initial high cost as a new process ramps

Zinsner

Looks Like the (Successful) TSMC Playbook

Intel’s approach resembles that of contract chip manufacturer Taiwan Semiconductor Manufacturing Co. (TSMC), which makes chips for Nvidia, Apple, and AMD. During the chip shortage, TSMC guaranteed capacity to Nvidia, AMD, and STMicroelectronics, who signed long-term agreements (LTAs) to maintain consistent chip supply amid component shortages and tight manufacturing capacity.

Our ambitions are to be the second largest external foundry by 2030, that remains our goal. In the new model based on internal volume, we expect to be the second largest foundry next year with manufacturing revenue of greater than $20 billion.

Zinsner 

The new company structure will make fab selection more democratic for Intel’s internal business units, which could go to outside fabs for manufacturing. Intel’s future chip designs include chiplets, where a bunch of silicon modules made on different processes from manufacturers can be bunched into a single chip package.

Today, roughly 20% to 25% of our silicon is manufactured externally, and Meteor Lake, which we’ll launch in the second half of this year, benefits from both internal and external wafer supply.

Zinsner 

Intel’s decoupling of fabless business from its foundry will allow Intel’s internal chip development process to operate at its own pace. Most importantly, competition from other fabless companies, which possibly could include AMD and Nvidia, could speed up Intel’s internal chip design efforts.

The manufacturing group will now face the same market dynamics as their foundry counterparts. They will need to compete for volume through performance and price as internal customers will have the option to leverage third-party foundries.

Zinsner 

The independent fabless and fab models will face more exposure to market dynamics, which drive competition.

As we shift to the internal foundry model, business units will now be charged the market price based on test time

Jason Grebe, Corporate Vice President and General Manager for the Corporate Planning Group at Intel

Intel’s internal foundry business and fabless businesses will also get independent financial books in 2024, which will provide better awareness of how each of the businesses are doing. The separation of accounting will also add clarity to the costs and revenue generated by each unit. Previously, Intel’s accounting shared profit and loss statements, and the foundry was seen as a liability when not running at full capacity.

Intel’s chip development in the past was closely tied to its two-year advances in manufacturing. But those advances have been stuck in neutral at the 14-nm and 10-nm processes over the last decade.

Gelsinger’s vision to play out in four years.

In 2021, Intel CEO Pat Gelsinger set a goal to advance five generations over the next four years. The first of the advances is on the Intel 7-nanometer process which is being used for the Sapphire Rapids server and the Raptor Lake PC chips.

The next PC chip, Meteor Lake, is being released next year on the Intel 4 process. Intel next year will ship server chips, Granite Rapids and Sierra Forest, which will be made on the Intel 3 process, and is expected to deliver 18% more performance per watt compared to chips based on the Intel 4 process.

A new PC chip called Arrow Lake next year, will lead off Intel’s 20A, and Clearwater Forest will be made on the most advanced 18A node.

Analysts site Intel’s challenge as the company’s inability to deliver chips on time, which allowed TSMC to take over as a leader. But Intel is confident it will regain leadership from TSMC on the 18A node, which packs cutting-edge technologies like gate-all-around (GAA), which is considered a big advance in transistor technology to bring better performance and more energy efficiency.

We continue to be confident that we will sign up our first Intel 18A external foundry customer this year

Zinsner 

Intel has shared no manufacturing roadmap beyond its most advanced Intel 18A process, but the company seems to be abandoning its tick-tock model of two-year node advances on manufacturing.

We are planning for longer tails on our process technologies to try to get more return on those investments

Intel’s Grebe

Of course there are two sides to the discussion. There have been calls over the years for Intel to spin off its manufacturing and fabless units, but there are benefits in having both under the same roof, Intel executives said during the call.

The manufacturing unit will speed up Intel’s chip design, and vice versa. As a relatively new player, Intel also needs to build credibility as a reliable manufacturer of chips, and Intel’s upcoming server and PC chips on the advanced nodes will prove to potential customers that it is up to the task.

Intel Gets a Nod from NVIDIA.

Nvidia CEO Jensen Huang last month gave a vote of confidence on Intel’s manufacturing, saying that a test chip received from Intel made on a next-generation process was showing good results. Nvidia is investigating the idea of diversifying its chip supplies to include manufacturers outside TSMC.

We think that our ability to use our internal customers as what we call customer zero to ramp volume on the new nodes is a benefit to the external foundry customers who then get to benefit from the use of those nodes.

Zinsner 

Subscribe to the latest Electronics Manufacturing News

By pressing the Subscribe button, you confirm that you have read and are agreeing to our Privacy Policy and Terms of Use
Advertisement