Intel Foundry and Arm to Collaborate on 1.8nm Mobile SoCs

Intel
(Image credit: Intel)

Arm and Intel Foundry Services on Wednesday announced plans to conduct design technology co-optimization (DTCO) and system technology co-optimization (STCO) for Arm's mobile IP on the Intel 18A fabrication technology (1.8nm-class). The plan will enable clients of Arm and IFS to maximize performance, lower power consumption, and optimize die sizes of their upcoming SoCs involving Arm's IP.

Under the agreement, Intel Foundry Services and Arm will co-optimize Arm's IP and Intel's 18A fabrication process to increase the performance, power, area, and cost advantages of the new node. The two companies will initially focus on mobile SoC designs, but may eventually expand their collaboration to automotive, aerospace, datacenter, Internet of Things, and government applications. As part of the pact, Arm and IFS will develop reference design and optimized process developers kits for mobile SoCs.

Modern chip manufacturing technologies and processor designs are extremely complex and expensive. To maximize advantages of each new node for a particular design, foundries and chip developers these days optimize transistor design, libraries, standard cells, chip layout, and interconnects — just to name a few of the things involved in DTCO methodology. 

When it comes to Intel's 18A fabrication process, there are a lot of things that can be optimized on node and design level to extract more PPAC advantages out of the node. One of the key innovations of Intel 18A is usage of gate-all-around (GAA) transistors that Intel calls RibbonFET. In GAA transistors, the channels are horizontally oriented and completely enclosed by gates. These GAA channels are created through epitaxy and selective material removal, enabling designers to fine-tune them by altering the transistor channels' width to get higher performance or lower power consumption. If everything works well, such control allows them to reduce transistor leakage current and performance variability — this provides great opportunities for DTCO.

Yet another advantage of Intel's 18A is its backside power delivery network (PDN) called PowerVia. In order to efficiently provide power and respond quickly to the behavior of a modern processor, which can vary significantly depending on the workload, the PDN needs to be customized for a specific design and process technology, which gives plenty of opportunities for DTCO. Client and smartphone SoCs should be optimized for burst behavior, while datacenter SoCs should be optimized for constant high loads — which is why it is important that (for now) Intel and Arm will only address smartphone SoCs.

"Intel's collaboration with Arm will expand the market opportunity for IFS and open up new options and approaches for any fabless company that wants to access best-in-class CPU IP and the power of an open system foundry with leading-edge process technology," said Intel CEO Pat Gelsinger.

One important thing that to note about Intel's 18A is that this process technology will be used to make chips at different locations that IFS will operate around the world. This will be an advantage of this fabrication process as there are fabless chip designers seeking for localization of chip production

"As the demands for compute and efficiency become increasingly complex, our industry must innovate on many new levels. Arm's collaboration with Intel enables IFS as a critical foundry partner for our customers as we deliver the next generation of world-changing products built on Arm," said Arm CEO Rene Haas.

Anton Shilov
Freelance News Writer

Anton Shilov is a Freelance News Writer at Tom’s Hardware US. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

  • Giroro
    Intel's 18A node still isn't 1.8 nm nor "1.8 nm class". It's not expected to have any features that are even remotely close to 1.8nm lithography.

    It's extremely disappointing that their "cynical size-agnostic rebrand to confuse people into to thinking their transistors are smaller and therefore better" actually worked as effectively as it has.
    Are we also going to forget that their little efficiency cores are not full-sized fully-featured cores?
    Reply
  • rluker5
    Neither are Zen cores or Skylake cores compared to p cores.
    Reply
  • JamesJones44
    Giroro said:
    Intel's 18A node still isn't 1.8 nm nor "1.8 nm class". It's not expected to have any features that are even remotely close to 1.8nm lithography.

    It's extremely disappointing that their "cynical size-agnostic rebrand to confuse people into to thinking their transistors are smaller and therefore better" actually worked as effectively as it has.
    Are we also going to forget that their little efficiency cores are not full-sized fully-featured cores?

    This is an industry standard these days. TSMC's 4nm node is actually 5nm, but it's enhanced so they give it a new number. Advertised NM in the "chip" world became obsolete as a unit of comparison long ago.
    Reply
  • cyrusfox
    If you can't beat them (ARM) Join them (invite them to use your latest node). Interesting times ahead.
    Lets see how these multiple nodes roll out.
    Reply
  • jkflipflop98
    Giroro said:
    Intel's 18A node still isn't 1.8 nm nor "1.8 nm class". It's not expected to have any features that are even remotely close to 1.8nm lithography.

    It's extremely disappointing that their "cynical size-agnostic rebrand to confuse people into to thinking their transistors are smaller and therefore better" actually worked as effectively as it has.
    Are we also going to forget that their little efficiency cores are not full-sized fully-featured cores?

    Right. Right.

    Now point to the structure on TSMC's 3nm node that's actually 3nm. We'll wait.
    Reply
  • kjfatl
    Giroro said:
    Intel's 18A node still isn't 1.8 nm nor "1.8 nm class". It's not expected to have any features that are even remotely close to 1.8nm lithography.

    It's extremely disappointing that their "cynical size-agnostic rebrand to confuse people into to thinking their transistors are smaller and therefore better" actually worked as effectively as it has.
    Are we also going to forget that their little efficiency cores are not full-sized fully-featured cores?
    For decades Intel reported actual geometry numbers while their competition used numbers that were 30% smaller than actual. They did this specifically to make uninformed investors think they were somehow better than Intel and it worked. Since Intel was unable to end this deceptive practice done by TSMC, Samsung, Global Foundries and others., they began reporting using the same yardstick. Does this suddenly make their product significantly better and turn them back into a technology leader. The answer in the eyes of most investors is YES! They are suddenly a process generation ahead of where they were under the old naming convention.

    In a year or two, you can expect TSMC and Samsung to announce their 10A process based on 1/3 wavelength numbering. Actual features will be 30A
    Reply
  • bit_user
    cyrusfox said:
    If you can't beat them (ARM) Join them (invite them to use your latest node).
    This isn't entirely new. Intel previously announced plans to fab 3rd party SoCs, including some with ARM cores, like 5 years ago.
    Reply
  • TerryLaze
    bit_user said:
    This isn't entirely new. Intel previously announced plans to fab 3rd party SoCs, including some with ARM cores, like 5 years ago.
    Intel has been making their own RISC/ARM CPUs since all the way back in the 80' ,that's a few years more than just 5.
    https://en.wikipedia.org/wiki/Intel_i960
    cyrusfox said:
    If you can't beat them (ARM) Join them (invite them to use your latest node). Interesting times ahead.
    Lets see how these multiple nodes roll out.
    The last time ARM could realistically beat intel was back in the 90' where amiga apple and other micros all moved to risc CPUs, that ship has sailed now.
    ARM caters to a completely different market now from what intel is.
    Reply
  • bit_user
    TerryLaze said:
    Intel has been making their own RISC/ARM CPUs since all the way back in the 80' ,that's a few years more than just 5.
    https://en.wikipedia.org/wiki/Intel_i960
    That's misleading.

    First, don't confuse RISC with ARM. Not to get in a debate over whether ARM is truly RISC, but it just doesn't benefit anyone's understanding to muddle them together. Second, the i960 has been discontinued for about 15 years, now.

    They also inherited a line of ARM-based microcontrollers from DEC, called StrongARM, which they followed with XScale and sold off to Marvell, in 2006.

    More importantly, this isn't Intel designing ARM cores (unlike the above examples), but rather partnering with ARM to help them optimize their IP for Intel's 18A process node.

    TerryLaze said:
    The last time ARM could realistically beat intel was back in the 90'
    Cool story. Let's wait and see how Nvidia's Grace compares to Sapphire Rapids.

    On a per-vCPU basis, Amazon's Graviton 3 already trashed their Ice Lake SP instances (as well as Milan).

    TerryLaze said:
    ARM caters to a completely different market now from what intel is.
    No, but you saying that tells me Intel is definitely worried about ARM.

    Sure, ARM plays more in the embedded space than Intel. ARM also dominates phones, unlike Intel. However, ARM is gaining penetration into the laptop segment and on pace to overtake x86 in the cloud, in just a few years.
    Reply
  • JamesJones44
    bit_user said:
    Cool story. Let's wait and see how Nvidia's Grace compares to Sapphire Rapids.

    On a per-vCPU basis, Amazon's Graviton 3 already trashed their Ice Lake SP instances (as well as Milan).

    Adding to that, Apple Silicon holds it's own against Intel/AMDs laptop parts while using less power.

    ARM is very capable. Vertical scale has always been the question for RISC based processors, for now I personally think this is still an open question, but it has been slowly catching up to its CISC based counterparts now the power consumption has become a concern at all levels of processing.
    Reply