Inside Intel's CPU Packaging Factory: From Wafer to Chip

Final Thoughts

Intel has a tremendous arsenal of cutting-edge chip packaging tech, and that will help it not only move to chiplet-based designs (called 'tiles' in Intel parlance) with Meteor Lake but also to sell its packaging services to other companies as part of its Intel Foundry Services initiative. In fact, Intel's first IFS customers were attracted to its packaging services. With the current shortage of packaging production capacity that has led to Nvidia's AI GPU shortages (which TSMC says won't be resolved for a year and a half), it's a safe bet that Intel will see a lot more interest in those services over the coming year.

On the packaging front, Intel's biggest near-term objective is to build the scale of its advanced packaging operations. Meteor Lake represents the first time Intel has used advanced packaging in its high-volume client processor lineup, and that can be a tricky endeavor. The consumer market demands massive volume shipments, and Intel plans to quadruple its advanced packaging capacity by 2025, part of which involves investing $7 billion in new facilities in Kulim in addition to its investments in Penang and Rio Ranco in the United States. It's even pledged to build an advanced packaging plant in Poland, too, though the timeline of that effort remains in flux.

However, Intel also has a challenge: Cost. Advanced packaging invariably costs more than the standard single-die packaging process used with the majority of today's chips, and in the consumer market, that could limit its applicability to higher-end models that tend to chase the edge of performance. We can expect those costs to eventually diminish over time, but they remain a factor with Intel's Meteor Lake line of chips, and perhaps with the follow-on Arrow and Lunar Lake processors that will employ similar 3D Foveros designs.

Intel logo

(Image credit: Tom's Hardware)

Yes, the ability to package a chip with multiple other older and less expensive process nodes can reduce overall cost, and Intel is using that to its advantage by employing externally sourced tiles from TSMC for its GPU, SoC, and I/O tiles. However, it's still more expensive than using other simpler forms of packaging, like the SERDES-based interconnects that AMD uses with its Ryzen and EPYC lineups that also leverage a mix of different process nodes. In order for advanced packaging to pay off from an economic standpoint, the end products will have to deliver enough of a performance or power advantage to justify the extra cost.

Intel hasn't revealed the details of the initial Meteor Lake chips, but they apparently won't be used for high-end desktop models — Intel will purportedly continue to rely on single-die processors for that segment of the market, while Meteor Lake will likely step in primarily for mobile applications. It also isn't clear if Meteor Lake will comprise all of Intel's mobile offerings or just a subset of the higher-end models.

Of course, Intel also punches out plenty of high-volume multi-die server chips, like Sapphire Rapids. But here we see Intel still relying upon single-die processors for the mid-range and lower-end models, presumably due to cost. Economies of scale should help with that, but we also expect that Emerald Rapids and Granite Rapids could take a similar approach, with multi-die chips only used for the upper echelons of the product stack.

The key to reducing cost is often simply to gain the efficiencies of mass production, and here Intel has a solid and rapid expansion plan that could put it at the forefront of advanced packaging production capacity over the next few years. Not to mention plenty of experience in high-volume packaging operations. However, other companies, like TSMC, also have promising tech that will keep Intel on its toes as its primary competitor fuels its other competitors.

No other IDM operates on the leading edge at the scale that Intel does, but the company's past missteps on process node tech and the chip design side of its operations have left a cloud over the company, one that is just now beginning to show a sliver of light around the edges. However, unlike the negative press we often read about the state of Intel, the employees at its Malaysian facilities are full of optimism with a shared goal of returning the company to its former glory.

Intel's advances in process technology as it moves to develop an unprecedented five nodes (technically six) in four years are critical to regaining its leadership spot, but until the full recovery plan has time to play out, its packaging tech could be the unsung hero that helps defray the impact of being a node or two behind the competition. 

Paul Alcorn
Managing Editor: News and Emerging Tech

Paul Alcorn is the Managing Editor: News and Emerging Tech for Tom's Hardware US. He also writes news and reviews on CPUs, storage, and enterprise hardware.

  • Roland Of Gilead
    "If successful, those advanced nodes will grant Intel the lead over TSMC for the first time in years, " - so, the reasoning being here, that TSMC will be standing by scratching their arses during this same period. And thus Intel will overtake them!? Okay.
  • PaulAlcorn
    Intel will match TSMCs std cell density with the Intel 4 process by the end of this year. After that, roadmap projections indicate Intel will pull ahead.
  • Elusive Ruse
    Nice one, @PaulAlcorn, must have been a dream tour!
  • jkflipflop98
    Roland Of Gilead said:
    "If successful, those advanced nodes will grant Intel the lead over TSMC for the first time in years, " - so, the reasoning being here, that TSMC will be standing by scratching their arses during this same period. And thus Intel will overtake them!? Okay.

    No, you just inserted your own thoughts. No one is saying TSMC is going to stand still. We're saying they aren't going to be able to keep up.
  • vertuallinsanity
    Intel utilized TMSC as a stand-in *partner* during a period of machinery teething issues.

    The intention of that partnership, for certain products, was always temporary in nature or Intel would have gone fab-less similar to AMD. Intel utilized TSMC to prevent bring-to-market delay while simultaneously reworking their fab processes and equipment to a viable state.

    The article seems to imply TSMC makes their own processors or gpus and they're in direct competiton with Intel. They don't and they arent.

    TSMC and/or Intel may trade process enhancement "blows" but Intel makes Intel CPUs for Intel and TSMC makes 'em for "anybody".

    The article implication could apply to Apple, NVidia, AMD or others competing with Intel on similar node products.

    Intel will continue to work with TSMC on GPUs and other product lines deemed fiscally feasible. Intel and TSMC are still partners.

    There is no way anyone from Tom's hardware was allowed a tour of Intel's new ish.

  • Roland Of Gilead
    jkflipflop98 said:
    No, you just inserted your own thoughts. No one is saying TSMC is going to stand still. We're saying they aren't going to be able to keep up.
    No, you just inserted your own thoughts, in relation to my post!

    If you didn't notice, this article is one sided! I did certainly enjoy the tour and article in relation to Intel's tech, but there is zero context as to why TSMC 'might' fall behind.

    You are right, no one is saying TSMC is gonna fall behind, but without context how is a reader supposed to know that? By checking the forums posts for an addendum article that gives the reasons why TSMC might fall behind? A bit of balance to illustrate these comments would help.
  • td47
    It would be interesting to know how Intel prevents IP and manufacturing methods/secrets from being exfiltrated from the Chengdu facility in China - given that country's penchant f or stealing and copying the Western Tech to try to catch up!