Industry Expert Blogs

Arm and ecosystem partners are looking to accelerate the next evolution of silicon.

A key challenge our partners are consistently looking to solve is: How can we continue to push performance boundaries, with maximum efficiency, while managing costs associated with manufacturing and yield? Today, as the ever more complex AI-accelerated computing landscape evolves, a key solution emerging is chiplets.

Chiplets are designed to be combined to create larger and more complex systems that can be packaged and sold as a single solution, made of a number of smaller dies instead of one single larger monolithic die. This creates interesting new design possibilities, with one of the most exciting being a potential route to custom silicon for manufacturers who historically chose off-the-shelf solutions. The excitement is focused on the concept of composability – putting together a bespoke complete solution by reusing a set of existing or standardized chiplets, each optimized for cost-performance. This reusability and standardization can lead to a multi-vendor chiplet supply chain that could enable both new and existing players to access performance and differentiation opportunities.

Delivering the common frameworks required to accelerate the chiplet market

To realize this opportunity, designers will need alignment on many non-differentiating choices in chiplet partitioning, and the associated impact on interface standardization. A common framework is needed to enable the next generation of silicon designers, and requires significant and focused collaboration and investment. Today I’m excited to provide updates on two new initiatives that will help to deliver this framework:

• Arm Chiplet System Architecture (CSA): Arm and a group of more than 20 partners are analyzing and defining optimal partitioning choices for chiplet-based systems. The goal is to develop the Arm Chiplet System Architecture (CSA), which will enable greater reuse of components (physical design IP, soft IP etc) between multiple suppliers. The group covers multiple market segments from mobile to automotive to infrastructure, exploring how we can better standardize around system design choices for different chiplet types, such as how to partition an Arm-based system across multiple chiplets, or their high-level properties (e.g. requirements for system memory or a Root of Trust).
• Updating AMBA to standardize protocols for chiplets: For more than 27 years, AMBA has been a foundational open industry standard. AMBA specifications like AXI and CHI have been used in billions of devices.
  • AMBA CHI is high speed, credited, and packetized, which makes it ideal for chiplets. The previously announced AMBA CHI C2C specification leverages the existing on-chip CHI protocol and defines how it is packetized, enabling it to be transported chip(let)-to-chip(let). We’re pleased to share today that the open specification is now formally released, after collaboration with a diverse set of partners across the industry. You can read more in this blog.
  • There are several existing AXI-based designs that are critical to enable in chiplets. We’ve committed to deliver an open AXI C2C specification that can be adopted by silicon vendors currently enjoying the benefits of AXI in their single chip designs.

Complex Problems Need Collaboration

Our investments into AMBA and CSA will enable partners to decompose an Arm-based system across multiple chiplets, in the same way a monolithic chip is composed of IP blocks. Alongside these new standards, there are several Arm-specific and industry-wide layers we need to continue to collaborate on:

• Physical layer: Industry standards. such as UCIe, are needed to define the physical layer for transporting data between chiplets within a package (for AMBA, PCIe and other protocols) and other non-differentiating aspects of system aggregation from chiplets.
• Protocols: We expect industry standard protocols, such as PCIe and CXL, alongside on-SoC interconnect protocols like AMBA, to enable the market to aggregate well-defined peripherals from across a motherboard into a package.
• High-level properties and partitioning: For Arm-based systems using AMBA protocols, there is an infinite flexibility to disaggregate the SoC into chiplets. CSA will help the Arm-ecosystem to reach consensus around the most valuable partitioning schemes to reduce fragmentation.

A diverse chiplet market, built on Arm

While mass market adoption of chiplets is likely to take many years to reach mass-market adoption, we’re excited by the potential these standards have to expedite the journey towards chiplet-based systems. The flexibility of the Arm platform is enabling the emerging chiplet ecosystem today – it’s a natural progression in our heritage of enabling partners to build custom silicon solutions quickly with the flexibility they need. As we continue to deliver the compute, performance efficiency and software solutions the industry demands in an increasingly complex landscape, we are bringing the ecosystem together around new critical standards and programs, like the recently announced Arm Total Design, that will enable a thriving, diverse chiplet ecosystem built on Arm.

To get involved in Arm CSA, partners can reach out to csa-feedback@arm.com.