China Develops Domestic Chiplet Interface

Semiconductor
(Image credit: Shutterstock)

As China is accelerating its semiconductor self-sufficiency efforts, the newly formed China Chiplet League this week introduced its homegrown Chiplet Interconnect Interface Standard, according to a DigiTimes report citing the Chinese website Cailianshe. The new interface is meant to enable custom multi-chiplet designs developed by China-based companies and made in the People's Republic.

China's original Chiplet Interconnect Interface Standard, also known as the ACC 1.0 (Advanced Cost-driven Chiplet Interface 1.0), is being developed by a group of companies specializing in chip design, IP as well as packaging, testing and assembly services. The China Chiplet League seems to be coordinated by the Institute for Interdisciplinary Information Core Technology, though its exact role is not particularly clear. The ultimate goal of the China Chiplet League is to ensure that ACC 1.0 will be a cost effective and feasible solution for the country's chip designers. 

Developing standards for chiplets is particularly important these days, as it is getting harder and more expensive for chipmakers to tangibly increase transistor density every 18 months. The industry is slowly moving to multi-chiplet designs as it enables developers to through in as many transistors as they need to achieve their performance goals without making their designs excessively expensive (as large monolithic chips made on production nodes like 5 nm and below are prohibitively expensive).  

One of the advantages of multi-chiplet designs is that they can be designed by various companies and produced by different foundries at different nodes. To development the chiplet ecosystem ensure that chiplets from different vendors are compatible with each other, leading chip designers, producers and packaging specialists like AMD, ASE, Intel, Microsoft, Samsung, Qualcomm and TSMC have formed the Universal Chiplet Interconnect Express (UCIe) alliance. 

The Chinese semiconductor industry also has to adopt chiplet designs, mostly because foundries like Semiconductor Manufacturing International Corp. and Hua Hong can produce chips only using trailing nodes and therefore their clients cannot be competitive with companies that use TSMC's leading-edge nodes as far as monolithic designs are concerned. But with multi-chiplet designs, they have much better chances.  

Meanwhile, it does not make a lot of sense for Chinese companies to adopt UCIe since the U.S. government may restrict exports of advanced chiplets to People's Republic. Furthermore, chiplets made on 5nm-class and more sophisticated nodes may be physically and electrically incompatible with chiplets produced at a 28nm fabrication process.  

In fact, the Institute for Interdisciplinary Information Core Technology admits that the industry in China is currently in its developmental stage and will have to play catch-up in the current global situations.  

The organization emphasized the need for collaboration among Chinese suppliers across the entire supply chain, from upstream to downstream, to establish a comprehensive chiplet ecosystem. By fostering a virtuous cycle of downstream demand driving upstream investment, the China Chiplet League can enable economies of scale in mass production, facilitating performance breakthroughs. 

 

Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. 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.

  • cyrusfox
    Steep climb for them. Good luck
    Its the world vs china on advance packaging as almost everyone that is a maker in this space is part of the UCIe consortium
    https://www.uciexpress.org/membershiphttps://www.uciexpress.org/copy-of-membershiphttps://www.uciexpress.org/adopter-membersAll three links provide different list of companies part of the interconnect consortium
    Reply
  • bit_user
    So, this is interesting... In the short-term, perhaps the motivation is to have something that's more cost-effective to deploy on older nodes.

    If we wind the clock forward a several years, it could serve as in impediment to the inter-operation and inter-working between (i.e. Chinese and Western-designed) chiplets, that is the core mission of UCIe. Ultimately, it puts us on the track of fragmentation and a schism in the chiplet market. Along with some external leverage, this can work against UCIe-based designs and turn support for that standard into a competitive liability.
    Reply
  • gg83
    The China Chiplet league! Is the Marvel or DC?
    Reply
  • gg83
    bit_user said:
    So, this is interesting... In the short-term, perhaps the motivation is to have something that's more cost-effective to deploy on older nodes.

    If we wind the clock forward a several years, it could serve as in impediment to the inter-operation and inter-working between (i.e. Chinese and Western-designed) chiplets, that is the core mission of UCIe. Ultimately, it puts us on the track of fragmentation and a schism in the chiplet market. Along with some external leverage, this can work against UCIe-based designs and turn support for that standard into a competitive liability.
    So true. China is big enough to flip the UCIe on its head down the road.
    Reply
  • bit_user
    gg83 said:
    So true. China is big enough to flip the UCIe on its head down the road.
    It could happen. The camel's nose under the tent will be when western companies start making chiplets, or at least special versions, with some future revision of ACC. That will be needed, if you want to sell your chiplets into the future Chinese ecosystem. Once enough western chiplets have ACC versions, then you could see ACC chiplets being adopted by western customers, as it would enable them to use some lower-cost Chinese chiplets.

    Or, ACC could just die out. Maybe things smooth over with China, one way or another, and they begin to embrace UCIe as they move onto newer process nodes where it's viable.
    Reply