Intel is increasing its reliance on erstwhile rival Taiwan Semiconductor Manufacturing Co. (TSMC) in its attempt to boost sales and eventually regain dominance as the world leader in manufacturing scale and chip process technology.

California-based Intel will join Apple to order the world’s first 3-nm chips from TSMC as the Taiwan chip foundry ramps up its newest process this year, according to three analysts surveyed. Intel and Apple are likely to be the only two TSMC customers at that most-advanced node during the ramp-up, according to the analysts.

At Intel’s investor meeting yesterday, CEO Pat Gelsinger reiterated a promise to “regain leadership” in the chip business. Gelsinger, who has been at the helm at Intel for a year, said the company will advance “five nodes in four years”, after which chip nomenclature will shift to angstroms from nanometers.

Intel’s leap depends on TSMC’s help at the 5- and 3-nm nodes. One of the challenges will be combining chiplets from TSMC with other chiplets made internally by Intel into one device like the Ponte Vecchio; that will involve matching chiplets made in TSMC’s 5nm process with Intel’s own silicon, using Intel’s new packaging technologies, which include embedded multi-die interconnect bridge (EMIB) and Foveros.

It’s part of a shift from planar to 3D chips that vastly expand semiconductor real estate in a bid to increase transistor density and ideally extend Intel’s Moore’s Law for decades to come. Intel expects to be able to integrate 1 trillion transistors on a single chip by 2030.

Intel’s plan will count heavily on ASML’s high NA extreme ultraviolet (EUV) lithography tools. Intel is ASML’s first high-NA customer in an EUV business (which currently includes only a handful of customers, including TSMC, Samsung, Intel and Micron). TSMC foundry rivals Samsung and Intel are leading in the adoption of gate all around 3D technology combined with EUV in hopes of surpassing the FinFET and EUV knowhow that TSMC has used to maintain sales and process leadership.

Analyst comments

Samsung will do risk production for GAA early in 2022 and wafer-in volume by the end of 2022, according to the Credit Suisse report.

Samsung GAA will start with the company’s own Exynos product in the first half of 2023, while TSMC ramps up the new Apple-Intel orders, followed by a second wave of business from potentially more customers sometime during 2023-24, the report says.

“We also believe Intel would still announce more details on its products in TSMC in 2023-24, (and) on 3nm in the next few quarters while it tries to improve its in-house manufacturing,” according to the report. TSMC will address $5 billion of sales that are a chunk of Intel’s $24 billion total addressable market in chip manufacturing, according to the report.

The views of Arete Research senior analyst Brett Simpson are similar.

“Intel is engaged with TSMC on N3 for its next-generation graphic processors (GPUs) in 2023, and the ramp plans here look bigger than we first thought. This is a material engagement,” he says.

Intel’s GPU rivals include Nvidia and others. GPUs are not a product area that Intel has addressed historically.

“We thought they would go slow, but the ramp plans for GPUs look to be over 10K wafers per month. That signals Intel will be a node ahead of GPU rivals next year,” according to one analyst surveyed, who asked for anonymity.

The big prize for TSMC is strategic CPU tapeouts — but there is a strong inclination for Intel to keep its core tech in-house for now, Simpson told EE Times in an email. “As Intel shifts more towards ‘fab-agnostic’ chip design, it is possible to see more outsourcing of its CPUs most likely in the 2024 timeframe,” he said.

Intel is timing its leap to angstroms from nanometers even as it is placing new bets on breaking into the GPU business and adopting GAA and high-NA lithography.

And it is embarking on this program while most of the chip industry still faces supply-chain issues.

“Semiconductor equipment companies are bursting at the seams in supporting TSMC, Intel and Samsung capex plans. Any meaningful shift by Intel to use TSMC for CPUs will need to be part of a careful long-term supply planning process. For example, if Intel is unable to source the tools they need to grow their business, then a wider commitment to TSMC’s advanced nodes would make strategic sense,” according to Simpson.

The roadmap for the Intel-TSMC partnership is visible up to the 3-nm generation, after which the two are likely to become more like competitors. Intel expects TSMC’s nanometer narrative to change to angstroms with the introduction of Intel’s Meteor Lake and Arrow Lake products in 2023.

“I believe Intel will outsource some production, primarily for TSMC’s 3 nm and 5 nm, and a bit for 6 nm,” Mark Li, a senior research analyst with Bernstein & Co., told EE Times. “TSMC and Intel have been preparing for awhile, and the notable volume will start in early 2023.”

Intel general manager of technology development Ann Kelleher yesterday pledged to regain “leadership” in process technology at the 20-angstrom node, or 20A, in Intel’s parlance. The 2023-2024 period is when Intel expects to change chip nomenclature and retake technology leadership.

In the meantime, top-line sales for TSMC, Intel and their customers like Apple and MediaTek will be an important factor for the chip foundries in their effort to dominate the business. Leader TSMC has had annual sales growth near or exceeding 20 percent for years, helping to fund expansion of new fabs that can cost as much as $30 billion. Intel, which weeks ago announced its Silicon Heartland fab project in Ohio, yesterday said its annual sales growth won’t reach double digits for several more years.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved